Glean SDK

Glean logo

The Glean SDK is a modern approach for a Telemetry library and is part of the Glean project.

To contact us you can:

The source code is available on GitHub.

Using this book

This book is divided into 4 main chapters:

Using the Glean SDK

If you want to use the Glean SDK to report data then this is the section you should read. It explains the first steps from integrating Glean into your project, contains details about all available metric types and how to do send your own custom pings.

Metrics collected by the Glean SDK

This chapter lists all metrics collected by the Glean SDK itself.

Developing the Glean SDK

This chapter describes how to develop the Glean SDK and its various implementations. This is relevant if you plan to contribute to the Glean SDK code base.

This Week in Glean

“This Week in Glean” is a series of blog posts that the Glean Team at Mozilla is using to try to communicate better about our work. They could be release notes, documentation, hopes, dreams, or whatever: so long as it is inspired by Glean.

License

The Glean SDK Source Code is subject to the terms of the Mozilla Public License v2.0. You can obtain a copy of the MPL at https://mozilla.org/MPL/2.0/.

Using the Glean SDK

In this chapter we describe how to use the Glean SDK in your own libraries and applications.

Adding Glean to your project

Before using Glean

Products using the Glean SDK to collect telemetry must:

  • add documentation for any new metric collected with the library in its repository (see an example);
  • include the markdown-formatted documentation generated from the metrics.yaml and pings.yaml files in the project's documentation;
  • go through data review for the newly collected data by following this process;
  • provide a way for users to turn data collection off (e.g. providing settings to control Glean.setUploadEnabled()).

Usage

Integrating with your project

Setting up the dependency

Glean is published on maven.mozilla.org. To use it, you need to add the following to your project's top-level build file, in the allprojects block (see e.g. Glean's own build.gradle):

repositories {
    maven {
       url "https://maven.mozilla.org/maven2"
    }
}

Each module that uses Glean needs to specify it in its build file, in the dependencies block. Add this to your Gradle configuration:

implementation "org.mozilla.components:service-glean:{latest-version}"

Important: the {latest-version} placeholder in the above link should be replaced with the version of Android Components used by the project.

The Glean SDK is released as part of android-components. Therefore, it follows android-components' versions. The android-components release page can be used to determine the latest version.

For example, if version 33.0.0 is used, then the include directive becomes:

implementation "org.mozilla.components:service-glean:33.0.0"

Requirements

  • Python >= 3.6

Setting up the dependency

Glean can be consumed through Carthage, a dependency manager for macOS and iOS. For consuming the latest version of Glean, add the following line to your Cartfile:

github "mozilla/glean" "{latest-version}"

Important: the {latest-version} placeholder should be replaced with the version number of the latest Glean SDK release. You can find the version number on the release page.

Then check out and build the new dependency:

carthage update --platform iOS

Integrating with the build system

For integration with the build system you can follow the Carthage Quick Start steps.

  1. After building the dependency one drag the built .framework binaries from Carthage/Build/iOS into your application's Xcode project.

  2. On your application targets' Build Phases settings tab, click the + icon and choose New Run Script Phase. If you already use Carthage for other dependencies, extend the existing step. Create a Run Script in which you specify your shell (ex: /bin/sh), add the following contents to the script area below the shell:

    /usr/local/bin/carthage copy-frameworks
    
  3. Add the path to the Glean framework under "Input Files":

    $(SRCROOT)/Carthage/Build/iOS/Glean.framework
    
  4. Add the paths to the copied framework to the "Output Files":

    $(BUILT_PRODUCTS_DIR)/$(FRAMEWORKS_FOLDER_PATH)/Glean.framework
    

We recommend using a virtual environment for your work to isolate the dependencies for your project. There are many popular abstractions on top of virtual environments in the Python ecosystem which can help manage your project dependencies.

The Python Glean bindings currently have prebuilt wheels on PyPI for x86_64 Windows, Linux and macOS.

If you're running one of those platforms and have your virtual environment set up and activated, you can install Glean into it using:

$ python -m pip install glean_sdk

If you are not on one of these platforms, you will need to build the Glean Python bindings from source using these instructions.

TODO. To be implemented in bug 1643568.

Adding new metrics

All metrics that your project collects must be defined in a metrics.yaml file.

The format of that file is documented with glean_parser. To learn more, see adding new metrics.

Important: as stated before, any new data collection requires documentation and data-review. This is also required for any new metric automatically collected by the Glean SDK.

In order for the Glean SDK to generate an API for your metrics, two Gradle plugins must be included in your build:

The Glean Gradle plugin is distributed through Mozilla's Maven, so we need to tell your build where to look for it by adding the following to the top of your build.gradle:

buildscript {
    repositories {
        // Include the next clause if you are tracking snapshots of android components
        maven {
            url "https://snapshots.maven.mozilla.org/maven2"
        }
        maven {
            url "https://maven.mozilla.org/maven2"
        }

        dependencies {
            classpath "org.mozilla.components:tooling-glean-gradle:{android-components-version}"
        }
    }
}

Important: as above, the {android-components-version} placeholder in the above link should be replaced with the version number of android components used in your project.

The JetBrains Python plugin is distributed in the Gradle plugin repository, so it can be included with:

plugins {
    id "com.jetbrains.python.envs" version "0.0.26"
}

Right before the end of the same file, we need to apply the Glean Gradle plugin. Set any additional parameters to control the behavior of the Glean Gradle plugin before calling apply plugin.

// Optionally, set any parameters to send to the plugin.
ext.gleanGenerateMarkdownDocs = true
apply plugin: "org.mozilla.telemetry.glean-gradle-plugin"

Note: Earlier versions of Glean used a Gradle script (sdk_generator.gradle) rather than a Gradle plugin. Its use is deprecated and projects should be updated to use the Gradle plugin as described above.

The metrics.yaml file is parsed at build time and Swift code is generated. Add a new metrics.yaml file to your Xcode project.

Follow these steps to automatically run the parser at build time:

  1. Download the sdk_generator.sh script from the Glean repository:

    https://raw.githubusercontent.com/mozilla/glean/{latest-release}/glean-core/ios/sdk_generator.sh
    

    Important: as above, the {latest-version} placeholder should be replaced with the version number of Glean SDK release used in this project.

  2. Add the sdk_generator.sh file to your Xcode project.

  3. On your application targets' Build Phases settings tab, click the + icon and choose New Run Script Phase. Create a Run Script in which you specify your shell (ex: /bin/sh), add the following contents to the script area below the shell:

    bash $PWD/sdk_generator.sh
    
  4. Add the path to your metrics.yaml and (optionally) pings.yaml under "Input files":

    $(SRCROOT)/{project-name}/metrics.yaml
    $(SRCROOT)/{project-name}/pings.yaml
    
  5. Add the path to the generated code file to the "Output Files":

    $(SRCROOT)/{project-name}/Generated/Metrics.swift
    

    Important: The parser now generates a single file called Metrics.swift (since Glean v31.0.0).

  6. If you are using Git, add the following lines to your .gitignore file:

    .venv/
    {project-name}/Generated
    

    This will ignore files that are generated at build time by the sdk_generator.sh script. They don't need to be kept in version control, as they can be re-generated from your metrics.yaml and pings.yaml files.

Important information about Glean and embedded extensions: Metric collection is a no-op in application extensions and Glean will not run. Since extensions run in a separate sandbox and process from the application, Glean would run in an extension as if it were a completely separate application with different client ids and storage. This complicates things because Glean doesn’t know or care about other processes. Because of this, Glean is purposefully prevented from running in an application extension and if metrics need to be collected from extensions, it's up to the integrating application to pass the information to the base application to record in Glean.

For Python, the metrics.yaml file must be available and loaded at runtime.

If your project is a script (i.e. just Python files in a directory), you can load the metrics.yaml using:

from glean import load_metrics

metrics = load_metrics("metrics.yaml")

# Use a metric on the returned object
metrics.your_category.your_metric.set("value")

If your project is a distributable Python package, you need to include the metrics.yaml file using one of the myriad ways to include data in a Python package and then use pkg_resources.resource_filename() to get the filename at runtime.

from glean import load_metrics
from pkg_resources import resource_filename

metrics = load_metrics(resource_filename(__name__, "metrics.yaml"))

# Use a metric on the returned object
metrics.your_category.your_metric.set("value")

The documentation for your application or library's metrics and pings are written in metrics.yaml and pings.yaml. However, you should also provide human-readable markdown files based on this information, and this is a requirement for Mozilla projects using Glean. For other languages and platforms, this transformation is done automatically as part of the build. However, for Python the integration to automatically generate docs is an additional step.

Glean provides a commandline tool for automatically generating markdown documentation from your metrics.yaml and pings.yaml files. To perform that translation, run glean_parser's translate command:

python3 -m glean_parser translate -f markdown -o docs metrics.yaml pings.yaml

To get more help about the commandline options:

python3 -m glean_parser translate --help

We recommend integrating this step into your project's documentation build. The details of that integration is left to you, since it depends on the documentation tool being used and how your project is set up.

TODO. To be implemented in bug 1643568.

Adding custom pings

Please refer to the custom pings documentation.

Important: as stated before, any new data collection requires documentation and data-review. This is also required for any new metric automatically collected by the Glean SDK.

Parallelism

All of Glean's target languages use a separate worker thread to do most of Glean's work, including any I/O. This thread is fully managed by Glean as an implementation detail. Therefore, users should be free to use the Glean API wherever it is most convenient, without worrying about the performance impact of updating metrics and sending pings.

Since Glean performs disk and networking I/O, it tries to do as much of its work as possible on separate threads and processes. Since there are complex trade-offs and corner cases to support Python parallelism, it is hard to design a one-size-fits-all approach.

Default behavior

When using the Python bindings, most of Glean's work is done on a separate thread, managed by Glean itself. Glean releases the Global Interpreter Lock (GIL), therefore your application's threads should not be in contention with Glean's thread.

Glean installs an atexit handler so the Glean thread can cleanly finish when your application exits. This handler will wait up to 30 seconds for any pending work to complete.

In addition, by default ping uploading is performed in a separate child process. This process will continue to upload any pending pings even after the main process shuts down. This is important for commandline tools where you want to return control to the shell as soon as possible and not be delayed by network connectivity.

Cases where subprocesses aren't possible

The default approach may not work with applications built using PyInstaller or similar tools which bundle an application together with a Python interpreter making it impossible to spawn new subprocesses of that interpreter. For these cases, there is an option to ensure that ping uploading occurs in the main process. To do this, set the allow_multiprocessing parameter on the glean.Configuration object to False.

Using the multiprocessing module

Additionally, the default approach does not work if your application uses the multiprocessing module for parallelism. Glean can not wait to finish its work in a multiprocessing subprocess, since atexit handlers are not supported in that context.
Therefore, if Glean detects that it is running in a multiprocessing subprocess, all of its work that would normally run on a worker thread will run on the main thread. In practice, this should not be a performance issue: since the work is already in a subprocess, it will not block the main process of your application.

Testing metrics

In order to make testing metrics easier 'out of the box', all metrics include a set of test API functions in order to facilitate unit testing. These include functions to test whether a value has been stored, and functions to retrieve the stored value for validation. For more information, please refer to Unit testing Glean metrics.

Adding metadata about your project to the pipeline

In order for data to be collected from your project, its application id must be registered in the pipeline.

File a data engineering bug to enable your product's application id.

The General API

Glean has a minimal API available on its top-level Glean object. This API allows one to enable and disable upload, register custom pings and set experiment data.

Important: The Glean SDK should only be initialized from the main application, not individual libraries.

If you are adding Glean support to a library, you can safely skip this section.

The API

The Glean SDK provides a general API that supports the following operations. See below for language-specific details.

OperationDescriptionNotes
initializeConfigure and initialize the Glean SDK.Initializing the Glean SDK
setUploadEnabledEnable or disable Glean collection and upload.Enabling and disabling Metrics
getUploadEnabledGet whether or not Glean is allowed to record and upload data.
registerPingsRegister custom pings generated from pings.yaml.Custom pings
setExperimentActiveIndicate that an experiment is running.Using the Experiments API
setExperimentInactiveIndicate that an experiment is no longer running..Using the Experiments API

Initializing the Glean SDK

The following steps are required for applications using the Glean SDK, but not libraries.

Note: The initialize function must be called, even if telemetry upload is disabled. Glean needs to perform maintenance tasks even when telemetry is disabled, and because Glean does this as part of its initialization, it is required to always call the initialize function. Otherwise, Glean won't be able to clean up collected data, disable queuing of pre-init tasks, or perform other required operations.

Note: The Glean SDK does not support use across multiple processes, and must only be initialized on the application's main process. Initializing in other processes is a no-op. Additionally, Glean must be initialized on the main (UI) thread of the applications main process. Failure to do so will throw an IllegalThreadStateException.

An excellent place to initialize Glean is within the onCreate method of the class that extends Android's Application class.

import org.mozilla.yourApplication.GleanMetrics.Pings

class SampleApplication : Application() {

    override fun onCreate() {
        super.onCreate()

        // If you have custom pings in your application, you must register them
        // using the following command. This command should be omitted for
        // applications not using custom pings.
        Glean.registerPings(Pings)

        // Initialize the Glean library.
        Glean.initialize(
            applicationContext,
            // Here, `settings()` is a method to get user preferences, specific to
            // your application and not part of the Glean SDK API.
            uploadEnabled = settings().isTelemetryEnabled
        )
    }
}

Once initialized, if uploadEnabled is true, the Glean SDK will automatically start collecting baseline metrics and sending its pings, according to their respective schedules.
If uploadEnabled is false, any persisted metrics, events and pings (other than first_run_date) are cleared, and subsequent calls to record metrics will be no-ops.

The Glean SDK should be initialized as soon as possible, and importantly, before any other libraries in the application start using Glean. Library code should never call Glean.initialize, since it should be called exactly once per application.

Note: if the application has the concept of release channels and knows which channel it is on at run-time, then it can provide the Glean SDK with this information by setting it as part of the Configuration object parameter of the Glean.initialize method. For example:

Glean.initialize(applicationContext, Configuration(channel = "beta"))

Note: When the Glean SDK is consumed through Android Components, it is required to configure an HTTP client to be used for upload. For example:

// Requires `org.mozilla.components:concept-fetch`
import mozilla.components.concept.fetch.Client
// Requires `org.mozilla.components:lib-fetch-httpurlconnection`.
// This can be replaced by other implementations, e.g. `lib-fetch-okhttp`
// or an implementation from `browser-engine-gecko`.
import mozilla.components.lib.fetch.httpurlconnection.HttpURLConnectionClient

import mozilla.components.service.glean.config.Configuration
import mozilla.components.service.glean.net.ConceptFetchHttpUploader

val httpClient = ConceptFetchHttpUploader(lazy { HttpURLConnectionClient() as Client })
val config = Configuration(httpClient = httpClient)
Glean.initialize(context, uploadEnabled = true, configuration = config)

Note: The Glean SDK does not support use across multiple processes, and must only be initialized on the application's main process.

An excellent place to initialize Glean is within the application(_:) method of the class that extends the UIApplicationDelegate class.

import Glean
import UIKit

@UIApplicationMain
class AppDelegate: UIResponder, UIApplicationDelegate {
    func application(_: UIApplication, didFinishLaunchingWithOptions _: [UIApplication.LaunchOptionsKey: Any]?) -> Bool {
        // If you have custom pings in your application, you must register them
        // using the following command. This command should be omitted for
        // applications not using custom pings.
        Glean.shared.registerPings(GleanMetrics.Pings)

        // Initialize the Glean library.
        Glean.shared.initialize(
            // Here, `Settings` is a method to get user preferences specific to
            // your application, and not part of the Glean SDK API.
            uploadEnabled = Settings.isTelemetryEnabled
        )
    }
}

Once initialized, if uploadEnabled is true, the Glean SDK will automatically start collecting baseline metrics and sending its pings, according to their respective schedules.
If uploadEnabled is false, any persisted metrics, events and pings (other than first_run_date) are cleared, and subsequent calls to record metrics will be no-ops.

The Glean SDK should be initialized as soon as possible, and importantly, before any other libraries in the application start using Glean. Library code should never call Glean.shared.initialize, since it should be called exactly once per application.

Note: if the application has the concept of release channels and knows which channel it is on at run-time, then it can provide the Glean SDK with this information by setting it as part of the Configuration object parameter of the Glean.shared.initialize method. For example:

Glean.shared.initialize(Configuration(channel: "beta"))

The main control for Glean is on the glean.Glean singleton.

The Glean SDK should be initialized as soon as possible, and importantly, before any other libraries in the application start using Glean. Library code should never call Glean.initialize, since it should be called exactly once per application.

from glean import Glean

Glean.initialize(
    application_id="my-app-id",
    application_version="0.1.0",
    upload_enabled=True,
)

Once initialized, if upload_enabled is true, the Glean SDK will automatically start collecting baseline metrics and sending its pings, according to their respective schedules.
If upload_enabled is false, any persisted metrics, events and pings (other than first_run_date) are cleared, and subsequent calls to record metrics will be no-ops.

Additional configuration is available on the glean.Configuration object, which can be passed into Glean.initialize().

Unlike Android and Swift, the Python bindings do not automatically send any pings. See the custom pings documentation about adding custom pings and sending them.

The main control for Glean is on the GleanInstance singleton.

The Glean SDK should be initialized as soon as possible, and importantly, before any other libraries in the application start using Glean. Library code should never call Glean.initialize, since it should be called exactly once per application.

using static Mozilla.Glean.Glean;

GleanInstance.Initialize(
    applicationId: "my.app.id",
    applicationVersion: "0.1.1",
    uploadEnabled: true,
    configuration: new Configuration(),
    dataDir: gleanDataDir
    );

Behavior when uninitialized

Metric recording that happens before Glean is initialized is queued and applied when Glean is initialized. To avoid unbounded memory growth the queue is bounded (currently to a maximum of 100 tasks). Further recordings are dropped. The number of recordings dropped, if any, is recorded in the glean.error.preinit_tasks_overflow metric.

Custom ping submission will not fail before initialization. Collection and upload of the custom ping is delayed until Glean is initialized. Built-in pings are only available after initialization.

Enabling and disabling metrics

Glean.setUploadEnabled() should be called in response to the user enabling or disabling telemetry.

Glean.shared.setUploadEnabled() should be called in response to the user enabling or disabling telemetry.

Glean.set_upload_enabled() should be called in response to the user enabling or disabling telemetry.

GleanInstance.SetUploadEnabled() should be called in response to the user enabling or disabling telemetry.

The application should provide some form of user interface to call this method.

When going from enabled to disabled, all pending events, metrics and pings are cleared, except for first_run_date. When re-enabling, core Glean metrics will be recomputed at that time.

Adding new metrics

When adding a new metric, the workflow is:

  • Consider the question you are trying to answer with this data, and choose the metric type and parameters to use.
  • Add a new entry to metrics.yaml.
  • Add code to your project to record into the metric by calling the Glean SDK.

Important: Any new data collection requires documentation and data-review. This is also required for any new metric automatically collected by the Glean SDK.

Choosing a metric type

The following is a set of questions to ask about the data being collected to help better determine which metric type to use.

Is it a single measurement?

If the value is true or false, use a boolean metric.

If the value is a string, use a string metric. For example, to record the name of the default search engine. Beware: string metrics are exceedingly general, and you are probably best served by selecting the most specific metric for the job, since you'll get better error checking and richer analysis tools for free. For example, avoid storing a number in a string metric --- you probably want a counter metric instead.

If you need to store multiple string values in a metric, use a string list metric. For example, you may want to record the list of other Mozilla products installed on the device.

For all of the simple metric types in this section that measure single values, it is especially important to consider how the lifetime of the value relates to the ping it is being sent in. Since these metrics don't perform any aggregation on the client side, when a ping containing the metric is submitted, it will contain only the "last known" value for the metric, potentially resulting in data loss. There is further discussion of metric lifetimes below.

Are you counting things?

If you want to know how many times something happened, use a counter metric. If you are counting a group of related things, or you don't know what all of the things to count are at build time, use a labeled counter metric.

If you need to know when the things being counted happened relative to other things, consider using an event.

Are you measuring time?

If you need to record an absolute time, use a datetime metric. Datetimes are recorded in the user's local time, according to their device's real time clock, along with a timezone offset from UTC. Datetime metrics allow specifying the resolution they are collected at, and to stay lean, they should only be collected at the minimum resolution required to answer your question.

If you need to record how long something takes you have a few options.

If you need to measure the total time spent doing a particular task, look to the timespan metric. Timespan metrics allow specifying the resolution they are collected at, and to stay lean, they should only be collected at the minimum resolution required to answer your question. Note that this metric should only be used to measure time on a single thread. If multiple overlapping timespans are measured for the same metric, an invalid state error is recorded.

If you need to measure the relative occurrences of many timings, use a timing distribution. It builds a histogram of timing measurements, and is safe to record multiple concurrent timespans on different threads.

If you need to know the time between multiple distinct actions that aren't a simple "begin" and "end" pair, consider using an event.

Do you need to know the order of events relative to other events?

If you need to know the order of actions relative to other actions, such as, the user performed tasks A, B, and then C, and this is meaningfully different from the user performing tasks A, C and then B, (in other words, the order is meaningful beyond just the fact that a set of tasks were performed), use an event metric.

Importantly, events are the most expensive metric type to record, transmit, store and analyze, so they should be used sparingly, and only when none of the other metric types are sufficient for answering your question.

For how long do you need to collect this data?

Think carefully about how long the metric will be needed, and set the expires parameter to disable the metric at the earliest possible time. This is an important component of Mozilla's lean data practices.

When the metric passes its expiration date (determined at build time), it will automatically stop collecting data.

When a metric's expiration is within in 14 days, emails will be sent from telemetry-alerts@mozilla.com to the notification_emails addresses associated with the metric. At that time, the metric should be removed, which involves removing it from the metrics.yaml file and removing uses of it in the source code. Removing a metric does not affect the availability of data already collected by the pipeline.

If the metric is still needed after its expiration date, it should go back for another round of data review to have its expiration date extended.

When should Glean automatically clear the measurement?

The lifetime parameter of a metric defines when its value will be cleared. There are three lifetime options available:

  • ping (default): The metric is cleared each time it is submitted in the ping. This is the most common case, and should be used for metrics that are highly dynamic, such as things computed in response to the user's interaction with the application.
  • application: The metric is related to an application run, and is cleared after the application restarts and any Glean-owned ping, due at startup, is submitted. This should be used for things that are constant during the run of an application, such as the operating system version. In practice, these metrics are generally set during application startup. A common mistake---using the ping lifetime for these type of metrics---means that they will only be included in the first ping sent during a particular run of the application.
  • user: Reach out to the Glean team before using this.. The metric is part of the user's profile and will live as long as the profile lives. This is often not the best choice unless the metric records a value that really needs to be persisted for the full lifetime of the user profile, e.g. an identifier like the client_id, the day the product was first executed. It is rare to use this lifetime outside of some metrics that are built in to the Glean SDK.

While lifetimes are important to understand for all metric types, they are particularly important for the metric types that record single values and don't aggregate on the client (boolean, string, labeled_string, string_list, datetime and uuid), since these metrics will send the "last known" value and missing the earlier values could be a form of unintended data loss.

A lifetime example

Let's work through an example to see how these lifetimes play out in practice. Let's suppose we have a user preference, "turbo mode", which defaults to false, but the user can turn it to true at any time. We want to know when this flag is true so we can measure its affect on other metrics in the same ping. In the following diagram, we look at a time period that sends 4 pings across two separate runs of the application. We assume here, that like Glean's built-in metrics ping, the developer writing the metric isn't in control of when the ping is submitted.

In this diagram, the ping measurement windows are represented as rectangles, but the moment the ping is "submitted" is represented by its right edge. The user changes the "turbo mode" setting from false to true in the first run, and then toggles it again twice in the second run.

Metric lifetime timeline

  • A. Ping lifetime, set on change: The value isn't included in Ping 1, because Glean doesn't know about it yet. It is included in the first ping after being recorded (Ping 2), which causes it to be cleared.

  • B. Ping lifetime, set on init and change: The default value is included in Ping 1, and the changed value is included in Ping 2, which causes it to be cleared. It therefore misses Ping 3, but when the application is started, it is recorded again and it is included in Ping 4. However, this causes it to be cleared again and it is not in Ping 5.

  • C. Application lifetime, set on change: The value isn't included in Ping 1, because Glean doesn't know about it yet. After the value is changed, it is included in Pings 2 and 3, but then due to application restart it is cleared, so it is not included until the value is manually toggled again.

  • D. Application, set on init and change: The default value is included in Ping 1, and the changed value is included in Pings 2 and 3. Even though the application startup causes it to be cleared, it is set again, and all subsequent pings also have the value.

  • E. User, set on change: The default value is missing from Ping 1, but since user lifetime metrics aren't cleared unless the user profile is reset (e.g. on Android, when the product is uninstalled), it is included in all subsequent pings.

  • F. User, set on init and change: Since user lifetime metrics aren't cleared unless the user profile is reset, it is included in all subsequent pings. This would be true even if the "turbo mode" preference were never changed again.

Note that for all of the metric configurations, the toggle of the preference off and on during Ping 4 is completely missed. If you need to create a ping containing one, and only one, value for this metric, consider using a custom ping to create a ping whose lifetime matches the lifetime of the value.

What if none of these lifetimes are appropriate?

If the timing at which the metric is sent in the ping needs to closely match the timing of the metrics value, the best option is to use a custom ping to manually control when pings are sent.

This is especially useful when metrics need to be tightly related to one another, for example when you need to measure the distribution of frame paint times when a particular rendering backend is in use. If these metrics were in different pings, with different measurement windows, it is much harder to do that kind of reasoning with much certainty.

What should this new metric be called?

Reuse names from other applications

There's a lot of value using the same name for analogous metrics collected across different products. For example, BigQuery makes it simple to join columns with the same name across multiple tables. Therefore, we encourage you to investigate if a similar metric is already being collected by another product. If it is, there may be an opportunity for code reuse across these products, and if all the projects are using Glean, it's easy for libraries to send their own metrics. If sharing the code doesn't make sense, at a minimum we recommend using the same metric name for similar actions and concepts whenever possible.

Make names unique within an application

Metric identifiers (the combination of a metric's category and name) must be unique across all metrics that are sent by a single application. This includes not only the metrics defined in the app's metrics.yaml, but the metrics.yaml of any Glean-using library that the application uses, including Glean itself. Therefore, care should be taken to name things specifically enough so as to avoid namespace collisions. In practice, this generally involves thinking carefully about the category of the metric, more than the name.

Note: Duplicate metric identifiers are not currently detected at build time. See bug 1578383 for progress on that. However, the probe_scraper process, which runs nightly, will detect duplicate metrics and e-mail the notification_emails associated with the given metrics.

Be as specific as possible

More broadly, you should choose the names of metrics to be as specific as possible. It is not necessary to put the type of the metric in the category or name, since this information is retained in other ways through the entire end-to-end system.

For example, if defining a set of events related to search, put them in a category called search, rather than just events or search_events. The events word here would be redundant.

What if none of these metric types is the right fit?

The current set of metrics Glean supports is based on known common use cases, but new use cases are discovered all the time.

Please reach out to us on #glean:mozilla.org. If you think you need a new metric type, we have a process for that.

How do I make sure my metric is working?

Glean has rich support for writing unit tests involving metrics. Writing a good unit test is a large topic, but in general, you should write unit tests for all new telemetry that does the following:

  • Performs the operation being measured.

  • Asserts that metrics contain the expected data, using the testGetValue API on the metric.

  • Where applicable, asserts that no errors are recorded, such as when values are out of range, using the testGetNumRecordedErrors API.

In addition to unit tests, it is good practice to validate the incoming data for the new metric on a pre-release channel to make sure things are working as expected.

Adding the metric to the metrics.yaml file

The metrics.yaml file defines the metrics your application or library will send. They are organized into categories. The overall organization is:

# Required to indicate this is a `metrics.yaml` file
$schema: moz://mozilla.org/schemas/glean/metrics/1-0-0

toolbar:
  click:
    type: event
    description: |
      Event to record toolbar clicks.
    notification_emails:
      - CHANGE-ME@example.com
    bugs:
      - https://bugzilla.mozilla.org/123456789/
    data_reviews:
      - http://example.com/path/to/data-review
    expires: 2019-06-01  # <-- Update to a date in the future

  double_click:
    ...

category2.subcategory:  # Categories can contain subcategories using `.`
  metric:
    ...

The details of the metric parameters are described in metric parameters.

The metrics.yaml file is used to generate code in the target language (e.g. Kotlin, Swift, ...) that becomes the public API to access your application's metrics.

Using the metric from your code

The reference documentation for each metric type goes into detail about using each metric type from your code.

One thing to note is that we try to adhere to the coding conventions of each language wherever possible, to the metric name in the metrics.yaml (which is in snake_case) may be changed to some other case convention, such as camelCase, when used from code.

Category and metric names in the metrics.yaml are in snake_case, but given the Kotlin coding standards defined by ktlint, these identifiers must be camelCase in Kotlin. For example, the metric defined in the metrics.yaml as:

views:
  login_opened:
    ...

is accessible in Kotlin as:

import org.mozilla.yourApplication.GleanMetrics.Views
GleanMetrics.Views.loginOpened...

Category and metric names in the metrics.yaml are in snake_case, but given the Swift coding standards defined by swiftlint, these identifiers must be camelCase in Swift. For example, the metric defined in the metrics.yaml as:

views:
  login_opened:
    ...

is accessible in Kotlin as:

GleanMetrics.Views.loginOpened...

Category and metric names in the metrics.yaml are in snake_case, which matches the PEP8 standard, so no translation is needed for Python.

TODO. To be implemented in this bug.

Metric parameters

Required metric parameters

  • type: Required. Specifies the type of a metric, like "counter" or "event". This defines which operations are valid for the metric, how it is stored and how data analysis tooling displays it. See the list of supported metric types.

Important: Once a metric is released in a product, its type should not be changed. If any data was collected locally with the older type, and hasn't yet been sent in a ping, recording data with the new type may cause any old persisted data to be lost for that metric. See this comment for an extended explanation of the different scenarios.

  • description: Required. A textual description of the metric for humans. It should describe what the metric does, what it means for analysts, and its edge cases or any other helpful information.

    The description field may contain markdown syntax.

  • notification_emails: Required. A list of email addresses to notify for important events with the metric or when people with context or ownership for the metric need to be contacted.

  • bugs: Required. A list of bugs (e.g. Bugzilla or GitHub) that are relevant to this metric. For example, bugs that track its original implementation or later changes to it.

    Each entry should be the full URL to the bug in an issue tracker. The use of numbers alone is deprecated and will be an error in the future.

  • data_reviews: Required. A list of URIs to any data collection reviews responses relevant to the metric.

  • expires: Required. May be one of the following values:

    • <build date>: An ISO date yyyy-mm-dd in UTC on which the metric expires. For example, 2019-03-13. This date is checked at build time. Except in special cases, this form should be used so that the metric automatically "sunsets" after a period of time. Emails will be sent to the notification_emails addresses when the metric is about to expire. Generally, when a metric is no longer needed, it should simply be removed. This does not affect the availability of data already collected by the pipeline.
    • never: This metric never expires.
    • expired: This metric is manually expired.

Optional metric parameters

  • lifetime: Defines the lifetime of the metric. Different lifetimes affect when the metrics value is reset.

    • ping (default): The metric is cleared each time it is submitted in the ping. This is the most common case, and should be used for metrics that are highly dynamic, such as things computed in response to the user's interaction with the application.
    • application: The metric is related to an application run, and is cleared after the application restarts and any Glean-owned ping, due at startup, is submitted. This should be used for things that are constant during the run of an application, such as the operating system version. In practice, these metrics are generally set during application startup. A common mistake---using the ping lifetime for these type of metrics---means that they will only be included in the first ping sent during a particular run of the application.
    • user: Reach out to the Glean team before using this.. The metric is part of the user's profile and will live as long as the profile lives. This is often not the best choice unless the metric records a value that really needs to be persisted for the full lifetime of the user profile, e.g. an identifier like the client_id, the day the product was first executed. It is rare to use this lifetime outside of some metrics that are built in to the Glean SDK.
  • send_in_pings: Defines which pings the metric should be sent on. If not specified, the metric is sent on the "default ping", which is the events ping for events and the metrics ping for everything else. Most metrics don't need to specify this unless they are sent on custom pings.

  • disabled: (default: false) Data collection for this metric is disabled. This is useful when you want to temporarily disable the collection for a specific metric without removing references to it in your source code. Generally, when a metric is no longer needed, it should simply be removed. This does not affect the availability of data already collected by the pipeline.

  • version: (default: 0) The version of the metric. A monotonically increasing integer value. This should be bumped if the metric changes in a backward-incompatible way.

Unit testing Glean metrics

In order to support unit testing inside of client applications using the Glean SDK, a set of testing API functions have been included. The intent is to make the Glean SDK easier to test 'out of the box' in any client application it may be used in. These functions expose a way to inspect and validate recorded metric values within the client application but are restricted to test code only through visibility annotations (@VisibleForTesting(otherwise = VisibleForTesting.NONE) for Kotlin, internal methods for Swift).

General test API method semantics

In order to prevent issues with async calls when unit testing Glean, it is important to put the Glean SDK into testing mode by applying the JUnit GleanTestRule to your test class. When the Glean SDK is in testing mode, it enables uploading and clears the recorded metrics at the beginning of each test run. The rule can be used as shown below:

@RunWith(AndroidJUnit4::class)
class ActivityCollectingDataTest {
    // Apply the GleanTestRule to set up a disposable Glean instance.
    // Please note that this clears the Glean data across tests.
    @get:Rule
    val gleanRule = GleanTestRule(ApplicationProvider.getApplicationContext())

    @Test
    fun checkCollectedData() {
      // The Glean SDK testing API can be called here.
    }
}

This will ensure that metrics are done recording when the other test functions are used.

To check if a value exists (i.e. it has been recorded), there is a testHasValue() function on each of the metric instances:

assertTrue(GleanMetrics.Search.defaultSearchEngineUrl.testHasValue())

To check the actual values, there is a testGetValue() function on each of the metric instances. It is important to check that the values are recorded as expected, since many of the metric types may truncate or error-correct the value. This function will return a datatype appropriate to the specific type of the metric it is being used with:

assertEquals("https://example.com/search?", GleanMetrics.Search.defaultSearchEngineUrl.testGetValue())

Note that each of these functions has its visibility limited to the scope of unit tests by making use of the @VisibleForTesting annotation, so the IDE should complain if you attempt to use them inside of client code.

NOTE: There's no automatic test rule for Glean tests implemented.

In order to prevent issues with async calls when unit testing Glean, it is important to put the Glean SDK into testing mode. When the Glean SDK is in testing mode, it enables uploading and clears the recorded metrics at the beginning of each test run.

Activate it by resetting Glean in your test's setup:

@testable import Glean
import XCTest

class GleanUsageTests: XCTestCase {
    override func setUp() {
        Glean.shared.resetGlean(clearStores: true)
    }

    // ...
}

This will ensure that metrics are done recording when the other test functions are used.

To check if a value exists (i.e. it has been recorded), there is a testHasValue() function on each of the metric instances:

XCTAssertTrue(GleanMetrics.Search.defaultSearchEngineUrl.testHasValue())

To check the actual values, there is a testGetValue() function on each of the metric instances. It is important to check that the values are recorded as expected, since many of the metric types may truncate or error-correct the value. This function will return a datatype appropriate to the specific type of the metric it is being used with:

XCTAssertEqual("https://example.com/search?", try GleanMetrics.Search.defaultSearchEngineUrl.testGetValue())

Note that each of these functions is marked as internal, you need to import Glean explicitly in test mode:

@testable import Glean

It is generally a good practice to "reset" Glean prior to every unit test that uses Glean, to prevent side effects of one unit test impacting others. Glean contains a helper function glean.testing.reset_glean() for this purpose. It has two required arguments: the application ID, and the application version. Each reset of Glean will create a new temporary directory for Glean to store its data in. This temporary directory is automatically cleaned up the next time Glean is reset or when the testing framework finishes.

The instructions below assume you are using pytest as the test runner. Other test-running libraries have similar features, but are different in the details.

Create a file conftest.py at the root of your test directory, and add the following to reset Glean at the start of every test in your suite:

import pytest
from glean import testing

@pytest.fixture(name="reset_glean", scope="function", autouse=True)
def fixture_reset_glean():
    testing.reset_glean(application_id="my-app-id", application_version="0.1.0")

To check if a value exists (i.e. it has been recorded), there is a test_has_value() function on each of the metric instances:

from glean import load_metrics
metrics = load_metrics("metrics.yaml")

# ...

assert metrics.search.search_engine_url.test_has_value()

To check the actual values, there is a test_get_value() function on each of the metric instances. It is important to check that the values are recorded as expected, since many of the metric types may truncate or error-correct the value. This function will return a datatype appropriate to the specific type of the metric it is being used with:

assert (
    "https://example.com/search?" ==
    metrics.search.default_search_engine_url.test_get_value()
)

TODO. To be implemented in bug 1648448.

Testing metrics for custom pings

In order to test metrics where the metric is included in more than one ping, the test functions take an optional pingName argument (ping_name in Python). This is the name of the ping that the metric is being sent in, such as "events" for the events ping, or "metrics" for the metrics ping. This could also be a custom ping name that the metric is being sent in. In most cases you should not have to supply the ping name to the test function and can just use the default which is the "default" ping that this metric is sent in. You should only need to provide a pingName if the metric is being sent in more than one ping in order to identify the correct metric store.

You can call the testHasValue() and testGetValue() functions with pingName like this:

GleanMetrics.Foo.uriCount.testHasValue("customPing")
GleanMetrics.Foo.uriCount.testGetValue("customPing")

Example of using the test API

Here is a longer example to better illustrate the intended use of the test API:

// Record a metric value with extra to validate against
GleanMetrics.BrowserEngagement.click.record(
    mapOf(
        BrowserEngagement.clickKeys.font to "Courier"
    )
)

// Record more events without extras attached
BrowserEngagement.click.record()
BrowserEngagement.click.record()

// Check if we collected any events into the 'click' metric
assertTrue(BrowserEngagement.click.testHasValue())

// Retrieve a snapshot of the recorded events
val events = BrowserEngagement.click.testGetValue()

// Check if we collected all 3 events in the snapshot
assertEquals(3, events.size)

// Check extra key/value for first event in the list
assertEquals("Courier", events.elementAt(0).extra["font"])

Here is a longer example to better illustrate the intended use of the test API:

// Record a metric value with extra to validate against
GleanMetrics.BrowserEngagement.click.record([.font: "Courier"])

// Record more events without extras attached
BrowserEngagement.click.record()
BrowserEngagement.click.record()

// Check if we collected any events into the 'click' metric
XCTAssertTrue(BrowserEngagement.click.testHasValue())

// Retrieve a snapshot of the recorded events
let events = try! BrowserEngagement.click.testGetValue()

// Check if we collected all 3 events in the snapshot
XCTAssertEqual(3, events.count)

// Check extra key/value for first event in the list
XCTAssertEqual("Courier", events[0].extra?["font"])

Here is a longer example to better illustrate the intended use of the test API:

from glean import load_metrics
metrics = load_metrics("metrics.yaml")

# Record a metric value with extra to validate against
metrics.url.visit.add(1)

# Check if we collected any events into the 'click' metric
assert metrics.url.visit.test_has_value()

# Retrieve a snapshot of the recorded events
assert 1 == metrics.url.visit.test_get_value()

TODO. To be implemented in bug 1648448.

Debugging products using the Glean SDK

Platform-specific debugging

  1. Debugging Android applications using the Glean SDK
  2. Debugging iOS applications using the Glean SDK
  3. Debugging Python applications using the Glean SDK

General debugging information

Available commands and query format

There are 3 available commands that you can use with the Glean SDK debug tools

  • logPings: This is either true or false and will cause pings that are submitted to also be echoed to the device's log
  • tagPings: This command will tag outgoing pings with the provided value, in order to identify them in the Glean Debug View. Tags need to be string with upper and lower case letters, numbers and dashes, with a max length of 20 characters.
  • sendPing: This command expects a string name of a ping to force immediate collection and submission of.

Different platforms have different ways to send these commands.

Important considerations when using Glean SDK debug tools

  • Options that are set using the flags are not immediately reset and will persist until the application is closed or manually reset.

  • There are a couple different ways in which to send pings using the Glean SDK debug tools.

    1. You can tag pings using the debug tools and trigger them manually using the UI. This should always produce a ping with all required fields.
    2. You can tag and send pings using the debug tools. This has the side effect of potentially sending a ping which does not include all fields because sendPing triggers pings to be sent before certain application behaviors can occur which would record that information. For example, duration is not calculated or included in a baseline ping sent with sendPing because it forces the ping to be sent before the duration metric has been recorded. Keep in mind that there may be nothing to send, in which case no ping is generated.
    3. You can trigger a command while the instrumented application is still running. This is useful for toggling commands or for triggering pings that have schedules that are difficult to trigger manually. This is especially useful if you need to trigger a ping submission after some activity within the application, such as with the metrics ping.

Glean Log messages

Glean logs warnings and errors through the platform-specific logging frameworks. See the platform-specific instructions for information on how to view the logs.

Implementation details

See Debug Pings.

Glean Debug Ping View

The Glean Debug Ping View enables you to easily see in real-time what data your application is sending.

This data is what actually arrives in our data pipeline, shown in a web interface that is automatically updated when new data arrives. Any data sent from a Glean SDK-instrumented application usually shows up within 10 seconds, updating the pages automatically. Pings are retained for 3 weeks.

What setup is needed for applications?

You need to tag each ping with a debugging tag. See the documentation on debugging for information on how to do this for each platform and/or language.

Troubleshooting

If nothing is showing up on the dashboard, and you see Glean must be enabled before sending pings. in the logs, then the application has disabled Glean. Check with the application author on how to re-enable it.

Debugging Android applications using the Glean SDK

The Glean SDK exports the GleanDebugActivity that can be used to toggle debugging features on or off. Users can invoke this special activity, at run-time, using the following adb command:

adb shell am start -n [applicationId]/mozilla.telemetry.glean.debug.GleanDebugActivity [extra keys]

Note: In previous versions of Glean the activity was available with the name mozilla.components.service.glean.debug.GleanDebugActivity.

If you're debugging an old build, try running: adb shell am start -n [applicationId]/mozilla.components.service.glean.debug.GleanDebugActivity [extra keys]

In the above:

  • [applicationId] is the product's application id as defined in the manifest file and/or build script. For the Glean sample application, this is org.mozilla.samples.gleancore for a release build and org.mozilla.samples.gleancore.debug for a debug build.

  • [extra keys] is a list of extra keys to be passed to the debug activity. See the documentation for the command line switches used to pass the extra keys. These are the currently supported keys:

keytypedescription
logPingsboolean (--ez)If set to true, pings are dumped to logcat; defaults to false
sendPingstring (--es)Sends the ping with the given name immediately
tagPingsstring (--es)Tags all outgoing pings as debug pings to make them available for real-time validation, on the Glean Debug View. The value must match the pattern [a-zA-Z0-9-]{1,20}

Note: Due to limitations on Android logcat message size, pings larger than 4KB are broken into multiple log messages when using logPings.

For example, to direct a release build of the Glean sample application to (1) dump pings to logcat, (2) tag the ping with the test-metrics-ping tag, and (3) send the "metrics" ping immediately, the following command can be used:

adb shell am start -n org.mozilla.samples.gleancore/mozilla.telemetry.glean.debug.GleanDebugActivity \
  --ez logPings true \
  --es sendPing metrics \
  --es tagPings test-metrics-ping

The logPings command doesn't trigger ping submission and you won't see any output until a ping has been sent. You can use the sendPing command to force a ping to be sent, but it could be more desirable to trigger the pings submission on their normal schedule. For instance, the baseline and events pings can be triggered by moving the app out of the foreground and the metrics ping can be triggered normally if it is overdue for the current calendar day.

Note: The device or emulator must be connected to the internet for this to work. Otherwise the job that sends the pings won't be triggered.

If no metrics have been collected, no pings will be sent unless send_if_empty is set on your ping. See the ping documentation for more information on ping scheduling to learn when pings are sent.

Options that are set using the adb flags are not immediately reset and will persist until the application is closed or manually reset.

Note: In previous versions of Glean the activity was available with the name mozilla.components.service.glean.debug.GleanDebugActivity.

If you're debugging an old build, try running:

adb shell am start -n org.mozilla.samples.gleancore/mozilla.components.service.glean.debug.GleanDebugActivity \
  --ez logPings true \
  --es sendPing metrics \
  --es tagPings test-metrics-ping

Glean Log messages

When running a Glean-powered app in the Android emulator or on a device connected to your computer via cable, there are several ways to read the log output.

Android Studio

Android Studio can show the logs of a connected emulator or device. To display the log messages for an app:

  1. Run an app on your device.
  2. Click View > Tool Windows > Logcat (or click Logcat in the tool window bar).

The Logcat window will show all log messages and allows to filter those by the application ID. Select the application ID of the product you're debugging. You can also filter by Glean only.

More information can be found in the View Logs with Logcat help article.

Command line

On the command line you can show all of the log output using:

adb logcat

This is the unfiltered output of all log messages. You can match for glean using grep:

adb logcat | grep -i glean

A simple way to filter for only the application that is being debugged is by using pidcat, a wrapper around adb, which adds colors and proper filtering by application ID and log level. Run it like this to filter for an application:

pidcat [applicationId]

In the above [applicationId] is the product's application id as defined in the manifest file and/or build script. For the Glean sample application, this is org.mozilla.samples.gleancore for a release build and org.mozilla.samples.gleancore.debug for a debug build.

Debugging iOS applications using the Glean SDK

For debugging and validation purposes on iOS, Glean makes use of a custom URL scheme which is implemented within the application that is consuming Glean. Glean provides some convenience functions to facilitate this, but it's up to the consuming application to enable this functionality. Applications that enable this Glean SDK feature will be able to launch the application from a URL with the Glean debug commands embedded in the URL itself.

Available commands and query format

There are 3 available commands that you can use with the Glean SDK debug tools

  • logPings: This is either true or false and will cause pings that are submitted to also be echoed to the device's log
  • tagPings: This command will tag outgoing pings with the provided value, in order to identify them in the Glean Debug View. Tags need to be string with upper and lower case letters, numbers and dashes, with a max length of 20 characters.
  • sendPing: This command expects a string name of a ping to force immediate collection and submission of.

The structure of the custom URL uses the following format:

<protocol>://glean?<command 1>=<paramter 1>&<command 2>=<parameter 2> ...

Where:

  • <protocol> is the "URL Scheme" that has been added for your app (see Instrumenting the application below), such as glean-sample-app.
  • This is followed by :// and then glean which is required for the Glean SDK to recognize the command is meant for it to process.
  • Following standard URL query format, the next character after glean is the ? indicating the beginning of the query.
  • This is followed by one or more queries in the form of <command>=<parameter>, where the command is one of the commands listed above, followed by an = and then the value or parameter to be used with the command.

There are a few things to consider when creating the custom URL:

  • Invalid commands will log an error and cause the entire URL to be ignored.
  • Not all commands are required to be encoded in the URL, you can mix and match the commands that you need.
  • Multiple instances of commands are not allowed in the same URL and, if present, will cause the entire URL to be ignored.
  • The logPings command doesn't trigger ping submission and you won't see any output until a ping has been submitted. You can use the sendPings command to force a ping to be sent, but it could be more desirable to trigger the pings submission on their normal schedule. For instance, the baseline and events pings can be triggered by moving the app out of the foreground and the metrics ping can be triggered normally if it is overdue for the current calendar day. See the ping documentation for more information on ping scheduling to learn when pings are sent.

Instrumenting the application for Glean debug functionality

In order to enable the debugging features in a Glean SDK consuming iOS application, it is necessary to add some information to the application's Info.plist, and add a line and possibly an override for a function in the AppDelegate.swift.

Register custom URL scheme in Info.plist

Note: If your application already has a custom URL scheme implemented, there is no need to implement a second scheme, you can simply use that and skip to the next section about adding the convenience method. If the app doesn't have a custom URL scheme implemented, then you will need to perform the following instructions to register your app to receive custom URLs.

Find and open the application's Info.plist and right click any blank area and select Add Row to create a new key.

You will be prompted to select a key from a drop-down menu, scroll down to and select URL types. This creates an array item, which can be expanded by clicking the triangle disclosure icon.

Select Item 0, click on it and click the disclosure icon to expand it and show the URL identifier line. Double-click the value field and fill in your identifier, typically the same as the bundle ID.

Right-click on Item 0 and select Add Row from the context menu. In the dropdown menu, select URL Schemes to add the item.

Click on the disclosure icon of URL Schemes to expand the item, double-click the value field of Item 0 and key in the value for your application's custom scheme. For instance, the Glean sample app uses glean-sample-app, which allows for custom URLs to be crafted using that as a protocol, for example: glean-sample-app://glean?logPings=true

Add the Glean.handleCustomUrl() convenience function and necessary overrides

In order to handle the incoming Glean SDK debug commands, it is necessary to implement the override in the application's AppDelegate.swift file. Within that function, you can make use of the convenience function provided in Glean handleCustomUrl(url: URL).

An example of a simple implementation of this would look like this:

func application(_: UIApplication,
                 open url: URL,
                 options _: [UIApplication.OpenURLOptionsKey: Any] = [:]) -> Bool {
    // ...

    // This does nothing if the url isn't meant for Glean.
    Glean.shared.handleCustomUrl(url: url)

    // ...

    return true
}

If you need additional help setting up a custom URL scheme in your application, please refer to Apple's documentation.

Invoking the Glean-iOS debug commands

Now that the app has the Glean SDK debug functionality enabled, there are a few ways in which we can invoke the debug commands.

Using a web browser

Perhaps the simplest way to invoke the Glean SDK debug functionality is to open a web browser and type/paste the custom URL into the address bar. This is especially useful on an actual device because there isn't a good way to launch from the command line and process the URL for an actual device.

Using the glean-sample-app as an example: to activate ping logging, tag the pings to go to the Glean Debug View, and force the events ping to be sent, enter the following URL in a web browser on the iOS device:

glean-sample-app://glean?logPings=true&tagPings=My-ping-tag&sendPing=events

This should cause iOS to prompt you with a dialog asking if you want to open the URL in the Glean Sample App, and if you select "Okay" then it will launch (or resume if it's already running) the application with the indicated commands and parameters and immediately force the collection and submission of the events ping.

Note: This method does not work if the browser you are using to input the command is the same application you are attempting to pass the Glean debug commands to. So, you couldn't use Firefox for iOS to trigger commands within Firefox for iOS.

It is also possible to encode the URL into a 2D barcode or QR code and launch the app via the camera app. After scanning the encoded URL, the dialog prompting to launch the app should appear as if the URL were entered into the browser address bar.

Using the command line

This method is useful for testing via the Simulator, which typically requires a Mac with Xcode installed, including the Xcode command line tools. In order to perform the same command as above with using the browser to input the URL, you can use the following command in the command line terminal of the Mac:

xcrun simctl openurl booted "glean-sample-app://glean?logPings=true&tagPings=My-ping-tag&sendPing=events"

This will launch the simulator and again prompt the user with a dialog box asking if you want to open the URL in the Glean Sample App (or whichever app you are instrumenting and testing).

Glean log messages

If debugging in the simulator, the logging messages can be seen in the console window within Xcode.

When running a Glean-powered app in the iOS Simulator or on a device connected to your computer via cable, the following steps will help in getting the logs:

  1. Plug in the device and open Xcode
  2. Choose Window -> Devices from the menu bar
  3. Under the DEVICES section in the left column, choose the device
  4. To see the device console, click the up-triangle at the bottom left of the right hand panel

Note: It is helpful to filter the console window by adding the keyword "glean" to the filter box. This will allow you to see only messages generated by the Glean SDK.

You can refer to this Apple technical Q&A for more information about debugging deployed iOS apps.

Debugging Python applications using the Glean SDK

Glean provides a couple of configuration flags to assist with debugging Python applications.

Tagging pings

The Glean.configuration.ping_tag property can be used to add a special flag to the HTTP header so that the ping will end up in the Glean Debug View.

You can set it after Glean.initialize is called:

from Glean import Glean, Configuration
Glean.initialize(
    application_id="my-app-id",
    application_version="0.1.0",
    upload_enabled=True,
)

# ...

Glean.configuration.ping_tag = "my-ping-tag"

After doing so, something like pings.custom_ping.submit() will send the custom ping to the Glean Debug View.

Logging pings

If the Glean.configuration.log_pings property is set to True, pings are logged to the console on DEBUG level whenever they are submitted. You can set this property in a similar way as the ping_tag property above.

Make sure that when you configure logging in your application, you set the level for the glean logger to DEBUG or higher. Otherwise pings won't be logged even if log_pings is set to True.

You can set the logging level for Glean to DEBUG as follows:

import logging

logging.getLogger("glean").setLevel(logging.DEBUG)

See the Python logging documentation for more information.

Error reporting

The Glean SDK records the number of errors that occur when metrics are passed invalid data or are otherwise used incorrectly. This information is reported back in special labeled counter metrics in the glean.error category. Error metrics are included in the same pings as the metric that caused the error. Additionally, error metrics are always sent in the metrics ping ping.

The following categories of errors are recorded:

  • invalid_value: The metric value was invalid.
  • invalid_label: The label on a labeled metric was invalid.
  • invalid_state: The metric caught an invalid state while recording.
  • invalid_overflow: The metric value to be recorded overflows the metric-specific upper range.

For example, if you had a string metric and passed it a string that was too long:

MyMetrics.stringMetric.set("this_string_is_longer_than_the_limit_for_string_metrics")

The following error metric counter would be incremented:

Glean.error.invalidValue["my_metrics.string_metric"].add(1)

Resulting in the following keys in the ping:

{
  "metrics": {
    "labeled_counter": {
      "glean.error.invalid_value": {
        "my_metrics.string_metric": 1
      }
    }
  }
}

If you have a debug build of the Glean SDK, details about the errors being recorded are included in the logs. This detailed information is not included in Glean pings.

Using the experiments API

The Glean SDK supports tagging all its pings with experiments annotations. The annotations are useful to report that experiments were active at the time the measurement were collected. The annotations are reported in the optional experiments entry in the ping_info section of all the Glean SDK pings.

Important: the experiment annotations set through this API are not persisted by the Glean SDK. The application or consuming library is responsible for setting the relevant experiment annotations at each run.

API

// Annotate Glean pings with experiments data.
Glean.setExperimentActive(
  experimentId = "blue-button-effective",
  branch = "branch-with-blue-button",
  extra: mapOf(
    "buttonLabel" to "test"
  )
)
// After the experiment terminates, the annotation
// can be removed.
Glean.setExperimentInactive("blue-button-effective")

Important: Experiment IDs and branch don't need to be pre-defined in the Glean SDK registry files. Please also note that the extra map is non-nested arbitrary String to String map.

There are test APIs available too:

// Was the experiment annotated in Glean pings?
assertTrue(Glean.testIsExperimentActive("blue-button-effective"))
// Was the correct branch reported?
assertEquals(
  "branch-with-blue-button", Glean.testGetExperimentData("blue-button-effective")?.branch
)
// Annotate Glean pings with experiments data.
Glean.shared.setExperimentActive(
  experimentId: "blue-button-effective",
  branch: "branch-with-blue-button",
  extra: ["buttonLabel": "test"]
)
// After the experiment terminates, the annotation
// can be removed.
Glean.shared.setExperimentInactive(experimentId: "blue-button-effective")

Important: Experiment IDs and branch don't need to be pre-defined in the Glean SDK registry files. Please also note that the extra is a non-nested Dictionary of type [String: String].

There are test APIs available too:

// Was the experiment annotated in Glean pings?
XCTAssertTrue(Glean.shared.testIsExperimentActive(experimentId: "blue-button-effective"))
// Was the correct branch reported?
XCTAssertEqual(
  "branch-with-blue-button",
  Glean.testGetExperimentData(experimentId: "blue-button-effective")?.branch
)
from glean import Glean

# Annotate Glean pings with experiments data.
Glean.set_experiment_active(
  experiment_id="blue-button-effective",
  branch="branch-with-blue-button",
  extra={
    "buttonLabel": "test"
  }
)

# After the experiment terminates, the annotation
# can be removed.
Glean.set_experiment_inactive("blue-button-effective")

Important: Experiment IDs and branch don't need to be pre-defined in the Glean SDK registry files. Please also note that the extra dict is non-nested arbitrary str to str mapping.

There are test APIs available too:

from glean import Glean

# Was the experiment annotated in Glean pings?
assert Glean.test_is_experiment_active("blue-button-effective")
# Was the correct branch reported?
assert (
    "branch-with-blue-button" ==
    Glean.test_get_experiment_data("blue-button-effective").branch
)

TODO. To be implemented in bug 1648412.

Limits

  • experimentId, branch, and the keys and values of the 'extra' field are fixed at a maximum length of 100 bytes. Longer strings used as ids, keys, or values are truncated to their respective maximum length. (Specifically, this is measured in the number of bytes when the string is encoded in UTF-8.)
  • extra map is limited to 20 entries. If passed a map which contains more elements than this, it is truncated to 20 elements. WARNING Which items are truncated is nondeterministic due to the unordered nature of maps and what's left may not necessarily be the first elements added.

NOTE: Any truncation that occurs when recording to the Experiments API will result in an invalid_value error being recorded. See Error Reporting for more information about this type of error.

Reference

Metrics

Not sure which metric type to use? These docs contain a series of questions that can help. Reference information about each metric type is linked below.

There are different metrics to choose from, depending on what you want to achieve:

  • Boolean: Records a single truth value, for example "is a11y enabled?"

  • Counter: Used to count how often something happens, for example, how often a certain button was pressed.

  • Labeled counter: Used to count how often something happens, for example which kind of crash occurred ("uncaught_exception" or "native_code_crash").

  • String: Records a single Unicode string value, for example the name of the OS.

  • Labeled strings: Records multiple Unicode string values, for example to record which kind of error occurred in different stages of a login process.

  • String List: Records a list of Unicode string values, for example the list of enabled search engines.

  • Timespan: Used to measure how much time is spent in a single task.

  • Timing Distribution: Used to record the distribution of multiple time measurements.

  • Memory Distribution: Used to record the distribution of memory sizes.

  • UUID: Used to record universally unique identifiers (UUIDs), such as a client ID.

  • Datetime: Used to record an absolute date and time, such as the time the user first ran the application.

  • Events: Records events e.g. individual occurrences of user actions, say every time a view was open and from where.

  • Custom Distribution: Used to record the distribution of a value that needs fine-grained control of how the histogram buckets are computed. Custom distributions are only available for values that come from Gecko.

  • Quantity: Used to record a single non-negative integer value. For example, the width of the display in pixels. Quantities are only available for values that come from Gecko.

Labeled metrics

There are two types of metrics listed above - labeled and unlabeled metrics. If a metric is labeled, it means that for a single metric entry you define in metrics.yaml, you can record into multiple metrics under the same name, each of the same type and identified by a different string label.

This is useful when you need to break down metrics by a label known at build time or run time. For example:

  • When you want to count a different set of sub-views that users interact with, you could use viewCount["view1"].add() and viewCount["view2"].add().

  • When you want to count errors that might occur for a feature, you could use errorCount[errorName].add().

Labeled metrics come in two forms:

  • Static labels: The labels are specified at build time in the metrics.yaml file. If a label that isn't part of this set is used at run time, it is converted to the special label __other__.

  • Dynamic labels: The labels aren't known at build time, so are set at run time. Only the first 16 labels seen by the Glean SDK will be tracked. After that, any additional labels are converted to the special label __other__.

Note: Be careful with using arbitrary strings as labels and make sure they can't accidentally contain identifying data (like directory paths or user input).

Label format

To ensure maximum support in database columns, labels must be made up of dot-separated identifiers with lowercase ASCII alphanumerics, containing underscores and dashes.

Specifically, they must conform to this regular expression:

^[a-z_][a-z0-9_-]{0,29}(\\.[a-z_][a-z0-9_-]{0,29})*$

Adding or changing metric types

Glean has a well-defined process for requesting changes to existing metric types or suggesting the implementation of new metric types:

  1. Glean consumers need to file a bug in the Data platforms & tools::Glean Metric Types component, filling in the provided form;
  2. The triage owner of the Bugzilla component prioritizes this within 6 business days and kicks off the decision making process.
  3. Once the decision process is completed, the bug is closed with a comment outlining the decision that was made.

Boolean

Booleans are used for simple flags, for example "is a11y enabled"?.

Configuration

Say you're adding a boolean to record whether a11y is enabled on the device. First you need to add an entry for the boolean to the metrics.yaml file:

flags:
  a11y_enabled:
    type: boolean
    description: >
      Records whether a11y is enabled on the device.
    lifetime: application
    ...

API

import org.mozilla.yourApplication.GleanMetrics.Flags

Flags.a11yEnabled.set(System.isAccesibilityEnabled())

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.Flags

// Was anything recorded?
assertTrue(Flags.a11yEnabled.testHasValue())
// Does it have the expected value?
assertTrue(Flags.a11yEnabled.testGetValue())
import org.mozilla.yourApplication.GleanMetrics.Flags;

Flags.INSTANCE.a11yEnabled.set(System.isAccessibilityEnabled());

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.Flags;

// Was anything recorded?
assertTrue(Flags.INSTANCE.a11yEnabled.testHasValue());
// Does it have the expected value?
assertTrue(Flags.INSTANCE.a11yEnabled.testGetValue());
Flags.a11yEnabled.set(self.isAccessibilityEnabled)

There are test APIs available too:

@testable import Glean

// Was anything recorded?
XCTAssertTrue(Flags.a11yEnabled.testHasValue())
// Does the counter have the expected value?
XCTAssertTrue(try Flags.a11yEnabled.testGetValue())
from glean import load_metrics
metrics = load_metrics("metrics.yaml")

metrics.flags.a11y_enabled.set(is_accessibility_enabled())

There are test APIs available too:

# Was anything recorded?
assert metrics.flags.a11y_enabled.test_has_value()
# Does it have the expected value?
assert True is metrics.flags.a11y_enabled.test_get_value()
using static Mozilla.YourApplication.GleanMetrics.FlagsOuter;

Flags.a11yEnabled.Set(System.IsAccessibilityEnabled());

There are test APIs available too:

using static Mozilla.YourApplication.GleanMetrics.FlagsOuter;

// Was anything recorded?
Assert.True(Flags.a11yEnabled.TestHasValue());
// Does it have the expected value?
Assert.True(Flags.a11yEnabled.TestGetValue());

Limits

  • None.

Examples

  • Is a11y enabled?

Recorded errors

  • None.

Reference

Labeled Booleans

Labeled booleans are used to record different related boolean flags.

Configuration

For example, you may want to record a set of flags related to accessibility (a11y).

accessibility:
  features:
    type: labeled_boolean
    description: >
      a11y features enabled on the device. ...
    labels:
      - screen_reader
      - high_contrast
    ...

Note: removing or changing labels, including their order in the registry file, is permitted. Avoid reusing labels that were removed in the past. It is best practice to add documentation about removed labels to the description field so that analysts will know of their existence and meaning in historical data. Special care must be taken when changing GeckoView metrics sent through the Glean SDK, as the index of the labels is used to report Gecko data through the Glean SDK.

API

Now you can use the labeled boolean from the application's code:

import org.mozilla.yourApplication.GleanMetrics.Accessibility
Accessibility.features["screen_reader"].set(isScreenReaderEnabled())
Accessibility.features["high_contrast"].set(isHighContrastEnabled())

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.Accessibility
// Was anything recorded?
assertTrue(Accessibility.features["screen_reader"].testHasValue())
assertTrue(Accessibility.features["high_contrast"].testHasValue())
// Do the booleans have the expected values?
assertEquals(True, Accessibility.features["screen_reader"].testGetValue())
assertEquals(False, Accessibility.features["high_contrast"].testGetValue())
// Did we record any invalid labels?
assertEquals(0, Accessibility.features.testGetNumRecordedErrors(ErrorType.InvalidLabel))
Accessibility.features["screen_reader"].set(isScreenReaderEnabled())
Accessibility.features["high_contrast"].set(isHighContrastEnabled())

There are test APIs available too:

@testable import Glean

// Was anything recorded?
XCTAssert(Accessibility.features["screen_reader"].testHasValue())
XCTAssert(Accessibility.features["high_contrast"].testHasValue())
// Do the booleans have the expected values?
XCTAssertEqual(true, try Accessibility.features["screen_reader"].testGetValue())
XCTAssertEqual(false, try Accessibility.features["high_contrast"].testGetValue())
// Were there any invalid labels?
XCTAssertEqual(0, Accessibility.features.testGetNumRecordedErrors(.invalidLabel))
from glean import load_metrics
metrics = load_metrics("metrics.yaml")

metrics.accessibility.features["screen_reader"].set(
    is_screen_reader_enabled()
)
metrics.accessibility.features["high_contrast"].set(
    is_high_contrast_enabled()
)

There are test APIs available too:

# Was anything recorded?
assert metrics.accessibility.features["screen_reader"].test_has_value()
assert metrics.accessibility.features["high_contrast"].test_has_value()
# Do the booleans have the expected values?
assert metrics.accessibility.features["screen_reader"].test_get_value()
assert not metrics.accessibility.features["high_contrast"].test_get_value()
# Did we record any invalid labels?
assert 0 == metrics.accessibility.features.test_get_num_recorded_errors(
    ErrorType.INVALID_LABEL
)

TODO. To be implemented in bug 1648436.

Limits

  • Labels must conform to the label formatting regular expression.

  • Labels support lowercase alphanumeric characters; they additionally allow for dots (.), underscores (_) and/or hyphens (-).

  • Each label must have a maximum of 60 bytes, when encoded as UTF-8.

  • If the labels are specified in the metrics.yaml, using any label not listed in that file will be replaced with the special value __other__.

  • If the labels aren't specified in the metrics.yaml, only 16 different dynamic labels may be used, after which the special value __other__ will be used.

Examples

  • Record a related set of boolean flags.

Recorded Errors

  • invalid_label: If the label contains invalid characters.

  • invalid_label: If the label exceeds the maximum number of allowed characters.

Reference

Counter

Used to count how often something happens, say how often a certain button was pressed. A counter always starts from 0. Each time you record to a counter, its value is incremented. Unless incremented by a positive value, a counter will not be reported in pings.

IMPORTANT: When using a counter metric, it is important to let the Glean metric do the counting. Using your own variable for counting and setting the counter yourself could be problematic because it will be difficult to reset the value at the exact moment that the value is sent in a ping. Instead, just use counter.add to increment the value and let Glean handle resetting the counter.

Configuration

Say you're adding a new counter for how often the refresh button is pressed. First you need to add an entry for the counter to the metrics.yaml file:

controls:
  refresh_pressed:
    type: counter
    description: >
      Counts how often the refresh button is pressed.
    ...

API

import org.mozilla.yourApplication.GleanMetrics.Controls

Controls.refreshPressed.add() // Adds 1 to the counter.
Controls.refreshPressed.add(5) // Adds 5 to the counter.

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.Controls

// Was anything recorded?
assertTrue(Controls.refreshPressed.testHasValue())
// Does the counter have the expected value?
assertEquals(6, Controls.refreshPressed.testGetValue())
// Did the counter record an negative value?
assertEquals(
    1, Controls.refreshPressed.testGetNumRecordedErrors(ErrorType.InvalidValue)
)
import org.mozilla.yourApplication.GleanMetrics.Controls;

Controls.INSTANCE.refreshPressed.add(); // Adds 1 to the counter.
Controls.INSTANCE.refreshPressed.add(5); // Adds 5 to the counter.

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.Controls;

// Was anything recorded?
assertTrue(Controls.INSTANCE.refreshPressed.testHasValue());
// Does the counter have the expected value?
assertEquals(6, Controls.INSTANCE.refreshPressed.testGetValue());
// Did the counter record an negative value?
assertEquals(
    1, Controls.INSTANCE.refreshPressed.testGetNumRecordedErrors(ErrorType.InvalidValue)
);
Controls.refreshPressed.add() // Adds 1 to the counter.
Controls.refreshPressed.add(5) // Adds 5 to the counter.

There are test APIs available too:

@testable import Glean

// Was anything recorded?
XCTAssert(Controls.refreshPressed.testHasValue())
// Does the counter have the expected value?
XCTAssertEqual(6, try Controls.refreshPressed.testGetValue())
// Did the counter record a negative value?
XCTAssertEqual(1, Controls.refreshPressed.testGetNumRecordedErrors(.invalidValue))
from glean import load_metrics
metrics = load_metrics("metrics.yaml")

metrics.controls.refresh_pressed.add()  # Adds 1 to the counter.
metrics.controls.refresh_pressed.add(5) # Adds 5 to the counter.

There are test APIs available too:

# Was anything recorded?
assert metrics.controls.refresh_pressed.test_has_value()
# Does the counter have the expected value?
assert 6 == metrics.controls.refresh_pressed.test_get_value()
# Did the counter record an negative value?
from glean.testing import ErrorType
assert 1 == metrics.controls.refresh_pressed.test_get_num_recorded_errors(
    ErrorType.INVALID_VALUE
)
using static Mozilla.YourApplication.GleanMetrics.Controls;

Controls.refreshPressed.Add(); // Adds 1 to the counter.
Controls.refreshPressed.Add(5); // Adds 5 to the counter.

There are test APIs available too:

using static Mozilla.YourApplication.GleanMetrics.Controls;

// Was anything recorded?
Assert.True(Controls.refreshPressed.TestHasValue());
// Does the counter have the expected value?
Assert.Equal(6, Controls.refreshPressed.TestGetValue());
// Did the counter record an negative value?
Assert.Equal(
    1, Controls.refreshPressed.TestGetNumRecordedErrors(ErrorType.InvalidValue)
);

Limits

  • Only increments, saturates at the largest value that can be represented as a 32-bit signed integer (2147483647).

Examples

  • How often was a certain button was pressed?

Recorded errors

  • invalid_value: If the counter is incremented by 0 or a negative value.

Reference

Labeled Counters

Labeled counters are used to record different related counts that should sum up to a total.

Configuration

For example, you may want to record a count of different types of crashes for your Android application, such as native code crashes and uncaught exceptions:

stability:
  crash_count:
    type: labeled_counter
    description: >
      Counts the number of crashes that occur in the application. ...
    labels:
      - uncaught_exception
      - native_code_crash
    ...

Note: removing or changing labels, including their order in the registry file, is permitted. Avoid reusing labels that were removed in the past. It is best practice to add documentation about removed labels to the description field so that analysts will know of their existence and meaning in historical data. Special care must be taken when changing GeckoView metrics sent through the Glean SDK, as the index of the labels is used to report Gecko data through the Glean SDK.

API

Now you can use the labeled counter from the application's code:

import org.mozilla.yourApplication.GleanMetrics.Stability
Stability.crashCount["uncaught_exception"].add() // Adds 1 to the "uncaught_exception" counter.
Stability.crashCount["native_code_crash"].add(3) // Adds 3 to the "native_code_crash" counter.

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.Stability
// Was anything recorded?
assertTrue(Stability.crashCount["uncaught_exception"].testHasValue())
assertTrue(Stability.crashCount["native_code_crash"].testHasValue())
// Do the counters have the expected values?
assertEquals(1, Stability.crashCount["uncaught_exception"].testGetValue())
assertEquals(3, Stability.crashCount["native_code_crash"].testGetValue())
// Were there any invalid labels?
assertEquals(0, Stability.crashCount.testGetNumRecordedErrors(ErrorType.InvalidLabel))
Stability.crashCount["uncaught_exception"].add() // Adds 1 to the "uncaught_exception" counter.
Stability.crashCount["native_code_crash"].add(3) // Adds 3 to the "native_code_crash" counter.

There are test APIs available too:

@testable import Glean

// Was anything recorded?
XCTAssert(Stability.crashCount["uncaught_exception"].testHasValue())
XCTAssert(Stability.crashCount["native_code_crash"].testHasValue())
// Do the counters have the expected values?
XCTAssertEqual(1, try Stability.crashCount["uncaught_exception"].testGetValue())
XCTAssertEqual(3, try Stability.crashCount["native_code_crash"].testGetValue())
// Were there any invalid labels?
XCTAssertEqual(0, Stability.crashCount.testGetNumRecordedErrors(.invalidLabel))
from glean import load_metrics
metrics = load_metrics("metrics.yaml")

# Adds 1 to the "uncaught_exception" counter.
metrics.stability.crash_count["uncaught_exception"].add()
# Adds 3 to the "native_code_crash" counter.
metrics.stability.crash_count["native_code_crash"].add(3)

There are test APIs available too:

# Was anything recorded?
assert metrics.stability.crash_count["uncaught_exception"].test_has_value()
assert metrics.stability.crash_count["native_code_crash"].test_has_value()
# Do the counters have the expected values?
assert 1 == metrics.stability.crash_count["uncaught_exception"].test_get_value()
assert 3 == metrics.stability.crash_count["native_code_crash"].test_get_value()
# Were there any invalid labels?
assert 0 == metrics.stability.crash_count.test_get_num_recorded_errors(
    ErrorType.INVALID_LABEL
)

TODO. To be implemented in bug 1648437.

Limits

  • Labels must conform to the label formatting regular expression.

  • Labels support lowercase alphanumeric characters; they additionally allow for dots (.), underscores (_) and/or hyphens (-).

  • Each label must have a maximum of 60 bytes, when encoded as UTF-8.

  • If the labels are specified in the metrics.yaml, using any label not listed in that file will be replaced with the special value __other__.

  • If the labels aren't specified in the metrics.yaml, only 16 different dynamic labels may be used, after which the special value __other__ will be used.

Examples

  • Record the number of times different kinds of crashes occurred.

Recorded Errors

  • invalid_label: If the label contains invalid characters.

  • invalid_label: If the label exceeds the maximum number of allowed characters.

Reference

Strings

This allows recording a Unicode string value with arbitrary content.

Note: Be careful using arbitrary strings and make sure they can't accidentally contain identifying data (like directory paths or user input).

Note: This is does not support recording JSON blobs - please get in contact with the Telemetry team if you're missing a type.

Configuration

Say you're adding a metric to find out what the default search in a browser is. First you need to add an entry for the metric to the metrics.yaml file:

search.default:
  name:
    type: string
    description: >
      The name of the default search engine.
    lifetime: application
    ...

API

import org.mozilla.yourApplication.GleanMetrics.SearchDefault

// Record a value into the metric.
SearchDefault.name.set("duck duck go")
// If it changed later, you can record the new value:
SearchDefault.name.set("wikipedia")

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.SearchDefault

// Was anything recorded?
assertTrue(SearchDefault.name.testHasValue())
// Does the string metric have the expected value?
// IMPORTANT: It may have been truncated -- see "Limits" below
assertEquals("wikipedia", SearchDefault.name.testGetValue())
// Was the string truncated, and an error reported?
assertEquals(1, SearchDefault.name.testGetNumRecordedErrors(ErrorType.InvalidValue))
import org.mozilla.yourApplication.GleanMetrics.SearchDefault;

// Record a value into the metric.
SearchDefault.INSTANCE.name.set("duck duck go");
// If it changed later, you can record the new value:
SearchDefault.INSTANCE.name.set("wikipedia");

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.SearchDefault

// Was anything recorded?
assertTrue(SearchDefault.INSTANCE.name.testHasValue());
// Does the string metric have the expected value?
// IMPORTANT: It may have been truncated -- see "Limits" below
assertEquals("wikipedia", SearchDefault.INSTANCE.name.testGetValue());
// Was the string truncated, and an error reported?
assertEquals(
    1,
    SearchDefault.INSTANCE.name.testGetNumRecordedErrors(
        ErrorType.InvalidValue
    )
);
// Record a value into the metric.
SearchDefault.name.set("duck duck go")
// If it changed later, you can record the new value:
SearchDefault.name.set("wikipedia")

There are test APIs available too:

@testable import Glean

// Was anything recorded?
XCTAssert(SearchDefault.name.testHasValue())
// Does the string metric have the expected value?
// IMPORTANT: It may have been truncated -- see "Limits" below
XCTAssertEqual("wikipedia", try SearchDefault.name.testGetValue())
// Was the string truncated, and an error reported?
XCTAssertEqual(1, SearchDefault.name.testGetNumRecordedErrors(.invalidValue))
from glean import load_metrics
metrics = load_metrics("metrics.yaml")

# Record a value into the metric.
metrics.search_default.name.set("duck duck go")
# If it changed later, you can record the new value:
metrics.search_default.name.set("wikipedia")

There are test APIs available too:

# Was anything recorded?
assert metrics.search_default.name.test_has_value()
# Does the string metric have the expected value?
# IMPORTANT: It may have been truncated -- see "Limits" below
assert "wikipedia" == metrics.search_default.name.test_get_value()
# Was the string truncated, and an error reported?
assert 1 == metrics.search_default.name.test_get_num_recorded_errors(
    ErrorType.INVALID_VALUE
)
using static Mozilla.YourApplication.GleanMetrics.SearchDefault;

// Record a value into the metric.
SearchDefault.name.Set("duck duck go");
// If it changed later, you can record the new value:
SearchDefault.name.Set("wikipedia");

There are test APIs available too:

using static Mozilla.YourApplication.GleanMetrics.SearchDefault;

// Was anything recorded?
Assert.True(SearchDefault.name.TestHasValue());
// Does the string metric have the expected value?
// IMPORTANT: It may have been truncated -- see "Limits" below
Assert.Equal("wikipedia", SearchDefault.name.TestGetValue());
// Was the string truncated, and an error reported?
Assert.Equal(
    1,
    SearchDefault.name.TestGetNumRecordedErrors(
        ErrorType.InvalidValue
    )
);

Limits

  • Fixed maximum string length: 100. Longer strings are truncated. This is measured in the number of bytes when the string is encoded in UTF-8.

Examples

  • Record the operating system name with a value of "android".

  • Recording the device model with a value of "SAMSUNG-SGH-I997".

Recorded errors

  • invalid_value: if the string is too long

Reference

Labeled Strings

Labeled strings record multiple Unicode string values, each under a different label.

Configuration

For example to record which kind of error occurred in different stages of a login process - "RuntimeException" in the "server_auth" stage or "invalid_string" in the "enter_email" stage:

login:
  errors_by_stage:
    type: labeled_string
    description: Records the error type, if any, that occur in different stages of the login process.
    labels:
      - server_auth
      - enter_email
    ...

API

Now you can use the labeled string from the application's code:

import org.mozilla.yourApplication.GleanMetrics.Login

Login.errorsByStage["server_auth"].set("Invalid password")

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.Login

// Was anything recorded?
assertTrue(Login.errorsByStage["server_auth"].testHasValue())

// Were there any invalid labels?
assertEquals(0, Login.errorsByStage.testGetNumRecordedErrors(ErrorType.InvalidLabel))
Login.errorsByStage["server_auth"].set("Invalid password")

There are test APIs available too:

@testable import Glean

// Was anything recorded?
XCTAssert(Login.errorsByStage["server_auth"].testHasValue())

// Were there any invalid labels?
XCTAssertEqual(0, Login.errorsByStage.testGetNumRecordedErrors(.invalidLabel))
from glean import load_metrics
metrics = load_metrics("metrics.yaml")

metrics.login.errors_by_stage["server_auth"].set("Invalid password")

There are test APIs available too:

# Was anything recorded?
assert metrics.login.errors_by_stage["server_auth"].test_has_value()

# Were there any invalid labels?
assert 0 == metrics.login.errors_by_stage.test_get_num_recorded_errors(
    ErrorType.INVALID_LABEL
)

TODO. To be implemented in bug 1648438.

Limits

  • Labels must conform to the label formatting regular expression.

  • Each label must have a maximum of 60 bytes, when encoded as UTF-8.

  • If the labels are specified in the metrics.yaml, using any label not listed in that file will be replaced with the special value __other__.

  • If the labels aren't specified in the metrics.yaml, only 16 different dynamic labels may be used, after which the special value __other__ will be used.

Examples

  • What kind of errors occurred at each step in the login process?

Recorded Errors

  • invalid_label: If the label contains invalid characters.

  • invalid_label: If the label exceeds the maximum number of allowed characters.

Reference

String List

Strings lists are used for recording a list of Unicode string values, such as the names of the enabled search engines.


Note: Be careful using arbitrary strings and make sure they can't accidentally contain identifying data (like directory paths or user input).


Configuration

First you need to add an entry for the counter to the metrics.yaml file:

search:
  engines:
    type: string_list
    description: >
      Records the name of the enabled search engines.
    lifetime: application
    ...

API

import org.mozilla.yourApplication.GleanMetrics.Search

// Add them one at a time
engines.forEach {
  Search.engines.add(it)
}

// Set them in one go
Search.engines.set(engines)

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.Search

// Was anything recorded?
assertTrue(Search.engines.testHasValue())
// Does it have the expected value?
// IMPORTANT: It may have been truncated -- see "Limits" below
assertEquals(listOf("Google", "DuckDuckGo"), Search.engines.testGetValue())
// Were any of the values too long, and thus an error was recorded?
assertEquals(1, Search.engines.testGetNumRecordedErrors(ErrorType.InvalidValue))
// Add them one at a time
for engine in engines {
    Search.engines.add(engine)
}

// Set them in one go
Search.engines.set(engines)

There are test APIs available too:

@testable import Glean

// Was anything recorded?
XCTAssert(Search.engines.testHasValue())
// Does it have the expected value?
// IMPORTANT: It may have been truncated -- see "Limits" below
XCTAssertEqual(["Google", "DuckDuckGo"], try Search.engines.testGetValue())
// Were any of the values too long, and thus an error was recorded?
XCTAssertEqual(1, Search.engines.testGetNumRecordedErrors(.invalidValue))
from glean import load_metrics

metrics = load_metrics("metrics.yaml")

# Add them one at a time
for engine in engines:
    metrics.search.engines.add(engine)

# Set them in one go
metrics.search.engines.set(engines)

There are test APIs available too:

# Was anything recorded?
assert metrics.search.engines.test_has_value()
# Does it have the expected value?
# IMPORTANT: It may have been truncated -- see "Limits" below
assert ["Google", "DuckDuckGo"] == metrics.search.engines.test_get_value()
# Were any of the values too long, and thus an error was recorded?
assert 1 == metrics.search.engines.test_get_num_recorded_errors(
    ErrorType.INVALID_VALUE
)

TODO. To be implemented in bug 1648441.

Limits

  • Fixed maximum string length: 50. Longer strings are truncated. This is measured in the number of bytes when the string is encoded in UTF-8.

  • Fixed maximum list length: 20 items. Additional strings are dropped.

Examples

  • The names of the enabled search engines.

Recorded errors

  • invalid_value: if the string is too long

  • invalid_value: if the list is too long

Reference

Timespan

Timespans are used to make a measurement of how much time is spent in a particular task.

To measure the distribution of multiple timespans, see Timing Distributions. To record absolute times, see Datetimes.

It is not recommended to use timespans in multiple threads, since calling start or stop out of order will be recorded as an invalid_state error.

Configuration

Timespans have a required time_unit parameter to specify the smallest unit of resolution that the timespan will record. The allowed values for time_unit are:

  • nanosecond
  • microsecond
  • millisecond
  • second
  • minute
  • hour
  • day

Consider the resolution that is required by your metric, and use the largest possible value that will provide useful information so as to not leak too much fine-grained information from the client. It is important to note that the value sent in the ping is truncated down to the nearest unit. Therefore, a measurement of 500 nanoseconds will be truncated to 0 microseconds.

Say you're adding a new timespan for the time spent logging into the app. First you need to add an entry for the counter to the metrics.yaml file:

auth:
  login_time:
    type: timespan
    description: >
      Measures the time spent logging in.
    time_unit: milliseconds
    ...

API

import org.mozilla.yourApplication.GleanMetrics.Auth

fun onShowLogin() {
    Auth.loginTime.start()
    // ...
}

fun onLogin() {
    Auth.loginTime.stop()
    // ...
}

fun onLoginCancel() {
    Auth.loginTime.cancel()
    // ...
}

The time reported in the telemetry ping will be timespan recorded during the lifetime of the ping.

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.Auth

// Was anything recorded?
assertTrue(Auth.loginTime.testHasValue())
// Does the timer have the expected value
assertTrue(Auth.loginTime.testGetValue() > 0)
// Was the timing recorded incorrectly?
assertEquals(1, Auth.loginTime.testGetNumRecordedErrors(ErrorType.InvalidValue))
import org.mozilla.yourApplication.GleanMetrics.Auth;

void onShowLogin() {
    Auth.INSTANCE.loginTime.start();
    // ...
}

void onLogin() {
    Auth.INSTANCE.loginTime.stop();
    // ...
}

void onLoginCancel() {
    Auth.INSTANCE.loginTime.cancel();
    // ...
}

The time reported in the telemetry ping will be timespan recorded during the lifetime of the ping.

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.Auth;

// Was anything recorded?
assertTrue(Auth.INSTANCE.loginTime.testHasValue());
// Does the timer have the expected value
assertTrue(Auth.INSTANCE.loginTime.testGetValue() > 0);
// Was the timing recorded incorrectly?
assertEquals(
    1,
    Auth.INSTANCE.loginTime.testGetNumRecordedErrors(
        ErrorType.InvalidValue
    )
);
func onShowLogin() {
    Auth.loginTime.start()
    // ...
}

func onLogin() {
    Auth.loginTime.stop()
    // ...
}

func onLoginCancel() {
    Auth.loginTime.cancel()
    // ...
}

The time reported in the telemetry ping will be timespan recorded during the lifetime of the ping.

There are test APIs available too:

@testable import Glean

// Was anything recorded?
XCTAssert(Auth.loginTime.testHasValue())
// Does the timer have the expected value
XCTAssert(try Auth.loginTime.testGetValue() > 0)
// Was the timing recorded incorrectly?
XCTAssertEqual(1, Auth.loginTime.testGetNumRecordedErrors(.invalidValue))
from glean import load_metrics
metrics = load_metrics("metrics.yaml")

def on_show_login():
    metrics.auth.login_time.start()
    # ...

def on_login():
    metrics.auth.login_time.stop()
    # ...

def on_login_cancel():
    metrics.auth.login_time.cancel()
    # ...

The time reported in the telemetry ping will be timespan recorded during the lifetime of the ping.

There are test APIs available too:

# Was anything recorded?
assert metrics.auth.login_time.test_has_value()
# Does the timer have the expected value
assert metrics.auth.login_time.test_get_value() > 0
# Was the timing recorded incorrectly?
assert 1 == metrics.auth.local_time.test_get_num_recorded_errors(
    ErrorType.INVALID_VALUE
)

TODO. To be implemented in bug 1648442.

Raw API

Note: The raw API was designed to support a specific set of use-cases. Please consider using the higher level APIs listed above.

It's possible to explicitly set the timespan value, in nanoseconds. This API should only be used if your library or application requires recording times in a way that can not make use of start/stop/cancel.

The raw API will not overwrite a running timer or existing timespan value.

import org.mozilla.yourApplication.GleanMetrics.HistorySync

val duration = SyncResult.status.syncs.took.toLong()
HistorySync.setRawNanos(duration)
let duration = SyncResult.status.syncs.took.toLong()
HistorySync.setRawNanos(duration)
import org.mozilla.yourApplication.GleanMetrics.HistorySync

val duration = SyncResult.status.syncs.took.toLong()
HistorySync.setRawNanos(duration)

TODO. To be implemented in bug 1648442.

Limits

  • Timings are recorded in nanoseconds.

    • On Android, the SystemClock.elapsedRealtimeNanos() function is used, so it is limited by the accuracy and performance of that timer. The time measurement includes time spent in sleep.

    • On iOS, the mach_absolute_time function is used, so it is limited by the accuracy and performance of that timer. The time measurement does not include time spent in sleep.

    • On Python 3.7 and later, time.monotonic_ns() is used. On earlier versions of Python, time.monotonics() is used, which is not guaranteed to have nanosecond resolution.

Examples

  • How much time is spent rendering the UI?

Recorded errors

  • invalid_value
    • If recording a negative timespan.
  • invalid_state
    • If starting a timer while a previous timer is running.
    • If stopping a timer while it is not running.
    • If trying to set a raw timespan while a timer is running.

Reference

Timing Distribution

Timing distributions are used to accumulate and store time measurement, for analyzing distributions of the timing data.

To measure the distribution of single timespans, see Timespans. To record absolute times, see Datetimes.

Timing distributions are recorded in a histogram where the buckets have an exponential distribution, specifically with 8 buckets for every power of 2. That is, the function from a value \( x \) to a bucket index is:

\[ \lfloor 8 \log_2(x) \rfloor \]

This makes them suitable for measuring timings on a number of time scales without any configuration.

Timings always span the full length between start and stopAndAccumulate. If the Glean upload is disabled when calling start, the timer is still started. If the Glean upload is disabled at the time stopAndAccumulate is called, nothing is recorded.

Multiple concurrent timespans in different threads may be measured at the same time.

Timings are always stored and sent in the payload as nanoseconds. However, the time_unit parameter controls the minimum and maximum values that will recorded:

  • nanosecond: 1ns <= x <= 10 minutes
  • microsecond: 1μs <= x <= ~6.94 days
  • millisecond: 1ms <= x <= ~19 years

Overflowing this range is considered an error and is reported through the error reporting mechanism. Underflowing this range is not an error and the value is silently truncated to the minimum value.

Additionally, when a metric comes from GeckoView (the geckoview_datapoint parameter is present), the time_unit parameter specifies the unit that the samples are in when passed to Glean. Glean will convert all of the incoming samples to nanoseconds internally.

Configuration

If you wanted to create a timing distribution to measure page load times, first you need to add an entry for it to the metrics.yaml file:

pages:
  page_load:
    type: timing_distribution
    description: >
      Counts how long each page takes to load
    ...

API

Now you can use the timing distribution from the application's code. Starting a timer returns a timer ID that needs to be used to stop or cancel the timer at a later point. Multiple intervals can be measured concurrently. For example, to measure page load time on a number of tabs that are loading at the same time, each tab object needs to store the running timer ID.

import mozilla.components.service.glean.GleanTimerId
import org.mozilla.yourApplication.GleanMetrics.Pages

val timerId : GleanTimerId

fun onPageStart(e: Event) {
    timerId = Pages.pageLoad.start()
}

fun onPageLoaded(e: Event) {
    Pages.pageLoad.stopAndAccumulate(timerId)
}

There are test APIs available too. For convenience, properties sum and count are exposed to facilitate validating that data was recorded correctly.

Continuing the pageLoad example above, at this point the metric should have a sum == 11 and a count == 2:

import org.mozilla.yourApplication.GleanMetrics.Pages

// Was anything recorded?
assertTrue(pages.pageLoad.testHasValue())

// Get snapshot.
val snapshot = pages.pageLoad.testGetValue()

// Does the sum have the expected value?
assertEquals(11, snapshot.sum)

// Usually you don't know the exact timing values, but how many should have been recorded.
assertEquals(2L, snapshot.count)

// Was an error recorded?
assertEquals(1, pages.pageLoad.testGetNumRecordedErrors(ErrorType.InvalidValue))
import mozilla.components.service.glean.GleanTimerId;
import org.mozilla.yourApplication.GleanMetrics.Pages;

GleanTimerId timerId;

void onPageStart(Event e) {
    timerId = Pages.INSTANCE.pageLoad.start();
}

void onPageLoaded(Event e) {
    Pages.INSTANCE.pageLoad.stopAndAccumulate(timerId);
}

There are test APIs available too. For convenience, properties sum and count are exposed to facilitate validating that data was recorded correctly.

Continuing the pageLoad example above, at this point the metric should have a sum == 11 and a count == 2:

import org.mozilla.yourApplication.GleanMetrics.Pages;

// Was anything recorded?
assertTrue(pages.INSTANCE.pageLoad.testHasValue());

// Get snapshot.
DistributionData snapshot = pages.INSTANCE.pageLoad.testGetValue();

// Does the sum have the expected value?
assertEquals(11, snapshot.getSum);

// Usually you don't know the exact timing values, but how many should have been recorded.
assertEquals(2L, snapshot.getCount);

// Was an error recorded?
assertEquals(
    1,
    pages.INSTANCE.pageLoad.testGetNumRecordedErrors(
        ErrorType.InvalidValue
    )
);
import Glean

var timerId : GleanTimerId

func onPageStart() {
    timerId = Pages.pageLoad.start()
}

func onPageLoaded() {
    Pages.pageLoad.stopAndAccumulate(timerId)
}

There are test APIs available too. For convenience, properties sum and count are exposed to facilitate validating that data was recorded correctly.

Continuing the pageLoad example above, at this point the metric should have a sum == 11 and a count == 2:

@testable import Glean

// Was anything recorded?
XCTAssert(pages.pageLoad.testHasValue())

// Get snapshot.
let snapshot = try! pages.pageLoad.testGetValue()

// Does the sum have the expected value?
XCTAssertEqual(11, snapshot.sum)

// Usually you don't know the exact timing values, but how many should have been recorded.
XCTAssertEqual(2, snapshot.count)

// Was an error recorded?
XCTAssertEqual(1, pages.pageLoad.testGetNumRecordedErrors(.invalidValue))
from glean import load_metrics
metrics = load_metrics("metrics.yaml")

class PageHandler:
    def __init__(self):
        self.timer_id = None

    def on_page_start(self, event):
        # ...
        self.timer_id = metrics.pages.page_load.start()

    def on_page_loaded(self, event):
        # ...
        metrics.pages.page_load.stop_and_accumulate(self.timer_id)

There are test APIs available too. For convenience, properties sum and count are exposed to facilitate validating that data was recorded correctly.

Continuing the page_load example above, at this point the metric should have a sum == 11 and a count == 2:

# Was anything recorded?
assert metrics.pages.page_load.test_has_value()

# Get snapshot.
snapshot = metrics.pages.page_load.test_get_value()

# Does the sum have the expected value?
assert 11 == snapshot.sum

# Usually you don't know the exact timing values, but how many should have been recorded.
assert 2 == snapshot.count

# Was an error recorded?
assert 1 == metrics.pages.page_load.test_get_num_recorded_errors(
    ErrorType.INVALID_VALUE
)

TODO. To be implemented in bug 1648443.

Limits

  • Timings are recorded in nanoseconds.

    • On Android, the SystemClock.elapsedRealtimeNanos() function is used, so it is limited by the accuracy and performance of that timer. The time measurement includes time spent in sleep.

    • On iOS, the mach_absolute_time function is used, so it is limited by the accuracy and performance of that timer. The time measurement does not include time spent in sleep.

    • On Python 3.7 and later, time.monotonic_ns() is used. On earlier versions of Python, time.monotonics() is used, which is not guaranteed to have nanosecond resolution.

  • The maximum timing value that will be recorded depends on the time_unit parameter:

    • nanosecond: 1ns <= x <= 10 minutes
    • microsecond: 1μs <= x <= ~6.94 days
    • millisecond: 1ms <= x <= ~19 years Longer times will be truncated to the maximum value and an error will be recorded.

Examples

  • How long does it take a page to load?

Recorded errors

  • invalid_value: If recording a negative timespan.
  • invalid_state: If a non-existing/stopped timer is stopped again.
  • invalid_overflow: If recording a time longer than the maximum for the given unit.

Reference

Memory Distribution

Memory distributions are used to accumulate and store memory sizes.

Memory distributions are recorded in a histogram where the buckets have an exponential distribution, specifically with 16 buckets for every power of 2. That is, the function from a value \( x \) to a bucket index is:

\[ \lfloor 16 \log_2(x) \rfloor \]

This makes them suitable for measuring memory sizes on a number of different scales without any configuration.

Configuration

If you wanted to create a memory distribution to measure the amount of heap memory allocated, first you need to add an entry for it to the metrics.yaml file:

memory:
  heap_allocated:
    type: memory_distribution
    description: >
      The heap memory allocated
    memory_unit: kilobyte
    ...

API

Now you can use the memory distribution from the application's code.

For example, to measure the distribution of heap allocations:

import org.mozilla.yourApplication.GleanMetrics.Memory

fun allocateMemory(nbytes: Int) {
    // ...
    Memory.heapAllocated.accumulate(nbytes / 1024)
}

There are test APIs available too. For convenience, properties sum and count are exposed to facilitate validating that data was recorded correctly.

Continuing the heapAllocated example above, at this point the metric should have a sum == 11 and a count == 2:

import org.mozilla.yourApplication.GleanMetrics.Memory

// Was anything recorded?
assertTrue(Memory.heapAllocated.testHasValue())

// Get snapshot
val snapshot = Memory.heapAllocated.testGetValue()

// Does the sum have the expected value?
assertEquals(11, snapshot.sum)

// Usually you don't know the exact memory values, but how many should have been recorded.
assertEquals(2L, snapshot.count)

// Did this record a negative value?
assertEquals(1, Memory.heapAllocated.testGetNumRecordedErrors(ErrorType.InvalidValue))
func allocateMemory(nbytes: UInt64) {
    // ...
    Memory.heapAllocated.accumulate(nbytes / 1024)
}

There are test APIs available too. For convenience, properties sum and count are exposed to facilitate validating that data was recorded correctly.

Continuing the heapAllocated example above, at this point the metric should have a sum == 11 and a count == 2:

@testable import Glean

// Was anything recorded?
XCTAssert(Memory.heapAllocated.testHasValue())

// Get snapshot
let snapshot = try! Memory.heapAllocated.testGetValue()

// Does the sum have the expected value?
XCTAssertEqual(11, snapshot.sum)

// Usually you don't know the exact memory values, but how many should have been recorded.
XCTAssertEqual(2, snapshot.count)

// Did this record a negative value?
XCTAssertEqual(1, Memory.heapAllocated.testGetNumRecordedErrors(.invalidValue))
from glean import load_metrics
metrics = load_metrics("metrics.yaml")

def allocate_memory(nbytes):
    # ...
    metrics.memory.heap_allocated.accumulate(nbytes / 1024)

There are test APIs available too. For convenience, properties sum and count are exposed to facilitate validating that data was recorded correctly.

Continuing the heapAllocated example above, at this point the metric should have a sum == 11 and a count == 2:

# Was anything recorded?
assert metrics.memory.head_allocated.test_has_value()

# Get snapshot
snapshot = metrics.memory.heap_allocated.test_get_value()

# Does the sum have the expected value?
assert 11 == snapshot.sum

# Usually you don't know the exact memory values, but how many should have been recorded.
assert 2 == snapshot.count

# Did this record a negative value?
assert 1 == metrics.memory.heap_allocated.test_get_num_recorded_errors(
    ErrorType.INVALID_VALUE
)

TODO. To be implemented in bug 1648439.

Limits

  • The maximum memory size that can be recorded is 1 Terabyte (240 bytes). Larger sizes will be truncated to 1 Terabyte.

Examples

  • What is the distribution of the size of heap allocations?

Recorded errors

  • invalid_value: If recording a negative memory size.
  • invalid_value: If recording a size larger than 1TB.

Reference

UUID

UUIDs are used to record values that uniquely identify some entity, such as a client id.

Configuration

You first need to add an entry for it to the metrics.yaml file:

user:
  client_id:
    type: uuid
    description: >
      A unique identifier for the client's profile
    lifetime: user
    ...

API

Now that the UUID is defined in metrics.yaml, you can use the metric to record values in the application's code.

import org.mozilla.yourApplication.GleanMetrics.User

User.clientId.generateAndSet() // Generate a new UUID and record it
User.clientId.set(UUID.randomUUID())  // Set a UUID explicitly

There are test APIs available too.

import org.mozilla.yourApplication.GleanMetrics.User

// Was anything recorded?
assertTrue(User.clientId.testHasValue())
// Was it the expected value?
assertEquals(uuid, User.clientId.testGetValue())
import org.mozilla.yourApplication.GleanMetrics.User;

User.INSTANCE.clientId.generateAndSet(); // Generate a new UUID and record it
User.INSTANCE.clientId.set(UUID.randomUUID());  // Set a UUID explicitly

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.User;

// Was anything recorded?
assertTrue(User.INSTANCE.clientId.testHasValue());
// Was it the expected value?
assertEquals(uuid, User.INSTANCE.clientId.testGetValue());
User.clientId.generateAndSet() // Generate a new UUID and record it
User.clientId.set(UUID())  // Set a UUID explicitly

There are test APIs available too.

@testable import Glean

// Was anything recorded?
XCTAssert(User.clientId.testHasValue())
// Was it the expected value?
XCTAssertEqual(uuid, try User.clientId.testGetValue())
import uuid

from glean import load_metrics
metrics = load_metrics("metrics.yaml")

# Generate a new UUID and record it
metrics.user.client_id.generate_and_set()
# Set a UUID explicitly
metrics.user.client_id.set(uuid.uuid4())

There are test APIs available too.

# Was anything recorded?
assert metrics.user.client_id.test_has_value()
# Was it the expected value?
assert uuid == metrics.user.client_id.test_get_value()
using static Mozilla.YourApplication.GleanMetrics.User;

User.clientId.GenerateAndSet(); // Generate a new UUID and record it
User.clientId.Set(System.Guid.NewGuid()); // Set a UUID explicitly

There are test APIs available too:

using static Mozilla.YourApplication.GleanMetrics.User;

// Was anything recorded?
Assert.True(User.clientId.TestHasValue());
// Was it the expected value?
Assert.Equal(uuid, User.clientId.TestGetValue());

Limits

  • None.

Examples

  • A unique identifier for the client.

Recorded errors

  • None.

Reference

Datetime

Datetimes are used to record an absolute date and time, for example the date and time that the application was first run.

The device's offset from UTC is recorded and sent with the Datetime value in the ping.

To record a single elapsed time, see Timespan. To measure the distribution of multiple timespans, see Timing Distributions.

Configuration

Datetimes have a required time_unit parameter to specify the smallest unit of resolution that the timespan will record. The allowed values for time_unit are:

  • nanosecond
  • microsecond
  • millisecond
  • second
  • minute
  • hour
  • day

Carefully consider the required resolution for recording your metric, and choose the coarsest resolution possible.

You first need to add an entry for it to the metrics.yaml file:

install:
  first_run:
    type: datetime
    time_unit: day
    description: >
      Records the date when the application was first run
    lifetime: user
    ...

API

import org.mozilla.yourApplication.GleanMetrics.Install

Install.firstRun.set() // Records "now"
Install.firstRun.set(Calendar(2019, 3, 25)) // Records a custom datetime

There are test APIs available too.

import org.mozilla.yourApplication.GleanMetrics.Install

// Was anything recorded?
assertTrue(Install.firstRun.testHasValue())
// Was it the expected value?
// NOTE: Datetimes always include a timezone offset from UTC, hence the
// "-05:00" suffix.
assertEquals("2019-03-25-05:00", Install.firstRun.testGetValueAsString())
// Was the value invalid?
assertEquals(1, Install.firstRun.testGetNumRecordedErrors(ErrorType.InvalidValue))
import org.mozilla.yourApplication.GleanMetrics.Install;

Install.INSTANCE.firstRun.set(); // Records "now"
Install.INSTANCE.firstRun.set(Calendar(2019, 3, 25)); // Records a custom datetime

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.Install;

// Was anything recorded?
assertTrue(Install.INSTANCE.firstRun.testHasValue());
// Was it the expected value?
// NOTE: Datetimes always include a timezone offset from UTC, hence the
// "-05:00" suffix.
assertEquals("2019-03-25-05:00", Install.INSTANCE.firstRun.testGetValueAsString());
// Was the value invalid?
assertEquals(1, Install.INSTANCE.firstRun.testGetNumRecordedErrors(ErrorType.InvalidValue));
Install.firstRun.set() // Records "now"

let dateComponents = DateComponents(
                        calendar: Calendar.current,
                        year: 2004, month: 12, day: 9, hour: 8, minute: 3, second: 29
                     )
Install.firstRun.set(dateComponents.date!) // Records a custom datetime

There are test APIs available too:

@testable import Glean

// Was anything recorded?
XCTAssert(Install.firstRun.testHasValue())
// Does the datetime have the expected value?
XCTAssertEqual(6, try Install.firstRun.testGetValue())
// Was the value invalid?
XCTAssertEqual(1, Install.firstRun.getNumRecordedErrors(.invalidValue))
import datetime

from glean import load_metrics
metrics = load_metrics("metrics.yaml")

# Records "now"
metrics.install.first_run.set()
# Records a custom datetime
metrics.install.first_run.set(datetime.datetime(2019, 3, 25))

There are test APIs available too.

# Was anything recorded?
assert metrics.install.first_run.test_has_value()

# Was it the expected value?
# NOTE: Datetimes always include a timezone offset from UTC, hence the
# "-05:00" suffix.
assert "2019-03-25-05:00" == metrics.install.first_run.test_get_value_as_str()
# Was the value invalid?
assert 1 == metrics.install.test_get_num_recorded_errors(
    ErrorType.INVALID_VALUE
)
using static Mozilla.YourApplication.GleanMetrics.Install;

// Records "now"
Install.firstRun.Set();
// Records a custom datetime
Install.firstRun.Set(new DateTimeOffset(2018, 2, 25, 11, 10, 0, TimeZone.CurrentTimeZone.BaseUtcOffset));

There are test APIs available too:

using static Mozilla.YourApplication.GleanMetrics.Install;

// Was anything recorded?
Assert.True(Install.firstRun.TestHasValue());
// Was it the expected value?
// NOTE: Datetimes always include a timezone offset from UTC, hence the
// "-05:00" suffix.
Assert.Equal("2019-03-25-05:00", Install.firstRun.TestGetValueAsString());
// Was the value invalid?
Assert.Equal(1, Install.firstRun.TestGetNumRecordedErrors(ErrorType.InvalidValue));

Limits

  • None.

Examples

  • When did the user first run the application?

Recorded errors

  • invalid_value: Setting the date time to an invalid value.

Reference

Events

Events allow recording of e.g. individual occurrences of user actions, say every time a view was open and from where. Each time you record an event, it records a timestamp, the event's name and a set of custom values.

Configuration

Say you're adding a new event for when a view is shown. First you need to add an entry for the event to the metrics.yaml file:

views:
  login_opened:
    type: event
    description: >
      Recorded when the login view is opened.
    ...
    extra_keys:
      source_of_login:
        description: The source from which the login view was opened, e.g. "toolbar".

API

Note that an enum has been generated for handling the extra_keys: it has the same name as the event metric, with Keys added.

import org.mozilla.yourApplication.GleanMetrics.Views

Views.loginOpened.record(mapOf(Views.loginOpenedKeys.sourceOfLogin to "toolbar"))

There are test APIs available too, for example:

import org.mozilla.yourApplication.GleanMetrics.Views

// Was any event recorded?
assertTrue(Views.loginOpened.testHasValue())
// Get a List of the recorded events.
val snapshot = Views.loginOpened.testGetValue()
// Check that two events were recorded.
assertEquals(2, snapshot.size)
val first = snapshot.single()
assertEquals("login_opened", first.name)
// Check that no errors were recorded
assertEquals(0, Views.loginOpened.testGetNumRecordedErrors(ErrorType.InvalidValue))

Note that an enum has been generated for handling the extra_keys: it has the same name as the event metric, with Keys added.

Views.loginOpened.record(extra: [.sourceOfLogin: "toolbar"])

There are test APIs available too, for example:

@testable import Glean

// Was any event recorded?
XCTAssert(Views.loginOpened.testHasValue())
// Get a List of the recorded events.
val snapshot = try! Views.loginOpened.testGetValue()
// Check that two events were recorded.
XCTAssertEqual(2, snapshot.size)
val first = snapshot[0]
XCTAssertEqual("login_opened", first.name)
// Check that no errors were recorded
XCTAssertEqual(0, Views.loginOpened.testGetNumRecordedErrors(.invalidValue))

Note that an enum has been generated for handling the extra_keys: it has the same name as the event metric, with _keys added.

from glean import load_metrics
metrics = load_metrics("metrics.yaml")

metrics.views.login_opened.record(
    {
        metrics.views.login_opened_keys.SOURCE_OF_LOGIN: "toolbar"
    }
)

There are test APIs available too, for example:

# Was any event recorded?
assert metrics.views.login_opened.test_has_value()
# Get a List of the recorded events.
snapshot = metrics.views.login_opened.test_get_value()
# Check that two events were recorded.
assert 2 == len(snapshot)
first = snapshot[0]
assert "login_opened" == first.name
# Check that no errors were recorded
assert 0 == metrics.views.login_opened.test_get_num_recorded_errors(
    ErrorType.INVALID_VALUE
)

TODO. To be implemented in bug 1648422.

Limits

  • When 500 events are queued on the client an events ping is immediately sent.

  • The extra_keys allows for a maximum of 10 keys.

  • The keys in the extra_keys list must be in dotted snake case, with a maximum length of 40 bytes in UTF-8.

  • The values in the extras object have a maximum length of 50 in UTF-8.

Examples

  • Every time a new tab is opened.

Recorded errors

  • invalid_value: if any of the values in the extras object are greater than 50 bytes in length.

Reference

Custom Distribution

Custom distributions are used to record the distribution of arbitrary values.

It should be used only when direct control over how the histogram buckets are computed is required. Otherwise, look at the standard distribution metric types:

Note: Custom distributions are currently only allowed for GeckoView metrics (the gecko_datapoint parameter is present) and thus have only a Kotlin API.

Configuration

Custom distributions have the following required parameters:

  • range_min: (Integer) The minimum value of the first bucket
  • range_max: (Integer) The minimum value of the last bucket
  • bucket_count: (Integer) The number of buckets
  • histogram_type:
    • linear: The buckets are evenly spaced
    • exponential: The buckets follow a natural logarithmic distribution

In addition, the metric should specify:

  • unit: (String) The unit of the values in the metric. For documentation purposes only -- does not affect data collection.

If you wanted to create a custom distribution of the peak number of pixels used during a checkerboard event, first you need to add an entry for it to the metrics.yaml file:

graphics:
  checkerboard_peak:
    type: custom_distribution
    description: >
      Peak number of CSS pixels checkerboarded during a checkerboard event.
    range_min: 1
    range_max: 66355200
    bucket_count: 50
    histogram_type: exponential
    unit: pixels
    gecko_datapoint: CHECKERBOARD_PEAK
    ...

API

Now you can use the custom distribution from the application's code.

import org.mozilla.yourApplication.GleanMetrics.Graphics

Graphics.checkerboardPeak.accumulateSamples([23])

There are test APIs available too. For convenience, properties sum and count are exposed to facilitate validating that data was recorded correctly.

import org.mozilla.yourApplication.GleanMetrics.Graphics

// Was anything recorded?
assertTrue(Graphics.checkerboardPeak.testHasValue())

// Get snapshot
val snapshot = Graphics.checkerboardPeak.testGetValue()

// Does the sum have the expected value?
assertEquals(11, snapshot.sum)

// Usually you don't know the exact timing values, but how many should have been recorded.
assertEquals(2L, snapshot.count())

/// Did the metric receive a negative value?
assertEquals(1, Graphics.checkerboardPeak.testGetNumRecordedErrors(ErrorType.InvalidValue))

Limits

  • The maximum value of bucket_count is 100.

  • Only non-negative values may be recorded.

Recorded errors

  • invalid_value: If recording a negative value.

Reference

Quantity

Used to record a single non-negative integer value. For example, the width of the display in pixels.

Note: Quantities are currently only allowed for GeckoView metrics (the gecko_datapoint parameter is present) and thus have only a Kotlin API.

Configuration

Say you're adding a new quantity for the width of the display in pixels. First you need to add an entry for the quantity to the metrics.yaml file:

gfx:
  display_width:
    type: quantity
    description: >
      The width of the display, in pixels.
    unit: pixels
    gecko_datapoint: DISPLAY_W_PIXELS
    ...

Note that quantities have a required unit parameter, which is a free-form string for documentation purposes.

API

import org.mozilla.yourApplication.GleanMetrics.Gfx

Gfx.displayWidth.set(width)

There are test APIs available too:

import org.mozilla.yourApplication.GleanMetrics.Gfx

// Was anything recorded?
assertTrue(Gfx.displayWidth.testHasValue())
// Does the quantity have the expected value?
assertEquals(6, Gfx.displayWidth.testGetValue())
// Did it record an error due to a negative value?
assertEquals(1, Gfx.displayWidth.testGetNumRecordedErrors(ErrorType.InvalidValue))

Limits

  • Quantities must be non-negative integers.

  • Quantities are only available for metrics that come from Gecko.

Examples

  • What is the width of the display, in pixels?

Recorded errors

  • invalid_value: If a negative value is passed in.

Reference

Glean Pings

Every Glean ping is in JSON format and contains one or more of the common sections with shared information data.

If data collection is enabled, the Glean SDK provides a set of built-in pings that are assembled out of the box without any developer intervention. The following is a list of these built-in pings:

  • baseline ping: A small ping sent every time the application goes to foreground and background. Going to foreground also includes when the application starts.
  • metrics ping: The default ping for metrics. Sent approximately daily.
  • events ping: The default ping for events. Sent every time the application goes to background or a certain number of events is reached.
  • deletion-request ping: Sent when the user disables telemetry in order to request a deletion of their data.

Applications can also define and send their own custom pings when the schedules of these pings is not suitable.

There is also a high-level overview of how the metrics and baseline pings relate and the timings they record.

Ping sections

There are two standard metadata sections that are added to most pings, in addition to their core metrics and events content (which are described in Adding new metrics).

  • The ping_info section contains core metadata that is included in every ping.

  • The client_info section contains information that identifies the client. It is included in most pings (including all built-in pings), but may be excluded from pings where we don't want to connect client information with the other metrics in the ping.

The ping_info section

The following fields are included in the ping_info section, for every ping. Optional fields are marked accordingly.

Field nameTypeDescription
seqCounterA running counter of the number of times pings of this type have been sent
experimentsObjectOptional. A dictionary of active experiments
start_timeDatetimeThe time of the start of collection of the data in the ping, in local time and with minute precision, including timezone information.
end_timeDatetimeThe time of the end of collection of the data in the ping, in local time and with minute precision, including timezone information. This is also the time this ping was generated and is likely well before ping transmission time.
reasonStringOptional. The reason the ping was submitted. The specific set of values and their meanings are defined for each metric type in the reasons field in the pings.yaml file.

All the metrics surviving application restarts (e.g. seq, ...) are removed once the application using the Glean SDK is uninstalled.

The client_info section

The following fields are included in the client_info section. Optional fields are marked accordingly.

Field nameTypeDescription
app_buildStringThe build identifier generated by the CI system (e.g. "1234/A"). For language bindings that provide automatic detection for this value, (e.g. Android/Kotlin), in the unlikely event that the build identifier can not be retrieved from the OS, it is set to inaccessible. For other language bindings, if the value was not provided through configuration, this metric gets set to Unknown.
app_channelStringOptional The product-provided release channel (e.g. "beta")
app_display_versionStringThe user-visible version string (e.g. "1.0.3"). In the unlikely event this value can not be obtained from the OS, it is set to "inaccessible". If it is accessible, but not set by the application, it is set to "Unknown".
architectureStringThe architecture of the device (e.g. "arm", "x86")
client_idUUIDOptional A UUID identifying a profile and allowing user-oriented correlation of data
device_manufacturerStringOptional The manufacturer of the device
device_modelStringOptional The model name of the device. On Android, this is Build.MODEL, the user-visible name of the device.
first_run_dateDatetimeThe date of the first run of the application, in local time and with day precision, including timezone information.
osStringThe name of the operating system (e.g. "linux", "Android", "ios")
os_versionStringThe user-visible version of the operating system (e.g. "1.2.3")
android_sdk_versionStringOptional. The Android specific SDK version of the software running on this hardware device (e.g. "23")
telemetry_sdk_buildStringThe version of the Glean SDK
localeStringOptional. The locale of the application during initialization (e.g. "es-ES"). If the locale can't be determined on the system, the value is "und", to indicate "undetermined".

All the metrics surviving application restarts (e.g. client_id, ...) are removed once the application using the Glean SDK is uninstalled.

The experiments object

This object (included in the ping_info section) contains experiments keyed by the experiment id. Each listed experiment contains the branch the client is enrolled in and may contain a string to string map with additional data in the extra key. Both the id and branch are truncated to 30 characters. See Using the Experiments API on how to record experiments data.

{
  "<id>": {
    "branch": "branch-id",
    "extra": {
      "some-key": "a-value"
    }
  }
}

Ping submission

The pings that the Glean SDK generates are submitted to the Mozilla servers at specific paths, in order to provide additional metadata without the need to unpack the ping payload.

Note: To keep resource usage in check, the Glean SDK allows only up to 10 ping submissions every 60 seconds. There are no exposed methods to change these rate limiting defaults yet, follow Bug 1647630 for updates.

A typical submission URL looks like

"<server-address>/submit/<application-id>/<doc-type>/<glean-schema-version>/<document-id>"

where:

  • <server-address>: the address of the server that receives the pings;
  • <application-id>: a unique application id, automatically detected by the Glean SDK; this is the value returned by Context.getPackageName();
  • <doc-type>: the name of the ping; this can be one of the pings available out of the box with the Glean SDK, or a custom ping;
  • <glean-schema-version>: the version of the Glean ping schema;
  • <document-id>: a unique identifier for this ping.

Submitted headers

A pre-defined set of headers is additionally sent along with the submitted ping:

HeaderValueDescription
Content-Typeapplication/json; charset=utf-8Describes the data sent to the server
User-AgentDefaults to e.g. Glean/0.40.0 (Kotlin on Android), where 0.40.0 is the Glean SDK version number and Kotlin on Android is the name of the language used by the binding that sent the request plus the name of the platform it is running on.Describes the application sending the ping using the Glean SDK
Datee.g. Mon, 23 Jan 2019 10:10:10 GMT+00:00Submission date/time in GMT/UTC+0 offset
X-Client-TypeGleanCustom header to support handling of Glean pings in the legacy pipeline
X-Client-Versione.g. 0.40.0The Glean SDK version, sent as a custom header to support handling of Glean pings in the legacy pipeline

Defining foreground and background state

These docs refer to application 'foreground' and 'background' state in several places.

Foreground

For Android, this specifically means the activity becomes visible to the user, it has entered the Started state, and the system invokes the onStart() callback.

Background

This specifically means when the activity is no longer visible to the user, it has entered the Stopped state, and the system invokes the onStop() callback.

This may occur, if the user uses Overview button to change to another app, the user presses the Back button and navigates to a previous application or the home screen, or if the user presses the Home button to return to the home screen. This can also occur if the user navigates away from the application through some notification or other means.

The system may also call onStop() when the activity has finished running, and is about to be terminated.

Foreground

For iOS, Glean attaches to the willEnterForegroundNotification. This notification is posted by the OS shortly before an app leaves the background state on its way to becoming the active app.

Background

For iOS, this specifically means when the app is no longer visible to the user, or when the UIApplicationDelegate receives the applicationDidEnterBackground event.

This may occur if the user opens the task switcher to change to another app, or if the user presses the Home button to show the home screen. This can also occur if the user navigates away from the app through a notification or other means.

Note: Glean does not currently support Scene based lifecycle events that were introduced in iOS 13.

Ping schedules and timings overview

Full reference details about the metrics and baseline ping schedules are detailed elsewhere.

The following diagram shows a typical timeline of a mobile application, when pings are sent and what timing-related information is included.

ping timeline diagram

There are two distinct runs of the application, where the OS shutdown the application at the end of Run 1, and the user started it up again at the beginning of Run 2.

There are three distinct foreground sessions, where the application was visible on the screen and the user was able to interact with it.

The rectangles for the baseline and metrics pings represent the measurement windows of those pings, which always start exactly at the end of the preceding ping. The ping_info.start_time and ping_info.end_time metrics included in these pings correspond to these beginning and the end of their measurement windows.

The baseline.duration metric (included only in baseline pings) corresponds to amount of time the application spent on the foreground, which, since measurement window always extend to the next ping, is not always the same thing as the baseline ping's measurement window.

The submission_timestamp is the time the ping was received at the telemetry endpoint, added by the ingestion pipeline. It is not exactly the same as ping_info.end_time, since there may be various networking and system latencies both on the client and in the ingestion pipeline (represented by the dotted horizontal line, not to scale). Also of note is that start_time/end_time are measured using the client's real-time clock in its local timezone, which is not a fully reliable source of time.

The "Baseline 4" ping illustrates an important corner case. When "Session 2" ended, the OS also shut down the entire process, and the Glean SDK did not have an opportunity to send a baseline ping immediately. In this case, it is sent at the next available opportunity when the application starts up again in "Run 2". This baseline ping is annotated with the reason code dirty_startup.

The "Metrics 2" ping likewise illustrates another important corner case. "Metrics 1" was able to be sent at the target time of 04:00 (local device time) because the application was currently running. However, the next time 04:00 came around, the application was not active, so the Glean SDK was unable to send a metrics ping. It is sent at the next available opportunity, when the application starts up again in "Run 2". This metrics ping is annotated with the reason code overdue.

The baseline ping

Description

This ping is intended to provide metrics that are managed by the library itself, and not explicitly set by the application or included in the application's metrics.yaml file.

Note: As the baseline ping was specifically designed for mobile operating systems, it is not sent when using the Glean Python bindings.

Scheduling

The baseline ping is automatically submitted with a reason: foreground when the application is moved to the foreground. These baseline pings do not contain duration.

The baseline ping is automatically submitted with a reason: background when the application is moved to the background. Occasionally, the baseline ping may fail to send when going to background (e.g. the process is killed quickly). In that case, it will be submitted at startup with a reason: dirty_startup, if the previous session was not cleanly closed. This only happens from the second start onward.

See also the ping schedules and timing overview.

Contents

The baseline ping includes the following fields:

Field nameTypeDescription
durationTimespanThe duration, in seconds, of the last foreground session. Only available if reason: background. 1
1

See also the ping schedules and timing overview for how the duration metric relates to other sources of timing in the baseline ping.

The baseline ping also includes the common ping sections found in all pings.

Querying ping contents

A quick note about querying ping contents (i.e. for sql.telemetry.mozilla.org): Each metric in the baseline ping is organized by its metric type, and uses a namespace of glean.baseline. For instance, in order to select duration you would use metrics.timespan['glean.baseline.duration']. If you were trying to select a String based metric such as os, then you would use metrics.string['glean.baseline.os']

Example baseline ping

{
  "ping_info": {
    "experiments": {
      "third_party_library": {
        "branch": "enabled"
      }
    },
    "seq": 0,
    "start_time": "2019-03-29T09:50-04:00",
    "end_time": "2019-03-29T09:53-04:00",
    "reason": "foreground"
  },
  "client_info": {
    "telemetry_sdk_build": "0.49.0",
    "first_run_date": "2019-03-29-04:00",
    "os": "Android",
    "android_sdk_version": "27",
    "os_version": "8.1.0",
    "device_manufacturer": "Google",
    "device_model": "Android SDK built for x86",
    "architecture": "x86",
    "app_build": "1",
    "app_display_version": "1.0",
    "client_id": "35dab852-74db-43f4-8aa0-88884211e545"
  },
  "metrics": {
    "timespan": {
      "glean.baseline.duration": {
        "value": 52,
        "time_unit": "second"
      }
    }
  }
}

The deletion-request ping

Description

This ping is submitted when a user opts out of sending technical and interaction data.

This ping contains the client id.

This ping is intended to communicate to the Data Pipeline that the user wishes to have their reported Telemetry data deleted. As such it attempts to send itself at the moment the user opts out of data collection, and continues to try and send itself.

Note: It is possible to send secondary ids in the deletion request ping. For instance, if the application is migrating from legacy telemetry to Glean, the legacy client ids can be added to the deletion request ping by creating a metrics.yaml entry for the id to be added with a send_in_pings value of deletion_request.

An example metrics.yaml entry might look like this:

legacy_client_id:
   type: uuid
   description:
     A UUID uniquely identifying the legacy client.
   send_in_pings:
     - deletion_request
   ...

Scheduling

The deletion-request ping is automatically submitted when upload is disabled in Glean. If upload fails, it is retried after Glean is initialized.

Contents

The deletion-request does not contain additional metrics aside from secondary ids that have been added.

Example deletion-request ping

{
  "ping_info": {
    "seq": 0,
    "start_time": "2019-12-06T09:50-04:00",
    "end_time": "2019-12-06T09:53-04:00"
  },
  "client_info": {
    "telemetry_sdk_build": "22.0.0",
    "first_run_date": "2019-03-29-04:00",
    "os": "Android",
    "android_sdk_version": "28",
    "os_version": "9",
    "device_manufacturer": "Google",
    "device_model": "Android SDK built for x86",
    "architecture": "x86",
    "app_build": "1",
    "app_display_version": "1.0",
    "client_id": "35dab852-74db-43f4-8aa0-88884211e545"
  },
  "metrics": {
    "uuid": {
      "legacy_client_id": "5faffa6d-6147-4d22-a93e-c1dbd6e06171"
    }
  }
}

The metrics ping

Description

The metrics ping is intended for all of the metrics that are explicitly set by the application or are included in the application's metrics.yaml file (except events). The reported data is tied to the ping's measurement window, which is the time between the collection of two metrics pings. Ideally, this window is expected to be about 24 hours, given that the collection is scheduled daily at 04:00. However, the metrics ping is only submitted while the application is actually running, so in practice, it may not meet the 04:00 target very frequently. Data in the ping_info section of the ping can be used to infer the length of this window and the reason that triggered the ping to be submitted. If the application crashes, unsent recorded metrics are sent along with the next metrics ping.

Additionally, it is undesirable to mix metric recording from different versions of the application. Therefore, if a version upgrade is detected, the metrics ping is collected immediately before further metrics from the new version are recorded.

Note: As the metrics ping was specifically designed for mobile operating systems, it is not sent when using the Glean Python bindings.

Scheduling

The desired behavior is to collect the ping at the first available opportunity after 04:00 local time on a new calendar day, but given constraints of the platform, it can only be submitted while the application is running. This breaks down into three scenarios:

  1. the application was just installed;
  2. the application was just upgraded (the version of the app is different from the last time the app was run);
  3. the application was just started (after a crash or a long inactivity period);
  4. the application was running at 04:00.

In the first case, since the application was just installed, if the due time for the current calendar day has passed, a metrics ping is immediately generated and scheduled for sending (reason code overdue). Otherwise, if the due time for the current calendar day has not passed, a ping collection is scheduled for that time (reason code today).

In the second case, if a version change is detected at startup, the metrics ping is immediately submitted so that metrics from one version are not aggregated with metrics from another version (reason code upgrade).

In the third case, if the metrics ping was not already collected on the current calendar day, and it is before 04:00, a collection is scheduled for 04:00 on the current calendar day (reason code today). If it is after 04:00, a new collection is scheduled immediately (reason code overdue). Lastly, if a ping was already collected on the current calendar day, the next one is scheduled for collecting at 04:00 on the next calendar day (reason code tomorrow).

In the fourth and last case, the application is running during a scheduled ping collection time. The next ping is scheduled for 04:00 the next calendar day (reason code reschedule).

More scheduling examples are included below.

See also the ping schedules and timing overview.

Contents

The metrics ping contains all of the metrics defined in metrics.yaml (except events) that don't specify a ping or where default is specified in their send in pings property.

Additionally, error metrics in the glean.error category are included in the metrics ping.

The metrics ping shall also include the common ping_info and 'client_info' sections.

Querying ping contents

Information about query ping contents is available in Accessing Glean data in the Firefox data docs.

Scheduling Examples

Crossing due time with the application closed

  1. The application is opened on Feb 7 on 15:00, closed on 15:05.

    • Glean records one metric A (say startup time in ms) during this measurement window MW1.
  2. The application is opened again on Feb 8 on 17:00.

  • Glean notes that we passed local 04:00 since MW1.

  • Glean closes MW1, with:

    • start_time=Feb7/15:00;
    • end_time=Feb8/17:00.
  • Glean records metric A again, into MW2, which has a start_time of Feb8/17:00.

Crossing due time and changing timezones

  1. The application is opened on Feb 7 on 15:00 in timezone UTC, closed on 15:05.

    • Glean records one metric A (say startup time in ms) during this measurement window MW1.
  2. The application is opened again on Feb 8 on 17:00 in timezone UTC+1.

    • Glean notes that we passed local 04:00 UTC+1 since MW1.

    • Glean closes MW1, with:

      • start_time=Feb7/15:00/UTC;
      • end_time=Feb8/17:00/UTC+1.
    • Glean records metric A again, into MW2.

The application doesn’t run in a week

  1. The application is opened on Feb 7 on 15:00 in timezone UTC, closed on 15:05.

    • Glean records one metric A (say startup time in ms) during this measurement window MW1.
  2. The application is opened again on Feb 16 on 17:00 in timezone UTC.

    • Glean notes that we passed local 04:00 UTC since MW1.

    • Glean closes MW1, with:

      • start_time=Feb7/15:00/UTC;
      • end_time=Feb16/17:00/UTC.
    • Glean records metric A again, into MW2.

The application doesn’t run for a week, and when it’s finally re-opened the timezone has changed

  1. The application is opened on Feb 7 on 15:00 in timezone UTC, closed on 15:05.

    • Glean records one metric A (say startup time in ms) during this measurement window MW1.
  2. The application is opened again on Feb 16 on 17:00 in timezone UTC+1.

    • Glean notes that we passed local 04:00 UTC+1 since MW1.

    • Glean closes MW1, with:

      • start_time=Feb7/15:00/UTC
      • end_time=Feb16/17:00/UTC+1.
    • Glean records metric A again, into MW2.

The user changes timezone in an extreme enough fashion that they cross 04:00 twice on the same date

  1. The application is opened on Feb 7 at 15:00 in timezone UTC+11, closed at 15:05.

    • Glean records one metric A (say startup time in ms) during this measurement window MW1.
  2. The application is opened again on Feb 8 at 04:30 in timezone UTC+11.

    • Glean notes that we passed local 04:00 UTC+11.

    • Glean closes MW1, with:

      • start_time=Feb7/15:00/UTC+11;
      • end_time=Feb8/04:30/UTC+11.
    • Glean records metric A again, into MW2.

  3. The user changes to timezone UTC-10 and opens the application at Feb 7 at 22:00 in timezone UTC-10

    • Glean records metric A again, into MW2 (not MW1, which was already sent).
  4. The user opens the application at Feb 8 05:00 in timezone UTC-10

    • Glean notes that we have not yet passed local 04:00 on Feb 9
    • Measurement window MW2 remains the current measurement window
  5. The user opens the application at Feb 9 07:00 in timezone UTC-10

    • Glean notes that we have passed local 04:00 on Feb 9

    • Glean closes MW2 with:

      • start_time=Feb8/04:30/UTC+11;
      • end_time=Feb9/19:00/UTC-10.
    • Glean records metric A again, into MW3.

The events ping

Description

The events ping's purpose is to transport all of the event metric information. If the application crashes, an events ping is generated next time the application starts with events that were not sent before the crash.

Scheduling

The events ping is collected under the following circumstances:

  1. Normally, it is collected when the application goes into the background, if there are any recorded events to send.

  2. When the queue of events exceeds Glean.configuration.maxEvents (default 500).

  3. If there are any unsent events found on disk when starting the application. It would be impossible to coordinate the timestamps across a reboot, so it's best to just collect all events from the previous run into their own ping, and start over.

All of these cases are handled automatically, with no intervention or configuration required by the application.

Note: Since the Python bindings don't have a concept of "going to background", case (1) above does not apply.

Contents

At the top-level, this ping contains the following keys:

  • ping_info: The information common to all pings.

  • events: An array of all of the events that have occurred since the last time the events ping was sent.

Each entry in the events array is an object with the following properties:

  • "timestamp": The milliseconds relative to the first event in the ping.

  • "category": The category of the event, as defined by its location in the metrics.yaml file.

  • "name": The name of the event, as defined in the metrics.yaml file.

  • "extra" (optional): A mapping of strings to strings providing additional data about the event. The keys are restricted to 40 characters and values in this map will never exceed 100 characters.

Example event JSON

{
  "ping_info": {
    "experiments": {
      "third_party_library": {
        "branch": "enabled"
      }
    },
    "seq": 0,
    "start_time": "2019-03-29T09:50-04:00",
    "end_time": "2019-03-29T10:02-04:00"
  },
  "client_info": {
    "telemetry_sdk_build": "0.49.0",
    "first_run_date": "2019-03-29-04:00",
    "os": "Android",
    "android_sdk_version": "27",
    "os_version": "8.1.0",
    "device_manufacturer": "Google",
    "device_model": "Android SDK built for x86",
    "architecture": "x86",
    "app_build": "1",
    "app_display_version": "1.0",
    "client_id": "35dab852-74db-43f4-8aa0-88884211e545"
  },
  "events": [
    {
      "timestamp": 123456789,
      "category": "examples",
      "name": "event_example",
      "extra": {
        "metadata1": "extra",
        "metadata2": "more_extra"
      }
    },
    {
      "timestamp": 123456791,
      "category": "examples",
      "name": "event_example"
    }
  ]
}

Custom pings

Applications can define metrics that are sent in custom pings. Unlike the built-in pings, custom pings are sent explicitly by the application.

This is useful when the scheduling of the built-in pings (metrics, baseline and events) are not appropriate for your data. Since the timing of the submission of custom pings is handled by the application, the measurement window is under the application's control.

This is especially useful when metrics need to be tightly related to one another, for example when you need to measure the distribution of frame paint times when a particular rendering backend is in use. If these metrics were in different pings, with different measurement windows, it is much harder to do that kind of reasoning with much certainty.

Defining a custom ping

Custom pings must be defined in a pings.yaml file, which is in the same directory alongside your app's metrics.yaml file.

Ping names are limited to lowercase letters from the ISO basic Latin alphabet and hyphens and a maximum of 30 characters.

Each ping has the following parameters:

  • description (required): A textual description describing the purpose of the ping. It may contain markdown syntax.
  • include_client_id (required): A boolean indicating whether to include the client_id in the client_info section).
  • send_if_empty (optional, default: false): A boolean indicating if the ping is sent if it contains no metric data.
  • reasons (optional, default: {}): The reasons that this ping may be sent. The keys are the reason codes, and the values are a textual description of each reason. The ping payload will (optionally) contain one of these reasons in the ping_info.reason field.

In addition to these parameters, pings also support the parameters related to data review and expiration defined in common metric parameters: description, notification_emails, bugs, and data_reviews.

For example, to define a custom ping called search specifically for search information:

# Required to indicate this is a `pings.yaml` file
$schema: moz://mozilla.org/schemas/glean/pings/1-0-0

search:
  description: >
    A ping to record search data.
  include_client_id: false
  notification_emails:
    - CHANGE-ME@example.com
  bugs:
    - http://bugzilla.mozilla.org/123456789/
  data_reviews:
    - http://example.com/path/to/data-review

Note: the names baseline, metrics, events, deletion-request and all-pings are reserved and may not be used as the name of a custom ping.

Loading custom ping metadata into your application or library

The Glean SDK build generates code from pings.yaml in a Pings object, which must be instantiated so Glean can send pings by name.

In Kotlin, this object must be registered with Glean from your startup code (such as in your application's onCreate method or a function called from that method).

import org.mozilla.yourApplication.GleanMetrics.Pings

...

override fun onCreate() {
    ...
    Glean.registerPings(Pings)
    ...
}

In Swift, this object must be registered with Glean from your startup code (such as in your application's UIApplicationDelegate application(_:didFinishLaunchingWithOptions:) method or a function called from that method).

import Glean

@UIApplicationMain
class AppDelegate: UIResponder, UIApplicationDelegate {
func application(_: UIApplication, didFinishLaunchingWithOptions _: [UIApplication.LaunchOptionsKey: Any]?) -> Bool {
    // ...
    Glean.shared.registerPings(GleanMetrics.Pings)
    // ...
}
}

For Python, the pings.yaml file must be available and loaded at runtime.

If your project is a script (i.e. just Python files in a directory), you can load the pings.yaml using:

from glean import load_pings

pings = load_pings("pings.yaml")

If your project is a distributable Python package, you need to include the metrics.yaml file using one of the myriad ways to include data in a Python package and then use pkg_resources.resource_filename() to get the filename at runtime.

from glean import load_pings
from pkg_resources import resource_filename

pings = load_pings(resource_filename(__name__, "pings.yaml"))

In C#, this object must be registered with Glean from your startup code (such as in your application's Main method or a function called from that method).

using static Mozilla.YourApplication.GleanMetrics.Pings;

...

class Program
{
    static void Main(string[] args)
    {
        ...
        Glean.RegisterPings(Pings);
        ...
    }
}

Sending metrics in a custom ping

To send a metric on a custom ping, you add the custom ping's name to the send_in_pings parameter in the metrics.yaml file.

For example, to define a new metric to record the default search engine, which is sent in a custom ping called search, put search in the send_in_pings parameter. Note that it is an error to specify a ping in send_in_pings that does not also have an entry in pings.yaml.

search.default:
  name:
    type: string
    description: >
      The name of the default search engine.
    send_in_pings:
      - search

If this metric should also be sent in the default ping for the given metric type, you can add the special value default to send_in_pings:

    send_in_pings:
      - search
      - default

Submitting a custom ping

To collect and queue a custom ping for eventual uploading, call the submit method on the PingType object that the Glean SDK generated for your ping.

By default, if the ping doesn't currently have any events or metrics set, submit will do nothing. However, if the send_if_empty flag is set to true in the ping definition, it will always be submitted.

For example, to submit the custom ping defined above:

import org.mozilla.yourApplication.GleanMetrics.Pings
Pings.search.submit(
    GleanMetrics.Pings.searchReasonCodes.performed
)
import Glean

GleanMetrics.Pings.shared.search.submit(
    reason: .performed
)
from glean import load_pings

pings = load_pings("pings.yaml")

pings.search.submit(pings.search_reason_codes.PERFORMED)
using static Mozilla.YourApplication.GleanMetrics.Pings;

Pings.search.Submit(
    GleanMetrics.Pings.searchReasonCodes.performed
);

If none of the metrics for the ping contain data the ping is not sent (unless send_if_empty is set to true in the definition file)

Unit testing Glean custom pings for Android

Applications defining custom pings can use use the strategy defined in this document to test these pings in unit tests.

General testing strategy

The schedule of custom pings depends on the specific application implementation, since it is up to the SDK user to define the ping semantics. This makes writing unit tests for custom pings a bit more involved.

One possible strategy could be to wrap the Glean SDK API call to send the ping in a function that can be mocked in the unit test. This would allow for checking the status and the values of the metrics contained in the ping at the time in which the application would have sent it.

Example testing of a custom ping

Let us start by defining a custom ping with a sample metric in it. Here is the pings.yaml file:

$schema: moz://mozilla.org/schemas/glean/pings/1-0-0

my-custom-ping:
  description: >
    This ping is intended to showcase the recommended testing strategy for
    custom pings.
  include_client_id: false
  bugs:
    - https://bugzilla.mozilla.org/1556985/
  data_reviews:
    - https://bugzilla.mozilla.org/show_bug.cgi?id=1556985
  notification_emails:
    - custom-ping-owner@example.com

And here is the metrics.yaml

$schema: moz://mozilla.org/schemas/glean/metrics/1-0-0

custom_ping_data:
  sample_string:
    type: string
    lifetime: ping
    description: >
      A sample string metric for demonstrating unit tests for custom pings.
    send_in_pings:
      - my-custom-ping
    bugs:
      - https://bugzilla.mozilla.org/1556985/
    data_reviews:
      - https://bugzilla.mozilla.org/show_bug.cgi?id=1556985
    notification_emails:
      - custom-ping-owner@example.com
    expires: "2019-10-01"

A potential usage of the Glean SDK generated API could be the following:

import my.component.GleanMetrics.Pings
import my.component.GleanMetrics.CustomPingData

class MyCustomPingScheduler {
  /**
   * HERE ONLY TO KEEP THE EXAMPLE SIMPLE.
   *
   * A function that consumes the Glean SDK generated metrics API to
   * record some data. It doesn't really need to be in a function, nor
   * in this class. The Glean SDK API can be called when the data is
   * generated.
   */
  fun addSomeData() {
    // Record some sample data.
    CustomPingData.sampleString.set("test-data")
  }

  /**
   * Called to implement the ping scheduling logic for 'my_custom_ping'.
   */
  fun schedulePing() {
    // ... some scheduling logic that will end up calling the function below.
    submitPing()
  }

  /**
   * Internal function to only be overridden in tests. This
   * calls the Glean SDK API to send custom pings.
   */
  @VisibleForTesting(otherwise = VisibleForTesting.NONE)
  internal fun submitPing() {
    Pings.MyCustomPing.submit()
  }
}

The following unit test intercepts the MyCustomPingScheduler.submitPing() call in order to perform the validation on the data. This specific example uses Mockito, but any other framework would work.

// Metrics and pings definitions.
import my.component.GleanMetrics.Pings
import my.component.GleanMetrics.CustomPingData

// Mockito imports for using spies.
import org.mockito.Mockito.spy
import org.mockito.Mockito.`when`

@RunWith(AndroidJUnit4::class)
class MyCustomPingSchedulerTest {
    // Apply the GleanTestRule to set up a disposable Glean instance.
    // Please note that this clears the Glean data across tests.
    @get:Rule
    val gleanRule = GleanTestRule(ApplicationProvider.getApplicationContext())

    @Test
    fun `verify custom ping metrics`() {
      val scheduler = spy(MyCustomPingScheduler())
      doAnswer {
        // Here we validate the content that goes into the ping.
        assertTrue(CustomPingData.sampleString.testHasValue())
        assertEquals("test-data", CustomPingData.sampleString.testGetValue())

        // We want to intercept this call, but we also want to make sure the
        // real Glean API is called in order to clear the ping store and to provide
        // consistent behaviour with respect to the application.
        it.callRealMethod()
      }.`when`(scheduler).submitPing()

      scheduler.addSomeData()
      scheduler.schedulePing()
    }
}
import Foundation
import Glean

// Use typealiases to simplify usage.
// This can be placed anywhere in your code to be available in all files.
typealias CustomPingData = GleanMetrics.CustomPingData
typealias Pings = GleanMetrics.Pings

class MyCustomPingScheduler {
    /**
     * HERE ONLY TO KEEP THE EXAMPLE SIMPLE.
     *
     * A function that consumes the Glean SDK generated metrics API to
     * record some data. It doesn't really need to be in a function, nor
     * in this class. The Glean SDK API can be called when the data is
     * generated.
     */
    func addSomeData() {
       // Record some sample data.
       CustomPingData.sampleString.set("test-data")
    }

    /**
     * Called to implement the ping scheduling logic for 'my_custom_ping'.
     */
    func schedulePing() {
        // ... some scheduling logic that will end up calling the function below.
        submitPing()
    }

    /**
     * Internal function to only be overridden in tests. This
     * calls the Glean SDK API to send custom pings.
     */
    internal func submitPing() {
        Pings.shared.myCustomPing.submit()
    }
}

The following unit test intercepts the MyCustomPingScheduler.submitPing() call in order to perform the validation on the data. This example uses a manual mock implementation, but you could use a framework for that.

@testable import YourApplication
import Glean
import XCTest

class MyCustomPingSchedulerMock: MyCustomPingScheduler {
    var submitWasCalled = false

    deinit {
        XCTAssertTrue(submitWasCalled, "submitPing should have been called once")
    }

    override func submitPing() {
        submitWasCalled = true

        XCTAssertTrue(CustomPingData.os.testHasValue())
        XCTAssertEqual("test-data", try! CustomPingData.os.testGetValue())

        super.submitPing()
    }
}

class MyCustomPingSchedulerTests: XCTestCase {
    override func setUp() {
        Glean.shared.resetGlean(clearStores: true)
    }

    func testCustomPingMetrics() {
        let scheduler = MyCustomPingSchedulerMock()
        scheduler.addSomeData()
        scheduler.schedulePing()
    }
}
import glean

metrics = glean.load_metrics("metrics.yaml")
pings = glean.load_pings("pings.yaml")


class MyCustomPingScheduler:
    def add_some_data(self):
        """
        HERE ONLY TO KEEP THE EXAMPLE SIMPLE.

        A function that consumes the Glean SDK generated metrics API to
        record some data. It doesn't really need to be in a function, nor
        in this class. The Glean SDK API can be called when the data is
        generated.
        """
        # Record some sample data.
        metrics.custom_ping_data.sample_string.set("test-data")

    def schedule_ping(self):
        """
        Called to implement the ping scheduling logic for 'my_custom_ping'.
        """
        # ... some scheduling logic that will end up calling the function below.
        self._submit_ping()

    def _submit_ping(self):
        """
        Internal function to only be overridden in tests.
        """
        pings.my_custom_ping.submit()

The following unit test intercepts the MyCustomPingScheduler._submit_ping() call in order to perform the validation on the data.

from unittest.mock import MagicMock

from glean import testing

import custom_ping_scheduler


# This will be run before every test in the entire test suite
def pytest_runtest_setup(item):
    testing.reset_glean(application_id="my-app", application_version="0.0.1")


def test_verify_custom_ping_metrics():
    scheduler = custom_ping_scheduler.MyCustomPingScheduler()

    original_submit_ping = scheduler._submit_ping

    def _submit_ping(self):
        # Here we validate the content that goes into the ping.
        assert (
            custom_ping_scheduler.metrics.custom_ping_data.sample_string.test_has_value()
        )
        assert (
            "test-data"
            == custom_ping_scheduler.metrics.custom_ping_data.sample_string.test_get_value()
        )

        # We want to intercept this call, but we also want to make sure the
        # real Glean API is called in order to clear the ping strong and to
        # provide consistent behavior with respect to the application.
        original_submit_ping(self)

    scheduler._submit_ping = MagicMock(_submit_ping)

    scheduler.add_some_data()
    scheduler.schedule_ping()

TODO. To be implemented in bug 1648446.

Android build script configuration options

This chapter describes build configuration options that control the behavior of Glean's Gradle plugin. These options are not usually required for normal use.

Options can be turned on by setting a variable on the Gradle ext object before applying the Glean Gradle plugin.

allowMetricsFromAAR

Normally, Glean looks for metrics.yaml and pings.yaml files in the root directory of the Glean-using project. However, in some cases, these files may need to ship inside the dependencies of the project. For example, this is used in the engine-gecko component to grab the metrics.yaml from the geckoview AAR.

ext.allowMetricsFromAAR = true

When this flag is set, every direct dependency of your library will be searched for a metrics.yaml file, and those metrics will be treated as the metrics as if they were defined by your library. That is, API wrappers accessible from your library will be generated for those metrics.

The metrics.yaml can be added to the dependency itself by calling this on each relevant build variant:

variant.packageLibraryProvider.get().from("${topsrcdir}/path/metrics.yaml")

gleanGenerateMarkdownDocs

The Glean SDK can automatically generate Markdown documentation for metrics and pings defined in the registry files, in addition to the metrics API code.

ext.gleanGenerateMarkdownDocs = true

Flipping the feature to true will generate a metrics.md file in $projectDir/docs at build-time.

gleanDocsDirectory

The gleanDocsDirectory can be used to customize the path of the documentation output directory. If gleanGenerateMarkdownDocs is disabled, it does nothing. Please note that only the metrics.md will be overwritten: any other file available in the target directory will be preserved.

ext.gleanDocsDirectory = "$rootDir/docs/user/telemetry"

gleanYamlFiles

By default, the Glean Gradle plugin will look for metrics.yaml and pings.yaml files in the same directory that the plugin is included from in your application or library. To override this, ext.gleanYamlFiles may be set to a list of explicit paths.

ext.gleanYamlFiles = ["$rootDir/glean-core/metrics.yaml", "$rootDir/glean-core/pings.yaml"]

Focused Use Cases

Here is a list of specific examples of using Glean to instrument different situations.

Instrumenting Android Crashes With Glean

This is a simple example illustrating a strategy or method for instrumenting crashes in Android applications using Glean.

Instrumenting Android Crashes With The Glean SDK

One of the things that might be useful to collect data on in an Android application is crashes. This guide will walk through a basic strategy for instrumenting an Android application with crash telemetry using a custom ping.

Note: This is a very simple example of instrumenting crashes using the Glean SDK. There will be challenges to using this approach in a production application that should be considered. For instance, when an app crashes it can be in an unknown state and may not be able to do things like upload data to a server. The recommended way of instrumenting crashes with Android Components is called lib-crash, which takes into consideration things like multiple processes and persistence.

Before You Start

There are a few things that need to be installed in order to proceed, mainly Android Studio. If you include the Android SDK, Android Studio can take a little while to download and get installed. This walk-through assumes some knowledge of Android application development. Knowing where to go to create a new project and how to add dependencies to a Gradle file will be helpful in following this guide.

Setup Build Configuration

Please follow the instruction in the "Adding Glean to your project" chapter in order to set up Glean in an Android project.

Add A Custom Metric

Since crashes will be instrumented with some custom metrics, the next step will be to add a metrics.yaml file to define the metrics used to record the crash information and a pings.yaml file to define a custom ping which will give some control over the scheduling of the uploading. See "Adding new metrics" for more information about adding metrics.

What metric type should be used to represent the crash data? While this could be implemented several ways, an event is an excellent choice, simply because events capture information in a nice concise way and they have a built-in way of passing additional information using the extras field. If it is necessary to pass along the cause of the exception or a few lines of description, events let us do that easily (with some limitations).

Now that a metric type has been chosen to represent the metric, the next step is creating the metrics.yaml. Inside of the root application folder of the Android Studio project create a new file named metrics.yaml. After adding the schema definition and event metric definition, the metrics.yaml should look like this:

# Required to indicate this is a `metrics.yaml` file
$schema: moz://mozilla.org/schemas/glean/metrics/1-0-0

crash:
  exception:
    type: event
    description: |
      Event to record crashes caused by unhandled exceptions
    notification_emails:
      - crashes@example.com
    bugs:
      - https://bugzilla.mozilla.org/show_bug.cgi?id=1582479
    data_reviews:
      - https://bugzilla.mozilla.org/show_bug.cgi?id=1582479
    expires:
      2021-01-01
    send_in_pings:
      - crash
    extra_keys:
      cause:
        description: The cause of the crash
      message:
        description: The exception message

As a brief explanation, this creates a metric called exception within a metric category called crash. There is a text description and the required notification_emails, bugs, data_reviews, and expires fields. The send_in_pings field is important to note here that it has a value of - crash. This means that the crash event metric will be sent via a custom ping named crash (which hasn't been created yet). Finally, note the extra_keys field which has two keys defined, cause and message. This allows for sending additional information along with the event to be associated with these keys.

Note: For Mozilla applications, a mandatory data review is required in order to collect information with the Glean SDK.

Add A Custom Ping

Define the custom ping that will help control the upload scheduling by creating a pings.yaml file in the same directory as the metrics.yaml file. For more information about adding custom pings, see the section on custom pings.
The name of the ping will be crash, so the pings.yaml file should look like this:

# Required to indicate this is a `pings.yaml` file
$schema: moz://mozilla.org/schemas/glean/pings/1-0-0

crash:
  description: >
    A ping to transport crash data
  include_client_id: true
  notification_emails:
    - crash@example.com
  bugs:
    - https://bugzilla.mozilla.org/show_bug.cgi?id=1582479
  data_reviews:
    - https://bugzilla.mozilla.org/show_bug.cgi?id=1582479

Before the newly defined metric or ping can be used, the application must first be built. This will cause the glean_parser to execute and generate the API files that represent the metric and ping that were newly defined.

Note: If changes to the YAML files aren't showing up in the project, try running the clean task on the project before building any time one of the Glean YAML files has been modified.

It is recommended that Glean be initialized as early in the application startup as possible, which is why it's good to use a custom Application, like the Glean Sample App GleanApplication.kt.

Initializing Glean in the Application.onCreate() is ideal for this purpose. Start by adding the import statement to allow the usage of the custom ping that was created, adding the following to the top of the file:

import org.mozilla.gleancrashexample.GleanMetrics.Pings

Next, register the custom ping by calling Glean.registerPings(Pings) in the onCreate() function, preferably before calling Glean.initialize(). The completed function should look something like this:

override fun onCreate() {
  super.onCreate()

  // Register the application's custom pings.
  Glean.registerPings(Pings)

  // Initialize the Glean library
  Glean.initialize(applicationContext)
}

This completes the registration of the custom ping with the Glean SDK so that it knows about it and can manage the storage and other important details of it like sending it when send() is called.

Instrument The App To Record The Event

In order to make the custom Application class handle uncaught exceptions, extend the class definition by adding Thread.UncaughtExceptionHandler as an inherited class like this:

class MainActivity : AppCompatActivity(), Thread.UncaughtExceptionHandler {
    ...
}

As part of implementing the Thread.UncaughtExceptionHandler interface, the custom Application needs to implement the override of the uncaughtException() function. An example of this override that records data and sends the ping could look something like this:

override fun uncaughtException(thread: Thread, exception: Throwable) {
    Crash.exception.record(
        mapOf(
            Crash.exceptionKeys.cause to exception.cause!!.toString(),
            Crash.exceptionKeys.message to exception.message!!
        )
    )
    Pings.crash.submit()
}

This records data to the Crash.exception metric from the metrics.yaml. The category of the metric is crash and the name is exception so it is accessed it by calling record() on the Crash.exception object. The extra information for the cause and the message is set as well. Finally, calling Pings.crash.submit() forces the crash ping to be scheduled to be sent.

The final step is to register the custom Application as the default uncaught exception handler by adding the following to the onCreate() function after Glean.initialize(this):

Thread.setDefaultUncaughtExceptionHandler(this)

Next Steps

This information didn't really get recorded by anything, as it would be rejected by the telemetry pipeline unless the application was already known. In order to collect telemetry from a new application, there is additional work that is necessary that is beyond the scope of this example. In order for data to be collected from your project, metadata must be added to the probe_scraper. The instructions for accomplishing this can be found in the probe_scraper documentation.

Metrics

This document enumerates the metrics collected by this project. This project may depend on other projects which also collect metrics. This means you might have to go searching through the dependency tree to get a full picture of everything collected by this project.

Pings

all-pings

These metrics are sent in every ping.

The following metrics are added to the ping:

NameTypeDescriptionData reviewsExtrasExpiration
glean.error.invalid_labellabeled_counterCounts the number of times a metric was set with an invalid label. The labels are the category.name identifier of the metric.1never
glean.error.invalid_overflowlabeled_counterCounts the number of times a metric was set a value that overflew. The labels are the category.name identifier of the metric.1never
glean.error.invalid_statelabeled_counterCounts the number of times a timing metric was used incorrectly. The labels are the category.name identifier of the metric.1never
glean.error.invalid_valuelabeled_counterCounts the number of times a metric was set to an invalid value. The labels are the category.name identifier of the metric.1never

baseline

This is a built-in ping that is assembled out of the box by the Glean SDK.

See the Glean SDK documentation for the baseline ping.

This ping is sent if empty.

This ping includes the client id.

Data reviews for this ping:

Bugs related to this ping:

Reasons this ping may be sent:

  • background: The ping was submitted before going to background.

  • dirty_startup: The ping was submitted at startup, because the application process was killed before the Glean SDK had the chance to generate this ping, when going to background, in the last session.

    *Note*: this ping will not contain the `glean.baseline.duration` metric.
    
  • foreground: The ping was submitted when the application went to foreground, which includes when the application starts.

    *Note*: this ping will not contain the `glean.baseline.duration` metric.
    

The following metrics are added to the ping:

NameTypeDescriptionData reviewsExtrasExpiration
glean.baseline.durationtimespanThe duration of the last foreground session.1never

deletion-request

This ping is submitted when a user opts out of sending technical and interaction data to Mozilla. This ping is intended to communicate to the Data Pipeline that the user wishes to have their reported Telemetry data deleted. As such it attempts to send itself at the moment the user opts out of data collection.

This ping is sent if empty.

This ping includes the client id.

Data reviews for this ping:

Bugs related to this ping:

This ping contains no metrics.

metrics

This is a built-in ping that is assembled out of the box by the Glean SDK.

See the Glean SDK documentation for the metrics ping.

This ping includes the client id.

Data reviews for this ping:

Bugs related to this ping:

Reasons this ping may be sent:

  • overdue: The last ping wasn't submitted on the current calendar day, but it's after 4am, so this ping submitted immediately

  • reschedule: A ping was just submitted. This ping was rescheduled for the next calendar day at 4am.

  • today: The last ping wasn't submitted on the current calendar day, but it is still before 4am, so schedule to send this ping on the current calendar day at 4am.

  • tomorrow: The last ping was already submitted on the current calendar day, so schedule this ping for the next calendar day at 4am.

  • upgrade: This ping was submitted at startup because the application was just upgraded.

The following metrics are added to the ping:

NameTypeDescriptionData reviewsExtrasExpiration
glean.error.preinit_tasks_overflowcounterThe number of tasks queued in the pre-initialization buffer. Only sent if the buffer overflows.1never
glean.upload.ping_upload_failurelabeled_counterCounts the number of ping upload failures, by type of failure. This includes failures for all ping types, though the counts appear in the next successfully sent metrics ping.1
  • status_code_4xx
  • status_code_5xx
  • status_code_unknown
  • unrecoverable
  • recoverable
never

Developing the Glean SDK

In this chapter we describe how to develop the Glean SDK.

Running the tests

Running all tests

The tests for all languages may be run from the command line:

make test

Windows Note: On Windows, make is not available by default. While not required, installing make will allow you to use the convenience features in the Makefile.

Running the Rust tests

The Rust tests may be run with the following command:

cargo test --all

Log output can be controlled via the environment variable RUST_LOG for the glean_core crate:

export RUST_LOG=glean_core=debug

When running tests with logging you need to tell cargo to not suppress output:

cargo test -- --nocapture

Tests run in parallel by default, leading to interleaving log lines. This makes it harder to understand what's going on. For debugging you can force single-threaded tests:

cargo test -- --nocapture --test-threads=1

Running the Kotlin/Android tests

From the command line

The full Android test suite may be run from the command line with:

./gradlew test

From Android Studio

To run the full Android test suite, in the "Gradle" pane, navigate to glean-core -> Tasks -> verification and double-click either testDebugUnitTest or testReleaseUnitTest (depending on whether you want to run in Debug or Release mode). You can save this task permanently by opening the task dropdown in the toolbar and selecting "Save glean.rs:glean:android [testDebugUnitTest] Configuration".

To run a single Android test, navigate to the file containing the test, and right click on the green arrow in the left margin next to the test. There you have a choice of running or debugging the test.

Running the Swift/iOS tests

From the command line

The full iOS test suite may be run from the command line with:

make test-swift

From Xcode

To run the full iOS test suite, run tests in Xcode (Product -> Test). To run a single Swift test, navigate to the file containing the test, and click on the arrow in the left margin next to the test.

Testing in CI

We run multiple tests on CI for every Pull Request and every commit to the main branch. These include:

  • Full Android tests
  • Full iOS tests
  • Rust tests
  • Rust, Kotlin and Swift source code formatting
  • Rust, Kotlin and Swift source code linting
  • Generating documentation from Rust, Kotlin, Swift and the book
  • Checking link validity of documentation
  • Deploying generated documentation

These checks are required to pass before a Pull Request is merged.

Documentation-only changes

Documentation is deployed from CI, we therefore need it to run on documentation changes. However, some of the long-running code tests can be skipped. For that add the following literal string to the last commit message to be pushed:

[doc only]

Skipping CI completely

It is possible to completely skip running CI on a given push by including the following literal string in the commit message:

[ci skip]

This should only be used for metadata files, such as those in .github, LICENSE or CODE_OF_CONDUCT.md.

Developing documentation

The documentation in this repository pertains to the Glean SDK. That is, the client-side code for Glean telemetry. Documentation for Glean in general, and the Glean-specific parts of the data pipeline and analysis is documented elsewhere in the firefox-data-docs repository.

The main narrative documentation is written in Markdown and converted to static HTML using mdbook.

API docs are also generated from docstrings for Rust, Kotlin, Swift and Python.

Building documentation

Building the narrative (book) documentation

The mdbook crate is required in order to build the narrative documentation:

cargo install mdbook

Then both the narrative and Rust API docs can be built with:

make docs
# ...or...
bin/build-rust-docs.sh
# ...or on Windows...
bin/build-rust-docs.bat

The built narrative documentation is saved in build/docs/book, and the Rust API documentation is saved in build/docs/docs.

Building API documentation

Kotlin API documentation is generated using dokka. It is automatically installed by Gradle.

To build the Kotlin API documentation:

make kotlin-docs

The generated documentation is saved in build/docs/javadoc.

Swift API documentation is generated using jazzy. It can be installed using:

  1. Install the latest Ruby: brew install ruby
  2. Make the installed Ruby available: export PATH=/usr/local/opt/ruby/bin:$PATH (and add that line to your .bashrc)
  3. Install the documentation tool: gem install jazzy

To build the Swift API documentation:

make swift-docs

The generated documentation is saved in build/docs/swift.

The Python API docs are generated using pdoc3. It is installed as part of creating the virtual environment for Python development.

To build the Python API documentation:

make python-docs

The generated documentation is saved in build/docs/python.

TODO. To be implemented in bug 1648410.

Checking links

Internal links within the documentation can be checked using the linkchecker tool. External links are currently not checked, since this takes considerable time and frequently fails in CI due to networking restrictions or issues.

Link checking requires building the narrative documentation as well as all of the API documentation for all languages. It is rare to build all of these locally (and in particular, the Swift API documentation can only be built on macOS), therefore it is reasonable to let CI catch broken link errors for you.

However, if you do want to run the linkchecker locally, it can be installed using pip or your system's package manager. Then, run linkchecker with:

make linkcheck

Spell checking

The narrative documentation (but not the API documentation) is spell checked using aspell.

On Unix-like platforms, it can be installed using your system's package manager:

sudo apt install aspell-en
# ...or...
sudo dnf install aspell-en
# ...or...
brew install aspell

Note that aspell 0.60.8 or later is required, as that is the first release with markdown support.

You can the spell check the narrative documentation using the following:

make spellcheck
# ...or...
bin/spellcheck.sh

This will bring up an interactive spell-checking environment in the console. Pressing a to add a word will add it to the project's local .dictionary file, and the changes should be committed to git.

Android bindings

The Glean SDK contains the Kotlin bindings for use by Android applications. It makes use of the underlying Rust component with a Kotlin-specific API on top. It also includes integrations into the Android platform.

Setup the Android Build Environment

Doing a local build of the Glean SDK:

This document describes how to make local builds of the Android bindings in this repository. Most consumers of these bindings do not need to follow this process, but will instead use pre-built bindings.

Prepare your build environment

Typically, this process only needs to be run once, although periodically you may need to repeat some steps (e.g., Rust updates should be done periodically).

Setting up Android dependencies

With Android Studio

The easiest way to install all the dependencies (and automatically handle updates), is by using Android Studio. Once this is installed, start Android Studio and open the Glean project. If Android Studio asks you to upgrade the version of Gradle, decline.

The following dependencies can be installed in Android Studio through Tools > SDK Manager > SDK Tools:

  • Android SDK Tools (may already be selected)
  • NDK r21
  • CMake
  • LLDB

You should be set to build Glean from within Android Studio now.

Manually

Set JAVA_HOME to be the location of Android Studio's JDK which can be found in Android Studio's "Project Structure" menu (you may need to escape spaces in the path).

Note down the location of your SDK. If installed through Android Studio you will find the Android SDK Location in Tools > SDK Manager.

Set the ANDROID_HOME environment variable to that path. Alternatively add the following line to the local.properties file in the root of the Glean checkout (create the file if it does not exist):

sdk.dir=/path/to/sdk

For the Android NDK:

  1. Download NDK r21 from https://developer.android.com/ndk/downloads.
  2. Extract it and put it somewhere ($HOME/.android-ndk-r21 is a reasonable choice, but it doesn't matter).
  3. Add the following line to the local.properties file in the root of the Glean checkout (create the file if it does not exist):
    ndk.dir=/path/to/.android-ndk-r21
    

Setting up Rust

Rust can be installed using rustup, with the following commands:

curl https://sh.rustup.rs -sSf | sh
rustup update

Platform specific toolchains need to be installed for Rust. This can be done using the rustup target command. In order to enable building for real devices and Android emulators, the following targets need to be installed:

rustup target add aarch64-linux-android
rustup target add armv7-linux-androideabi
rustup target add i686-linux-android
rustup target add x86_64-linux-android

Building

Before building:

  • Ensure your repository is up-to-date.
  • Ensure Rust is up-to-date by running rustup update.

With Android Studio

After importing the Glean project into Android Studio it should be all set up and you can build the project using Build > Make Project

Manually

The builds are all performed by ./gradlew and the general syntax used is ./gradlew project:task

Build the whole project and run tests:

./gradlew build

You can see a list of projects by executing ./gradlew projects and a list of tasks by executing ./gradlew tasks.

FAQ

  • Q: Android Studio complains about Python not being found when building.
  • A: Make sure that the python binary is on your PATH. On Windows, in addition to that, it might be required to add its full path to the rust.pythonCommand entry in $project_root/local.properties.

Android SDK / NDK versions

The Glean SDK implementation is currently build against the following versions:

For the full setup see Setup the Android Build Environment.

The versions are defined in the following files. All locations need to be updated on upgrades:

  • Documentation
    • this file (docs/dev/core/internal/sdk-ndk-versions.md)
    • dev/android/setup-android-build-environment.md
  • CI configuration
    • .circleci/config.yml
      • sdkmanager 'build-tools;28.0.3'
      • image: circleci/android:api-28-ndk
    • taskcluster/docker/linux/Dockerfile.
      • ENV ANDROID_BUILD_TOOLS "28.0.3"
      • ENV ANDROID_SDK_VERSION "3859397"
      • ENV ANDROID_PLATFORM_VERSION "28"
      • ENV ANDROID_NDK_VERSION "r21"

Working on unreleased Glean code in android-components

This is a companion to the equivalent instructions for the android-components repository.

Modern Gradle supports composite builds, which allows to substitute on-disk projects for binary publications. Composite builds transparently accomplish what is usually a frustrating loop of:

  1. change library
  2. publish library snapshot to the local Maven repository
  3. consume library snapshot in application

Preparation

Clone the Glean SDK and android-components repositories:

git clone https://github.com/mozilla/glean
git clone https://github.com/mozilla-mobile/android-components

Cargo build targets

By default when building Android Components using gradle, the gradle-cargo plugin will compile Glean for every possible platform, which might end up in failures. You can customize which targets are built in glean/local.properties (more information):

# For physical devices:
rust.targets=arm

# For unit tests:
# rust.targets=darwin # Or linux-*, windows-* (* = x86/x64)

# For emulator only:
rust.targets=x86

Substituting projects

android-components has custom build logic for dealing with composite builds, so you should be able to configure it by simply adding the path to the Glean repository in the correct local.properties file:

In android-components/local.properties:

substitutions.glean.dir=../glean

If this doesn't seem to work, or if you need to configure composite builds for a project that does not contain this custom logic, add the following to settings.gradle:

In android-components/settings.gradle:

includeBuild('../glean') {
  dependencySubstitution {
    substitute module('org.mozilla.telemetry:glean') with project(':glean')
    substitute module('org.mozilla.telemetry:glean-forUnitTests') with project(':glean')
  }
}

Composite builds will ensure the Glean SDK is build as part of tasks inside android-components.

Caveat

There's a big gotcha with library substitutions: the Gradle build computes lazily, and AARs don't include their transitive dependencies' JNI libraries. This means that in android-components, ./gradlew :service-glean:assembleDebug does not invoke :glean:cargoBuild, even though :service-glean depends on the substitution for :glean and even if the inputs to Cargo have changed! It's the final consumer of the :service-glean project (or publication) that will incorporate the JNI libraries.

In practice that means you should always be targeting something that produces an APK: a test or a sample module. Then you should find that the cargoBuild tasks are invoked as you expect.

Inside the android-components repository ./gradlew :samples-glean:connectedAndroidTest should work. Other tests like :service-glean:testDebugUnitTest or :support-sync-telemetry:testDebugUnitTest will currently fail, because the JNI libraries are not included.

Notes

This document is based on the equivalent documentation for application-services: Development with the Reference Browser

  1. Transitive substitutions (as shown above) work but require newer Gradle versions (4.10+).

Using locally-published Glean in Fenix

Note: This is a bit tedious, and you might like to try the substitution-based > approach documented in Working on unreleased Glean code in android-components. That approach is still fairly new, and the local-publishing approach in this document is necessary if it fails.

Note: This is Fenix-specific only in that some links on the page go to the mozilla-mobile/fenix repository, however these steps should work for e.g. reference-browser, as well. (Same goes for Lockwise, or any other consumer of Glean, but they may use a different structure -- Lockwise has no Dependencies.kt, for example)

Preparation

Clone the Glean SDK, android-components and Fenix repositories:

git clone https://github.com/mozilla/glean
git clone https://github.com/mozilla-mobile/android-components
git clone https://github.com/mozilla-mobile/fenix/

Local publishing

  1. Inside the glean repository root:

    1. In .buildconfig.yml, change libraryVersion to end in -TESTING$N1, where $N is some number that you haven't used for this before.

      Example: libraryVersion: 22.0.0-TESTING1

    2. Check your local.properties file, and add rust.targets=x86 if you're testing on the emulator, rust.targets=arm if you're testing on 32-bit arm (arm64 for 64-bit arm, etc). This will make the build that's done in the next step much faster.

    3. Run ./gradlew publishToMavenLocal. This may take a few minutes.

  2. Inside the android-components repository root:

    1. In .buildconfig.yml, change componentsVersion to end in -TESTING$N1, where $N is some number that you haven't used for this before.

      Example: componentsVersion: 24.0.0-TESTING1

    2. Inside buildSrc/src/main/java/Dependencies.kt, change mozilla_glean to reference the libraryVersion you published in step 2 part 1.

      Example: const val mozilla_glean = "22.0.0-TESTING1"

    3. Inside build.gradle, add mavenLocal() inside allprojects { repositories { <here> } }.

    4. Add the following block in the settings.gradle file, before any other statement in the script, in order to have the Glean Gradle plugin loaded from the local Maven repository.

    pluginManagement {
      repositories {
          mavenLocal()
          gradlePluginPortal()
      }
    }
    
    1. Inside the android-component's local.properties file, ensure substitutions.glean.dir is NOT set.

    2. Run ./gradlew publishToMavenLocal.

  3. Inside the fenix repository root:

    1. Inside build.gradle, add mavenLocal() inside allprojects { repositories { <here> } }.

    2. Inside buildSrc/src/main/java/Dependencies.kt, change mozilla_android_components to the version you defined in step 3 part 1.

      Example: const val mozilla_android_components = "24.0.0-TESTING1"

      In the same file change mozilla_glean to the version you defined in step 1 part 1.

      Example: const val mozilla_glean = "22.0.0-TESTING1"

    3. Change the Versions.mozilla_android_components.endsWith('-SNAPSHOT') line in app/build.gradle to Versions.mozilla_android_components.endsWith('-TESTING$N') where $N is the number to reference the version you published in part 2.

You should now be able to build and run Fenix (assuming you could before all this).

Caveats

  1. This assumes you have followed the Android/Rust build setup
  2. Make sure you're fully up to date in all repositories, unless you know you need to not be.
  3. This omits the steps if changes needed because, e.g. Glean made a breaking change to an API used in android-components. These should be understandable to fix, you usually should be able to find a PR with the fixes somewhere in the android-component's list of pending PRs (or, failing that, a description of what to do in the Glean changelog).
  4. Ask in the #glean channel on chat.mozilla.org.

Notes

This document is based on the equivalent documentation for application-services: Using locally-published components in Fenix


1

It doesn't have to end with -TESTING$N, it only needs to have the format -someidentifier. -SNAPSHOT$N is also very common to use, however without the numeric suffix, this has specific meaning to gradle, so we avoid it. Additionally, while the $N we have used in our running example has matched (e.g. all of the identifiers ended in -TESTING1, this is not required, so long as you match everything up correctly at the end).

iOS bindings

The Glean SDK contains the Swift bindings for use by iOS applications. It makes use of the underlying Rust component with a Swift-specific API on top. It also includes integrations into the iOS platform.

Setup the iOS Build Environment

Prepare your build environment

  1. Install Xcode 11.0
  2. Install Carthage: brew install carthage
  3. Ensure you have Python 3 installed: brew install python
  4. Install linting and formatting tools: brew install swiftlint swiftformat
  5. Run bin/bootstrap.sh to download dependencies.
  6. (Optional, only required for building on the CLI) Install xcpretty: gem install xcpretty

Setting up Rust

Rust can be installed using rustup, with the following commands:

  • curl https://sh.rustup.rs -sSf | sh
  • rustup update

Platform specific toolchains need to be installed for Rust. This can be done using the rustup target command. In order to enable building to real devices and iOS emulators, the following targets need to be installed:

rustup target add aarch64-apple-ios x86_64-apple-ios

Building

This should be relatively straightforward and painless:

  1. Ensure your repository is up-to-date.

  2. Ensure Rust is up-to-date by running rustup update.

  3. Run a build using the command make build-swift

    • To run tests use make test-swift

The above can be skipped if using Xcode. The project directory can be imported and all the build details can be left to the IDE.

Debug an iOS application against different builds of Glean

At times it may be necessary to debug against a local build of Glean or another git fork or branch in order to test new features or specific versions of Glean.

Since Glean is consumed through Carthage, this can be as simple as modifying the Cartfile for the consuming application.

Building against the latest Glean

For consuming the latest version of Glean, the Cartfile contents would include the following line:

github "mozilla/glean" "main"

This will fetch and compile Glean from the mozilla/glean GitHub repository from the "main" branch.

Building against a specific release of Glean

For consuming a specific version of Glean, you can specify a branch name, tag name, or commit ID, like this:

github "mozilla/glean" "v0.0.1"

Where v0.0.1 is a tagged release name, but could also be a branch name or a specific commit ID like 832b222

If the custom Glean you wish to build from is a different fork on GitHub, you could simply modify the Cartfile to point at your fork like this:

github "myGitHubHandle/glean" "myBranchName"

Replace the myGitHubHandle and myBranchName with your GitHub handle and branch name, as appropriate.

Build from a locally cloned Glean

You can also use Carthage to build from a local clone by replacing the Cartfile line with the following:

git "file:///Users/yourname/path/to/glean" "localBranchName"

Notice that the initial Carthage command is git now instead of github, and we need to use a file URL of the path to the locally cloned Glean

Perform the Carthage update

One last thing not to forget is to run the carthage update command in the directory your Cartfile resides in order to fetch and build Glean.

Once that is done, your local application should be building against the version of Glean that you specified in the Cartfile.

Python bindings

The Glean SDK contains Python bindings for use in Python applications and test frameworks. It makes use of the underlying Rust component with a Python-specific API on top.

Setup the Python Build Environment

This document describes how to set up an environment for the development of the Glean Python bindings.

Instructions for installing a copy of the Glean Python bindings into your own environment for use in your project are described in adding Glean to your project.

Prerequisites

Glean requires Python 3.6 or later.

Make sure it is installed on your system and accessible on the PATH as python3.

Setting up Rust

If you've already set up Rust for building Glean for Android or iOS, you already have everything you need and can skip this section.

Rust can be installed using rustup, with the following commands:

  • curl https://sh.rustup.rs -sSf | sh
  • rustup update

Create a virtual environment

It is recommended to do all Python development inside a virtual environment to make sure there are no interactions with other Python libraries installed on your system.

You may want to manage your own virtual environment if, for example, you are developing a library that is using Glean and you want to install Glean into it. If you are just developing Glean itself, however, Glean's Makefile will handle the creation and use of a virtual environment automatically (though the Makefile unfortunately does not work on Microsoft Windows).

The instructions below all have both the manual instructions and the Makefile shortcuts.

Manual method

The following instructions use the basic virtual environment functionality that comes in the Python standard library. Other higher level tools exist to manage virtual environments, such as pipenv and conda. These will work just as well, but are not documented here. Using an alternative tool would replace the instructions in this section only, but all other instructions below would remain the same.

To create a virtual environment, enter the following command, replacing <venv> with the path to the virtual environment you want to create. You will need to do this only once.

  $ python3 -m venv <venv>

Then activate the environment. You will need to do this for each new shell session when you want to develop Glean. The command to use depends on your shell environment.

PlatformShellCommand to activate virtual environment
POSIXbash/zshsource <venv>/bin/activate
fish. <venv>/bin/activate.fish
csh/tcshsource <venv>bin/activate.csh
PowerShell Core<venv>/bin/Activate.ps1
Windowscmd.exe<venv>\Scripts\activate.bat
PowerShell<venv>\Scripts\Activate.ps1

Lastly, install Glean's Python development dependencies into the virtual environment.

cd glean-core/python
pip install -r requirements_dev.txt

Makefile method

  $ make python-setup

The default location of the virtual environment used by the make file is glean-core/python/.venvX.Y, where X.Y is the version of Python in use. This makes it possible to build and test for multiple versions of Python in the same checkout.

Note: If you wish to change the location of the virtual environment that the Makefile uses, pass the GLEAN_PYENV environment variable: make python-setup GLEAN_PYENV=mypyenv.

By default, the Makefile installs the latest version available of each of Glean's dependencies. If you wish to install the minimum required versions instead, (useful primarily to ensure that Glean doesn't unintentionally use newer APIs in its dependencies) pass the GLEAN_PYDEPS=min environment variable: make python-setup GLEAN_PYDEPS=min.

Build the Python bindings

Manual method

First, rebuild the Rust core if any Rust changes were made:

  $ cargo build  # If there were Rust changes
  $ cd glean-core/python
  $ python setup.py install

Makefile method

This will implicitly setup the Python environment, rebuild the Rust core and then build the Python bindings.

  $ make build-python

Running the Python unit tests

Manual method

Make sure the Python bindings are built, then:

  $ cd glean-core/python
  $ py.test

Makefile method

The following will run the Python unit tests using py.test:

  $ make test-python

You can send extra parameters to the py.test command by setting the PYTEST_ARGS variable:

  $ make test-python PYTEST_ARGS="-s --pdb"

Viewing logging output

The Glean Python bindings have two sources of log messages: those that come from Python and those that come from Rust.

Python log messages

Python log messages are emitted using the Python standard library's logging module. This module provides a lot of possibilities for customization, but the easiest way to control the log level globally is with logging.basicConfig:

import logging
logging.basicConfig(level=logging.DEBUG)

Rust log messages

Rust log messages are emitted using env_logger. The log level can be controlled with the RUST_LOG environment variable:

import os
os.environ["RUST_LOG"] = "glean_core=DEBUG"

Linting, formatting and type checking

The Glean Python bindings use the following tools:

Manual method

  $ cd glean-core/python
  $ flake8 glean tests
  $ black glean tests
  $ mypy glean

Makefile method

To just check the lints:

  $ make pythonlint

To reformat the Python files in-place:

  $ make pythonfmt

Building the Python API docs

The Python API docs are built using pdoc3.

Manual method

  $ python -m pdoc --html glean --force -o build/docs/python

Makefile method

  $ make python-docs

Building wheels for Linux

Building on Linux using the above instructions will create Linux binaries that dynamically link against the version of libc installed on your machine. This generally will not be portable to other Linux distributions, and PyPI will not even allow it to be uploaded. In order to create wheels that can be installed on the broadest range of Linux distributions, the Python Packaging Authority's manylinux project maintains a Docker image for building compatible Linux wheels.

The CircleCI configuration handles making these wheels from tagged releases. If you need to reproduce this locally, see the CircleCI job pypi-linux-release for an example of how this Docker image is used in practice.

Building wheels for Windows

The official wheels for Windows are produced on a Linux virtual machine using the Mingw toolchain.

The CircleCI configuration handles making these wheels from tagged releases. If you need to reproduce this locally, see the CircleCI job pypi-windows-release for an example of how this is done in practice.

Rust component

The majority of the Glean SDK is implemented as a Rust component, to be usable across all platforms.

This includes:

  • Implementations of all metric types
  • A storage layer
  • A ping serializer

Dependency Management Guidelines

This repository uses third-party code from a variety of sources, so we need to be mindful of how these dependencies will affect our consumers. Considerations include:

We're still evolving our policies in this area, but these are the guidelines we've developed so far.

Rust Code

Unlike Firefox, we do not vendor third-party source code directly into the repository. Instead we rely on Cargo.lock and its hash validation to ensure that each build uses an identical copy of all third-party crates. These are the measures we use for ongoing maintenance of our existing dependencies:

  • Check Cargo.lock into the repository.
  • Generate built artifacts using the --locked flag to cargo build, as an additional assurance that the existing Cargo.lock will be respected.
  • Use cargo-deny for a basic license-compatibility check as part of CI, to guard against human error.

Adding a new dependency, whether we like it or not, is a big deal - that dependency and everything it brings with it will become part of Firefox-branded products that we ship to end users. We try to balance this responsibility against the many benefits of using existing code, as follows:

  • In general, be conservative in adding new third-party dependencies.
    • For trivial functionality, consider just writing it yourself. Remember the cautionary tale of left-pad.
    • Check if we already have a crate in our dependency tree that can provide the needed functionality.
  • Prefer crates that have a a high level of due-diligence already applied, such as:
  • Check that it is clearly licensed and is MPL-2.0 compatible.
  • Take the time to investigate the crate's source and ensure it is suitably high-quality.
    • Be especially wary of uses of unsafe, or of code that is unusually resource-intensive to build.
    • Development dependencies do not require as much scrutiny as dependencies that will ship in consuming applications, but should still be given some thought.
      • There is still the potential for supply-chain compromise with development dependencies!
  • Explicitly describe your consideration of these points in the PR that introduces the new dependency.

Updating to new versions of existing dependencies is a normal part of software development and is not accompanied by any particular ceremony.

Adding a new metric type

Data in the Glean SDK is stored in so-called metrics. You can find the full list of implemented metric types in the user overview.

Adding a new metric type involves defining the metric type's API, its persisted and in-memory storage as well as its serialization into the ping payload.

The metric type's API

A metric type implementation is defined in its own file under glean-core/src/metrics/, e.g. glean-core/src/metrics/counter.rs for a Counter.

Start by defining a structure to hold the metric's metadata:

#[derive(Clone, Debug)]
pub struct CounterMetric {
    meta: CommonMetricData
}

Implement the MetricType trait to create a metric from the meta data as well as expose the meta data. This also gives you a should_record method on the metric type.

impl MetricType for CounterMetric {
    fn meta(&self) -> &CommonMetricData {
        &self.meta
    }

    fn meta_mut(&mut self) -> &mut CommonMetricData {
        &mut self.meta
    }
}

Its implementation should have a way to create a new metric from the common metric data. It should be the same for all metric types.

impl CounterMetric {
    pub fn new(meta: CommonMetricData) -> Self {
        Self { meta }
    }
}

Implement each method for the type. The first argument to accept should always be glean: &Glean, that is: a reference to the Glean object, used to access the storage:

impl CounterMetric { // same block as above
    pub fn add(&self, glean: &Glean, amount: i32) {
        // Always include this check!
        if !self.should_record() {
            return;
        }

        // Do error handling here

        glean
            .storage()
            .record_with(&self.meta, |old_value| match old_value {
                Some(Metric::Counter(old_value)) => Metric::Counter(old_value + amount),
                _ => Metric::Counter(amount),
            })
    }
}

Use glean.storage().record() to record a fixed value or glean.storage.record_with() to construct a new value from the currently stored one.

The storage operation makes use of the metric's variant of the Metric enumeration.

The Metric enumeration

Persistence and in-memory serialization as well as ping payload serialization are handled through the Metric enumeration. This is defined in glean-core/src/metrics/mod.rs. Variants of this enumeration are used in the storage implementation of the metric type.

To add a new metric type, include the metric module and declare its use, then add a new variant to the Metric enum:


mod counter;

// ...

pub use self::counter::CounterMetric;

#[derive(Serialize, Deserialize, Debug, Clone)]
pub enum Metric {
    // ...
    Counter(i32),
}

Then modify the below implementation and define the right ping section name for the new type. This will be used in the ping payload:

impl Metric {
    pub fn ping_section(&self) -> &'static str {
        match self {
            // ...
            Metric::Counter(_) => "counter",
        }
    }
}

Finally, define the ping payload serialization (as JSON). In the simple cases where the in-memory representation maps to its JSON representation it is enough to call the json! macro.

impl Metric { // same block as above
    pub fn as_json(&self) -> JsonValue {
        match self {
            // ...
            Metric::Counter(c) => json!(c),
        }
    }
}

For more complex serialization consider implementing serialization logic as a function returning a serde_json::Value or another object that can be serialized.

For example, the DateTime serializer has the following entry, where get_iso_time_string is a function to convert from the DateTime metric representation to a string:

Metric::Datetime(d, time_unit) => json!(get_iso_time_string(*d, *time_unit)),

In the next step we will create the FFI wrapper and platform-specific wrappers.

Adding a new metric type - FFI layer

In order to use a new metric type over the FFI layer, it needs implementations in the FFI component.

FFI component

The FFI component implementation can be found in glean-core/ffi/src. Each metric type is implemented in its own module.

Add a new file named after your metric, e.g. glean-core/ffi/src/counter.rs, and declare it in glean-core/ffi/src/lib.rs with mod counter;.

In the metric type module define your metric type using the define_metric macro. This allows referencing the metric name and defines the global map as well as some common functions such as the constructor and destructor. Simple operations can be also defined in the same macro invocation:

use crate::{define_metric, handlemap_ext::HandleMapExtension, GLEAN};

define_metric!(CounterMetric => COUNTER_METRICS {
    new           -> glean_new_counter_metric(),
    destroy       -> glean_destroy_counter_metric,

    add -> glean_counter_add(amount: i32),
});

More complex operations need to be defined as plain functions. For example the test helper function for a counter metric can be defined as:

#[no_mangle]
pub extern "C" fn glean_counter_test_has_value(
    glean_handle: u64,
    metric_id: u64,
    storage_name: FfiStr,
) -> u8 {
    GLEAN.call_infallible(glean_handle, |glean| {
        COUNTER_METRICS.call_infallible(metric_id, |metric| {
            metric
                .test_get_value(glean, storage_name.as_str())
                .is_some()
        })
    })
}

Adding a new metric type - Kotlin

FFI

The platform-specific FFI wrapper needs the definitions of these new functions. For Kotlin this is in glean-core/android/src/main/java/mozilla/telemetry/glean/rust/LibGleanFFI.kt:

fun glean_new_counter_metric(category: String, name: String, send_in_pings: StringArray, send_in_pings_len: Int, lifetime: Int, disabled: Byte): Long
fun glean_destroy_counter_metric(handle: Long)
fun glean_counter_add(glean_handle: Long, metric_id: Long, amount: Int)

Kotlin API

Finally, create a platform-specific metric type wrapper. For Kotlin this would be glean-core/android/src/main/java/mozilla/telemetry/glean/private/CounterMetricType.kt:

class CounterMetricType(
    private var handle: Long,
    private val disabled: Boolean,
    private val sendInPings: List<String>
) {
    /**
     * The public constructor used by automatically generated metrics.
     */
    constructor(
        disabled: Boolean,
        category: String,
        lifetime: Lifetime,
        name: String,
        sendInPings: List<String>
    ) : this(handle = 0, disabled = disabled, sendInPings = sendInPings) {
        val ffiPingsList = StringArray(sendInPings.toTypedArray(), "utf-8")
        this.handle = LibGleanFFI.INSTANCE.glean_new_counter_metric(
                category = category,
                name = name,
                send_in_pings = ffiPingsList,
                send_in_pings_len = sendInPings.size,
                lifetime = lifetime.ordinal,
                disabled = disabled.toByte())
    }

    fun add(amount: Int = 1) {
        if (disabled) {
            return
        }

        @Suppress("EXPERIMENTAL_API_USAGE")
        Dispatchers.API.launch {
            LibGleanFFI.INSTANCE.glean_counter_add(
                Glean.handle,
                this@CounterMetricType.handle,
                amount)
        }
    }

    @VisibleForTesting(otherwise = VisibleForTesting.NONE)
    fun testHasValue(pingName: String = sendInPings.first()): Boolean {
        @Suppress("EXPERIMENTAL_API_USAGE")
        Dispatchers.API.assertInTestingMode()

        val res = LibGleanFFI.INSTANCE.glean_counter_test_has_value(Glean.handle, this.handle, pingName)
        return res.toBoolean()
    }
}

Adding a new metric type - Swift

FFI

Swift can re-use the generated C header file. Re-generate it with

make cbindgen

Swift API

Finally, create a platform-specific metric type wrapper. For Swift this would be glean-core/ios/Glean/Metrics/CounterMetric.swift:

public class CounterMetricType {
    let handle: UInt64
    let disabled: Bool
    let sendInPings: [String]

    /// The public constructor used by automatically generated metrics.
    public init(category: String, name: String, sendInPings: [String], lifetime: Lifetime, disabled: Bool) {
        self.disabled = disabled
        self.sendInPings = sendInPings
        self.handle = withArrayOfCStrings(sendInPings) { pingArray in
            glean_new_counter_metric(
                category,
                name,
                pingArray,
                Int32(sendInPings.count),
                lifetime.rawValue,
                disabled.toByte()
            )
        }
    }

    public func add(amount: Int32 = 1) {
        guard !self.disabled else { return }

        _ = Dispatchers.shared.launch {
            glean_counter_add(Glean.shared.handle, self.handle, amount)
        }
    }

    func testHasValue(_ pingName: String? = nil) -> Bool {
        let pingName = pingName ?? self.sendInPings[0]
        return glean_counter_test_has_value(Glean.shared.handle, self.handle, pingName) != 0
    }

    func testGetValue(_ pingName: String? = nil) throws -> Int32 {
        let pingName = pingName ?? self.sendInPings[0]

        if !testHasValue(pingName) {
            throw "Missing value"
        }

        return glean_counter_test_get_value(Glean.shared.handle, self.handle, pingName)
    }
}

Adding a new metric type - Python

FFI

Python can use the generated C header file directly, through cffi. Re-generate it with

make cbindgen

Python API

Finally, create a platform-specific metric type wrapper. For Python this would be glean-core/python/glean/metrics/counter.py:

class CounterMetricType:
    """
    This implements the developer facing API for recording counter metrics.

    Instances of this class type are automatically generated by
    `glean.load_metrics`, allowing developers to record values that were
    previously registered in the metrics.yaml file.

    The counter API only exposes the `CounterMetricType.add` method, which
    takes care of validating the input data and making sure that limits are
    enforced.
    """

    def __init__(
        self,
        disabled: bool,
        category: str,
        lifetime: Lifetime,
        name: str,
        send_in_pings: List[str],
    ):
        self._disabled = disabled
        self._send_in_pings = send_in_pings

        self._handle = _ffi.lib.glean_new_counter_metric(
            _ffi.ffi_encode_string(category),
            _ffi.ffi_encode_string(name),
            _ffi.ffi_encode_vec_string(send_in_pings),
            len(send_in_pings),
            lifetime.value,
            disabled,
        )

    def __del__(self):
        if getattr(self, "_handle", 0) != 0:
            _ffi.lib.glean_destroy_counter_metric(self._handle)

    def add(self, amount: int = 1):
        """
        Add to counter value.

        Args:
            amount (int): (default: 1) This is the amount to increment the
                counter by.
        """
        if self._disabled:
            return

        @Dispatcher.launch
        def add():
            _ffi.lib.glean_counter_add(self._handle, amount)

    def test_has_value(self, ping_name: Optional[str] = None) -> bool:
        """
        Tests whether a value is stored for the metric for testing purposes
        only.

        Args:
            ping_name (str): (default: first value in send_in_pings) The name
                of the ping to retrieve the metric for.

        Returns:
            has_value (bool): True if the metric value exists.
        """
        if ping_name is None:
            ping_name = self._send_in_pings[0]

        return bool(
            _ffi.lib.glean_counter_test_has_value(
                self._handle, _ffi.ffi_encode_string(ping_name)
            )
        )

    def test_get_value(self, ping_name: Optional[str] = None) -> int:
        """
        Returns the stored value for testing purposes only.

        Args:
            ping_name (str): (default: first value in send_in_pings) The name
                of the ping to retrieve the metric for.

        Returns:
            value (int): value of the stored metric.
        """
        if ping_name is None:
            ping_name = self._send_in_pings[0]

        if not self.test_has_value(ping_name):
            raise ValueError("metric has no value")

        return _ffi.lib.glean_counter_test_get_value(
            self._handle, _ffi.ffi_encode_string(ping_name)
        )

    def test_get_num_recorded_errors(
        self, error_type: ErrorType, ping_name: Optional[str] = None
    ) -> int:
        """
        Returns the number of errors recorded for the given metric.

        Args:
            error_type (ErrorType): The type of error recorded.
            ping_name (str): (default: first value in send_in_pings) The name
                of the ping to retrieve the metric for.

        Returns:
            num_errors (int): The number of errors recorded for the metric for
                the given error type.
        """
        if ping_name is None:
            ping_name = self._send_in_pings[0]

        return _ffi.lib.glean_counter_test_get_num_recorded_errors(
            self._handle, error_type.value, _ffi.ffi_encode_string(ping_name),
        )

The new metric type also needs to be imported from glean-core/python/glean/metrics/__init__.py:

from .counter import CounterMetricType


__all__ = [
    "CounterMetricType",
    # ...
]

It also must be added to the _TYPE_MAPPING in glean-core/python/glean/_loader.py:

_TYPE_MAPPING = {
    "counter": metrics.CounterMetricType,
    # ...
}

FFI layer

The core part of the Glean SDK provides an FFI layer for its API. This FFI layer can be consumed by platform-specific bindings, such as the Kotlin bindings for Android.

When to use what method of passing data between Rust and Java/Swift

There are a bunch of options here. For the purposes of our discussion, there are two kinds of values you may want to pass over the FFI.

  1. Types with identity (includes stateful types, resource types, or anything that isn't really serializable).
  2. Plain ol' data.

Types with identity

Examples of this are all metric type implementations, e.g. StringMetric. These types are complex, implemented in Rust and make use of the global Glean singleton on the Rust side, They have an equivalent on the wrapper side (Kotlin/Swift/Python), that forwards calls through FFI.

They all follow the same pattern:

A ConcurrentHandleMap stores all instances of the same type, A handle is passed back and forth as a u64 from Rust, Long from Kotlin, UInt64 from Swift or other equivalent type in other languages.

This is recommended for most cases, as it's the hardest to mess up. Additionally, for types T such that &T: Sync + Send, or that you need to call &mut self method, this is the safest choice.

Additionally, this will ensure panic-safety, as it will poison the internal Mutex, making further access impossible.

The ffi_support::handle_map docs are good, and under ConcurrentHandleMap include an example of how to set this up.

Plain Old Data

This includes both primitive values, strings, arrays, or arbitrarily nested structures containing them.

Primitives

Numeric primitives are the easiest to pass between languages. The main recommendation is: use the equivalent and same-sized type as the one provided by Rust.

There are a couple of exceptions/caveats, especially for Kotlin. All of them are caused by JNA/Android issues. Swift has very good support for calling over the FFI.

  1. bool: Don't use it. JNA doesn't handle it well. Instead, use a numeric type (like u8) and represent 0 for false and 1 for true for interchange over the FFI, converting back to Kotlin's Boolean or Swift's Bool after (as to not expose this somewhat annoying limitation in our public API). All wrappers already include utility functions to turn 8-bit integers (u8) back to booleans (toBool() or equivalent methods).

  2. usize/isize: These cause the structure size to be different based on the platform. JNA does handle this if you use NativeSize, but it's awkward. Use the size-defined integers instead, such as i64/i32 and their language-equivalents (Kotlin: Long/Int, Swift:UInt64/UInt32). Caution: In Kotlin integers are signed by default. You can use u64/u32 for Long/Int if you ensure the values are non-negative through asserts or error reporting code.

  3. char: Avoid these if possible. If you really have a use case consider u32 instead.

    If you do this, you should probably be aware of the fact that Java chars are 16 bit, and Swift Characters are actually strings (they represent Extended Grapheme Clusters, not codepoints).

Strings

These we pass as null-terminated UTF-8 C-strings.

For return values, used *mut c_char, and for input, use ffi_support::FfiStr

  1. If the string is returned from Rust to Kotlin/Swift, you need to expose a string destructor from your ffi crate. See ffi_support::define_string_destructor!).

    For converting to a *mut c_char, use either rust_string_to_c if you have a String, or opt_rust_string_to_c for Option<String> (None becomes std::ptr::null_mut()).

    Important: In Kotlin, the type returned by a function that produces this must be Pointer, and not String, and the parameter that the destructor takes as input must also be Pointer.

    Using String will almost work. JNA will convert the return value to String automatically, leaking the value Rust provides. Then, when passing to the destructor, it will allocate a temporary buffer, pass it to Rust, which we'll free, corrupting both heaps 💥. Oops!

  2. If the string is passed into Rust from Kotlin/Swift, the Rust code should declare the parameter as a FfiStr<'_>. and things should then work more or less automatically. The FfiStr has methods for extracting it's data as &str, Option<&str>, String, and Option<String>.

Aggregates

This is any type that's more complex than a primitive or a string (arrays, structures, and combinations there-in). There are two options we recommend for these cases:

  1. Passing data as JSON. This is very easy and useful for prototyping, but is much slower and requires a great deal of copying and redundant encode/decode steps (in general, the data will be copied at least 4 times to make this work, and almost certainly more in practice). It can be done relatively easily by derive(Serialize, Deserialize), and converting to a JSON string using serde_json::to_string.

    This is a viable option for test-only functions, where the overhead is not important.

  2. Use repr(C) structures and copy the data across the boundary, carefully replicating the same structure on the wrapper side. In Kotlin this will require @Structure.FieldOrder annotations. Swift can directly handle C types.

    Caution: This is error prone! Structures, enumerations and unions need to be kept the same across all layers (Rust, generated C header, Kotlin, Swift, ...). Be extra careful, avoid adding references to structures and ensure the structures are correctly freed inside Rust. Copy out the data and convert into language-appropriate types (e.g. convert *mut c_char into Swift/Kotlin strings) as soon as possible.

Contributing to the Glean SDK

Anyone is welcome to help with the Glean SDK project. Feel free to get in touch with other community members on chat.mozilla.org or through issues on GitHub or Bugzilla.

Participation in this project is governed by the Mozilla Community Participation Guidelines.

Bug Reports

To report issues or request changes, file a bug in Bugzilla in Data Platform & Tools :: Glean: SDK.

If you don't have a Bugzilla account, we also accept issues on GitHub.

Making Code Changes

To work on the code in this repository you will need to be familiar with the Rust programming language. You can get a working rust compiler and toolchain via rustup.

You can check that everything compiles by running the following from the root of your checkout:

make test-rust

If you plan to work on the Android component bindings, you should also review the instructions for setting up an Android build environment.

To run all Kotlin tests:

make test-kotlin

or run tests in Android Studio.

To run all Swift tests:

make test-swift

or run tests in Xcode.

Sending Pull Requests

Patches should be submitted as pull requests (PRs).

Before submitting a PR:

  • Your code must run and pass all the automated tests before you submit your PR for review.
  • "Work in progress" pull requests are allowed to be submitted, but should be clearly labeled as such and should not be merged until all tests pass and the code has been reviewed.
  • For changes to Rust code
    • make test-rust produces no test failures
    • make clippy runs without emitting any warnings or errors.
    • make rustfmt does not produce any changes to the code.
  • For changes to Kotlin code
    • make test-kotlin runs without emitting any warnings or errors.
    • make ktlint runs without emitting any warnings.
  • For changes to Swift code
    • make test-swift (or running tests in Xcode) runs without emitting any warnings or errors.
    • make swiftlint runs without emitting any warnings or errors.
    • make swiftfmt runs without emitting any warnings or producing changes.
  • Your patch should include new tests that cover your changes. It is your and your reviewer's responsibility to ensure your patch includes adequate tests.

When submitting a PR:

  • You agree to license your code under the project's open source license (MPL 2.0).
  • Base your branch off the current main.
  • Add both your code and new tests if relevant.
  • Please do not include merge commits in pull requests; include only commits with the new relevant code.

Code Review

This project is production Mozilla code and subject to our engineering practices and quality standards. Every patch must be peer reviewed by a member of the Glean core team.

Reviewers are defined in the CODEOWNERS file and are automatically added for every pull request. Every pull request needs to be approved by at least one of these people before landing.

The submitter needs to decide on their own discretion whether the changes require a look from more than a single reviewer or any outside developer. Reviewers can also ask for additional approval from other reviewers.

Release

See the Release process on how to release a new version of the Glean SDK.

Code Coverage

In computer science, test coverage is a measure used to describe the degree to which the source code of a program is executed when a particular test suite runs. A program with high test coverage, measured as a percentage, has had more of its source code executed during testing, which suggests it has a lower chance of containing undetected software bugs compared to a program with low test coverage. (Wikipedia)

Generating Kotlin reports locally

Locally you can generate a coverage report with the following command:

./gradlew -Pcoverage :glean:build

After that you'll find an HTML report at the following location:

glean-core/android/build/reports/jacoco/jacocoTestReport/jacocoTestReport/html/index.html

Generating Rust reports locally

Generating the Rust coverage report requires a significant amount of RAM during the build.

We use grcov to collect and aggregate code coverage information. Releases can be found on the grcov Release page.

The build process requires a Rust Nightly version. Install it using rustup:

rustup toolchain add nightly

To generate an HTML report, genhtml from the lcov package is required. Install it through your system's package manager.

After installation you can build the Rust code and generate a report:

export CARGO_INCREMENTAL=0
export RUSTFLAGS='-Zprofile -Ccodegen-units=1 -Cinline-threshold=0 -Clink-dead-code -Coverflow-checks=off -Zno-landing-pads'
cargo +nightly test

zip -0 ccov.zip `find . \( -name "glean*.gc*" \) -print`
grcov ccov.zip -s . -t lcov --llvm --branch --ignore-not-existing --ignore-dir "/*" -o lcov.info
genhtml -o report/ --show-details --highlight --ignore-errors source --legend lcov.info

After that you'll find an HTML report at the following location:

report/index.html

Generating Swift reports locally

Xcode automatically generates code coverage when running tests. You can find the report in the Report Navigator (View -> Navigators -> Show Report Navigator -> Coverage).

Generating Python reports locally

Python code coverage is determined using the coverage.py library.

Run

make python-coverage

to generate code coverage reports in the Glean virtual environment.

After running, the report will be in htmlcov.

Glean release process

The Glean SDK consists of multiple libraries for different platforms and targets. The main supported libraries are released as one. Development happens on the main repository https://github.com/mozilla/glean. See Contributing for how to contribute changes to the Glean SDK.

The development & release process roughly follows the GitFlow model.

Note: The rest of this section assumes that upstream points to the https://github.com/mozilla/glean repository, while origin points to the developer fork. For some developer workflows, upstream can be the same as origin.

Table of Contents:

Published artifacts

Standard Release

Releases can only be done by one of the Glean maintainers.

  • Main development branch: main
  • Main release branch: release
  • Specific release branch: release-vX.Y.Z
  • Hotfix branch: hotfix-X.Y.(Z+1)

Create a release branch

  1. Create a release branch from the main branch:
    git checkout -b release-v25.0.0 main
    
  2. Update the changelog .
    1. Add any missing important changes under the Unreleased changes headline.
    2. Commit any changes to the changelog file due to the previous step.
  3. Run bin/prepare-release.sh <new version> to bump the version number.
    1. The new version should be the next patch, minor or major version of what is currently released.
    2. Let it create a commit for you.
  4. Push the new release branch:
    git push upstream release-v25.0.0
    
  5. Wait for CI to finish on that branch and ensure it's green: https://circleci.com/gh/mozilla/glean/tree/release-v25.0.0
  6. Apply additional commits for bug fixes to this branch.
    • Adding large new features here is strictly prohibited. They need to go to the main branch and wait for the next release.

Finish a release branch

When CI has finished and is green for your specific release branch, you are ready to cut a release.

  1. Check out the main release branch:
    git checkout release
    
  2. Merge the specific release branch:
    git merge --no-ff release-v25.0.0
    
  3. Push the main release branch:
    git push upstream release
    
  4. Tag the release on GitHub:
    1. Draft a New Release in the GitHub UI (Releases > Draft a New Release).
    2. Enter v<myversion> as the tag. It's important this is the same as the version you specified to the prepare_release.sh script, with the v prefix added.
    3. Select the release branch as the target.
    4. Under the description, paste the contents of the release notes from CHANGELOG.md.
  5. Wait for the CI build to complete for the tag.
  6. Release the Rust crates:
    cd glean-core
    cargo publish --verbose
    cd ffi
    cargo publish --verbose
    
  7. Send a pull request to merge back the specific release branch to the development branch: https://github.com/mozilla/glean/compare/main...release-v25.0.0?expand=1
    • This is important so that no changes are lost.
    • This might have merge conflicts with the main branch, which you need to fix before it is merged.
  8. Once the above pull request lands, delete the specific release branch.

Hotfix release for latest version

If the latest released version requires a bug fix, a hotfix branch is used.

Create a hotfix branch

  1. Create a hotfix branch from the main release branch:
    git checkout -b hotfix-v25.0.1 release
    
  2. Run bin/prepare-release.sh <new version> to bump the version number.
    1. The new version should be the next patch version of what is currently released.
    2. Let it create a commit for you.
  3. Push the hotfix branch:
    git push upstream hotfix-v25.0.1
    
  4. Create a local hotfix branch for bugfixes:
    git checkout -b bugfix hotfix-v25.0.1
    
  5. Fix the bug and commit the fix in one or more separate commits.
  6. Push your bug fixes and create a pull request against the hotfix branch: https://github.com/mozilla/glean/compare/hotfix-v25.0.1...your-name:bugfix?expand=1
  7. When that pull request lands, wait for CI to finish on that branch and ensure it's green: https://circleci.com/gh/mozilla/glean/tree/hotfix-v25.0.1

Finish a hotfix branch

When CI has finished and is green for your hotfix branch, you are ready to cut a release, similar to a normal release:

  1. Check out the main release branch:
    git checkout release
    
  2. Merge the hotfix branch:
    git merge --no-ff hotfix-v25.0.1
    
  3. Push the main release branch:
    git push upstream release
    
  4. Tag the release on GitHub:
    1. Draft a New Release in the GitHub UI (Releases > Draft a New Release).
    2. Enter v<myversion> as the tag. It's important this is the same as the version you specified to the prepare_release.sh script, with the v prefix added.
    3. Select the release branch as the target.
    4. Under the description, paste the contents of the release notes from CHANGELOG.md.
  5. Wait for the CI build to complete for the tag.
  6. Release the Rust crates:
    cd glean-core
    cargo publish --verbose
    cd ffi
    cargo publish --verbose
    
  7. Send a pull request to merge back the hotfix branch to the development branch: https://github.com/mozilla/glean/compare/main...hotfix-v25.0.1?expand=1
    • This is important so that no changes are lost.
    • This might have merge conflicts with the main branch, which you need to fix before it is merged.
  8. Once the above pull request lands, delete the hotfix branch.

Hotfix release for previous version

If you need to release a hotfix for a previously released version (that is: not the latest released version), you need a support branch.

Note: This should rarely happen. We generally support only the latest released version of Glean.

Create a support and hotfix branch

  1. Create a support branch from the version tag and push it:
    git checkout -b support/v24.0 v24.0.0
    git push upstream support/v24.0
    
  2. Create a hotfix branch for this support branch:
    git checkout -b hotfix-v24.0.1 support/v24.0
    
  3. Fix the bug and commit the fix in one or more separate commits into your hotfix branch.
  4. Push your bug fixes and create a pull request against the support branch: https://github.com/mozilla/glean/compare/support/v24.0...your-name:hotfix-v24.0.1?expand=1
  5. When that pull request lands, wait for CI to finish on that branch and ensure it's green: https://circleci.com/gh/mozilla/glean/tree/support/v24.0

Finish a support branch

  1. Check out the support branch:
    git checkout support/v24.0
    
  2. Update the changelog .
    1. Add any missing important changes under the Unreleased changes headline.
    2. Commit any changes to the changelog file due to the previous step.
  3. Run bin/prepare-release.sh <new version> to bump the version number.
    1. The new version should be the next patch version of the support branch.
    2. Let it create a commit for you.
  4. Push the support branch:
    git push upstream support/v24.0
    
  5. Tag the release on GitHub:
    1. Draft a New Release in the GitHub UI (Releases > Draft a New Release).
    2. Enter v<myversion> as the tag. It's important this is the same as the version you specified to the prepare_release.sh script, with the v prefix added.
    3. Select the support branch (e.g. support/v24.0) as the target.
    4. Under the description, paste the contents of the release notes from CHANGELOG.md.
  6. Wait for the CI build to complete for the tag.
  7. Release the Rust crates:
    cd glean-core
    cargo publish --verbose
    cd ffi
    cargo publish --verbose
    
  8. Send a pull request to merge back any bug fixes to the development branch: https://github.com/mozilla/glean/compare/main...support/v24.0?expand=1
    • This is important so that no changes are lost.
    • This might have merge conflicts with the main branch, which you need to fix before it is merged.
  9. Once the above pull request lands, delete the support branch.

Upgrading android-components to a new version of Glean

On Android, Mozilla products consume the Glean SDK through its wrapper in android-components. Therefore, when a new Glean SDK release is made, android-components must also be updated.

After following one of the above instructions to make a Glean SDK release:

  1. Ensure that CI has completed and the artifacts are published on Mozilla's Maven repository.

  2. Create a pull request against android-components to update the Glean version with the following changes:

    • The Glean version is updated in the mozilla_glean variable in the buildSrc/src/main/java/Dependencies.kt file.

    • The relevant parts of the Glean changelog copied into the top part of the android-components changelog. This involves copying the Android-specific changes and the general changes to Glean, but can omit other platform-specific changes.

IMPORTANT: Until the Glean Gradle plugin work is complete, all downstream consumers of android-components will also need to update their version of Glean to match the version used in android-components so that their unit tests can run correctly.

In Fenix, for example, the Glean version is specified here.

Internal documentation

This chapter describes aspects of Glean that are internal implementation details.

This includes:

Reserved ping names

The Glean SDK reserves all ping names in send_in_pings starting with glean_.

This currently includes, but is not limited to:

  • glean_client_info
  • glean_internal_info

Additionally, only Glean may specify all-pings. This special value has no effect in the client, but indicates to the backend infrastructure that a metric may appear in any ping.

Clearing metrics when disabling/enabling Glean

When disabling upload (Glean.setUploadEnabled(false)), metrics are also cleared to prevent their storage on the local device, and lessen the likelihood of accidentally sending them. There are a couple of exceptions to this:

  • first_run_date is retained so it isn't reset if metrics are later re-enabled.

  • client_id is set to the special value "c0ffeec0-ffee-c0ff-eec0-ffeec0ffeec0". This should make it possible to detect the error case when metrics are sent from a client that has been disabled.

When re-enabling metrics:

  • first_run_date is left as-is. (It should remain a correct date of first run of the application, unaffected by disabling/enabling the Glean SDK).

  • The client_id is set to a newly-generated random UUID. It has no connection to the client_id used prior to disabling the Glean SDK.

  • Application lifetime metrics owned by Glean are regenerated from scratch so that they will appear in subsequent pings. This is the same process that happens during every startup of the application when the Glean SDK is enabled. The application is also responsible for setting any application lifetime metrics that it manages at this time.

  • Ping lifetime metrics do not need special handling. They will begin recording again after metric uploading is re-enabled.

Payload format

The main sections of a Glean ping are described in Ping Sections. This Payload format chapter describes details of the ping payload that are relevant for decoding Glean pings in the pipeline. This is less relevant for end users of the Glean SDK.

JSON Schema

Glean's ping payloads have a formal JSON schema defined in the mozilla-pipeline-schemas project. It is written as a set of templates that are expanded by the mozilla-pipeline-schemas build infrastructure into a fully expanded schema.

Metric types

Boolean

A Boolean is represented by its boolean value.

Example

true

Counter

A Counter is represented by its integer value.

Example

17

Quantity

A Quantity is represented by its integer value.

Example

42

String

A String is represented by its string value.

Example

"sample string"

String list

A String List is represented as an array of strings.

["sample string", "another one"]

Timespan

A Timespan is represented as an object of their duration as an integer and the time unit.

Field nameTypeDescription
valueIntegerThe value in the marked time unit.
time_unitStringThe time unit, see the timespan's configuration for valid values.

Example

{
    "time_unit": "milliseconds",
    "value": 10
}

Timing Distribution

A Timing distribution is represented as an object with the following fields.

Field nameTypeDescription
sumIntegerThe sum of all recorded values.
valuesMap<String, Integer>The values in each bucket. The key is the minimum value for the range of that bucket.

A contiguous range of buckets is always sent, so that the server can aggregate and visualize distributions, without knowing anything about the specific bucketing function used. This range starts with the first bucket with a non-zero accumulation, and ends at one bucket beyond the last bucket with a non-zero accumulation (so that the upper bound on the last bucket is retained).

For example, the following shows the recorded values vs. what is sent in the payload.

recorded:  1024: 2, 1116: 1,                   1448: 1,
sent:      1024: 2, 1116: 1, 1217: 0, 1327: 0, 1448: 1, 1579: 0

Example:

{
    "sum": 4612,
    "values": {
        "1024": 2,
        "1116": 1,
        "1217": 0,
        "1327": 0,
        "1448": 1,
        "1579": 0
    }
}

Memory Distribution

A Memory distribution is represented as an object with the following fields.

Field nameTypeDescription
sumIntegerThe sum of all recorded values.
valuesMap<String, Integer>The values in each bucket. The key is the minimum value for the range of that bucket.

A contiguous range of buckets is always sent. See timing distribution for more details.

Example:

{
    "sum": 3,
    "values": {
        "0": 1,
        "1": 3,
    }
}

UUID

A UUID is represented by the string representation of the UUID.

Example

"29711dc8-a954-11e9-898a-eb4ea7e8fd3f"

Datetime

A Datetime is represented by its ISO8601 string representation, truncated to the metric's time unit. It always includes the timezone offset.

Example

"2019-07-18T14:06:00.000+02:00"

Event

Events are represented as an array of objects, with one object for each event. Each event object has the following keys:

Field nameTypeDescription
timestampIntegerA monotonically increasing timestamp value, in milliseconds. To avoid leaking absolute times, the first timestamp in the array is always zero, and subsequent timestamps in the array are relative to that reference point.
categoryStringThe event's category. This comes directly from the category under which the metric was defined in the metrics.yaml file.
nameStringThe event's name, as defined in the metrics.yaml file.
extraObject (optional)Extra data associated with the event. Both the keys and values of this object are strings. The keys must be from the set defined for this event in the metrics.yaml file. The values have a maximum length of 50 bytes, when encoded as UTF-8.

Example

[
  {
    "timestamp": 0,
    "category": "app",
    "name": "ss_menu_opened"
  },
  {
    "timestamp": 124,
    "category": "search",
    "name": "performed_search",
    "extra": {
      "source": "default.action"
    }
  }
]

Also see the JSON schema for events.

To avoid losing events when the application is killed by the operating system, events are queued on disk as they are recorded. When the application starts up again, there is no good way to determine if the device has rebooted since the last run and therefore any timestamps recorded in the new run could not be guaranteed to be consistent with those recorded in the previous run. To get around this, on application startup, any queued events are immediately collected into pings and then cleared. These "startup-triggered pings" are likely to have a very short duration, as recorded in ping_info.start_time and ping_info.end_time (see the ping_info section). The maximum timestamp of the events in these pings are quite likely to exceed the duration of the ping, but this is to be expected.

Custom Distribution

A Custom distribution is represented as an object with the following fields.

Field nameTypeDescription
sumIntegerThe sum of all recorded values.
valuesMap<String, Integer>The values in each bucket. The key is the minimum value for the range of that bucket.

A contiguous range of buckets is always sent, so that the server can aggregate and visualize distributions, without knowing anything about the specific bucketing function used. This range starts with the first bucket (as specified in the range_min parameter), and ends at one bucket beyond the last bucket with a non-zero accumulation (so that the upper bound on the last bucket is retained).

For example, suppose you had a custom distribution defined by the following parameters:

  • range_min: 10
  • range_max: 200
  • bucket_count: 80
  • histogram_type: 'linear'

The following shows the recorded values vs. what is sent in the payload.

recorded:        12: 2,                      22: 1
sent:     10: 0, 12: 2, 14: 0, 17: 0, 19: 0, 22: 1, 24: 0

Example:

{
    "sum": 3,
    "values": {
        "10": 0,
        "12": 2,
        "14": 0,
        "17": 0,
        "19": 0,
        "22": 1,
        "24": 0
    }
}

Labeled metrics

Currently several labeled metrics are supported:

All are on the top-level represented in the same way, as an object mapping the label to the metric's value. See the individual metric types for details on the value payload:

Example for Labeled Counters

{
    "label1": 2,
    "label2": 17
}

Directory structure

This page describes the contents of the directories where Glean stores its data.

All Glean data is inside a single root directory with the name glean_data.

On Android, this directory lives inside the ApplicationInfo.dataDir directory associated with the application.

On iOS, this directory lives inside the Documents directory associated with the application.

For the Python bindings, if no directory is specified, it is stored in a temporary directory and cleared at exit.

Within the glean_data directory are the following contents:

  • db: Contains the rkv database used to persist ping and user lifetime metrics.

  • events: Contains flat files containing persisted events before they are collected into pings.

  • pending_pings: Pings are written here before they are picked up by the ping uploader to send to the submission endpoint.

  • deletion_request: The deletion-request ping is written here before it is picked up by the ping uploader. This directory is separate from the pending_pings directory above, in or for an uploader to pick up only deletion-request pings and send them after general upload is disabled.

  • tmp: Pings are written here and then moved to the pending_pings directory when finished to make sure that partially-written pings to not get queued for sending.
    (The standard system temporary directory is not used for this because it is not guaranteed to be on the same volume as the glean_data directory on Android).

Debug Pings

For debugging and testing purposes Glean allows to tag pings, which are then available in the Debug Ping Viewer1.

Pings are sent to the same endpoint as all pings, with the addition of one HTTP header:

X-Debug-ID: <tag>

<tag> is a alphanumeric string with a maximum length of 20 characters, used to identify pings in the Debug Ping Viewer.

See Debugging products using the Glean SDK for detailed information how to use this mechanism in applications.


1

Requires a Mozilla login.

Upload mechanism

The glean-core Rust crate does not handle the ping upload directly. Network stacks vastly differ between platforms, applications and operating systems. The Glean SDK leverages the available platform capabilities to implement any network communication.

Glean core controls all upload and coordinates the platform side with its own internals. All language bindings implement ping uploading around a common API and protocol.

Upload task API

The following diagram visualizes the communication between Glean core (the Rust crate), a Glean language binding (e.g. the Kotlin or Swift implementation) and a Glean end point server.

sequenceDiagram
    participant Glean core
    participant Glean wrapper
    participant Server

    Glean wrapper->>Glean core: get_upload_task()
    Glean core->>Glean wrapper: Task::Upload(PingRequest)
    Glean wrapper-->>Server: POST /submit/{task.id}
    Server-->>Glean wrapper: 200 OK
    Glean wrapper->>Glean core: upload_response(200)
    Glean wrapper->>Glean core: get_upload_task()
    Glean core->>Glean wrapper: Task::Done

Glean core will take care of file management, cleanup, rescheduling and rate limiting1.

1 Rate limiting is achieved by limiting the amount of times a language binding is allowed to get a Task::Upload(PingRequest) from get_upload_task in a given time interval. Currently, the default limit is for a maximum of 10 upload tasks every 60 seconds and there are no exposed methods that allow changing this default (follow Bug 1647630 for updates). If the caller has reached the maximum tasks for the current interval, they will get a Task::Wait regardless if there are other Task::Upload(PingRequest)s queued.

Available APIs

For direct Rust consumers the global Glean object provides these methods:


#![allow(unused_variables)]
fn main() {
/// Gets the next task for an uploader.
fn get_upload_task(&self) -> PingUploadTask

/// Processes the response from an attempt to upload a ping.
fn process_ping_upload_response(&self, uuid: &str, status: UploadResult)
}

See the documentation for further usage and explanation of the additional types:

For FFI consumers (e.g. Kotlin/Swift/Python implementations) these functions are available:


#![allow(unused_variables)]
fn main() {
/// Gets the next task for an uploader. Which can be either:
extern "C" fn glean_get_upload_task(result: *mut FfiPingUploadTask)

/// Processes the response from an attempt to upload a ping.
extern "C" fn glean_process_ping_upload_response(task: *mut FfiPingUploadTask, status: u32)
}

See the documentation for additional information about the types:

Implementations

Project NameLanguage BindingsOperating SystemApp Lifecycle TypeEnvironment Data source
glean-coreRustallallnone
glean-ffiCallallnone
glean-preview1RustWindows/Mac/LinuxDesktop applicationOS info build-time autodetected, app info passed in
Glean AndroidKotlin, JavaAndroidMobile appAutodetected from the Android environment
Glean iOSSwiftiOSMobile appAutodetected from the iOS environment
Glean.pyPythonWindows/Mac/LinuxallAutodetected at runtime
FOG2Rust/C++/JavaScriptas Firefox supportsDesktop applicationOS info build-time autodetected, app info passed in

1

glean-preview is an experimental crate for prototyping integration into Firefox. It it not recommended for general use. See Project FOG.

2

Firefox on Glean (FOG) is the name of the layer that integrates the Glean SDK into Firefox Desktop. It is currently being designed and implemented. It is being used as a test bed for how best to write a generic Rust language binding layer and so temporarily ties directly to glean-core instead of an API crate. Prospective non-mozilla-central Rust consumers of the Glean SDK should not follow its example and should instead follow bug 1631768 for updates on when the proper language binding crate will be available.

The following language-specific API docs are available:

Appendix

The following sections contain other material related to Glean.

Glossary

A glossary with explanations and background for wording used in the Glean project.

Glean

According to the dictionary the word “glean” means:

to gather information or material bit by bit

Glean is the combination of the Glean SDK, the Glean pipeline & Glean tools.

See also: Glean - product analytics & telemetry.

Glean Pipeline

The general data pipeline is the infrastructure that collects, stores, and analyzes telemetry data from our products and logs from various services. See An overview of Mozilla’s Data Pipeline.

The Glean pipeline additionally consists of

Glean SDK

The Glean SDK is the bundle of libraries with support for different platforms. The source code is available at https://github.com/mozilla/glean.

Glean SDK book

This documentation.

Glean tools

Glean provides additional tools for its usage:

Metric

Metrics are the individual things being measured using Glean. They are defined in metrics.yaml files, also known as registry files.

Glean itself provides some metrics out of the box.

Ping

A ping is an entity used to bundle related metrics. The Glean SDK provides default pings and allows for custom ping, see Glean Pings.

Submission

"To submit" means to collect & to enqueue a ping for uploading.

The Glean SDK stores locally all the metrics set by it or by its clients. Each ping has its own schedule to gather all its locally saved metrics and create a JSON payload with them. This is called "collection".

Upon successful collection, the payload is queued for upload, which may not happen immediately or at all (in case network connectivity is not available).

Unless the user has defined their own custom pings, they don’t need to worry too much about submitting pings. All the default pings have their scheduling and submission handled by the SDK.

This Week in Glean (TWiG)

This Week in Glean is a series of blog posts that the Glean Team at Mozilla is using to try to communicate better about our work.

Unreleased changes

Full changelog

v31.3.0 (2020-07-10)

Full changelog

  • General
    • Remove locale from baseline ping. (1609968, #1016)
    • Persist X-Debug-ID header on store ping. (1605097, #1042)
    • BUGFIX: raise an error if Glean is initialized with an empty string as the application_id (#1043).
  • Python
    • BUGFIX: correctly set the app_build metric to the newly provided application_build_id initialization option (#1031).
    • The Python bindings now report networking errors in the glean.upload.ping_upload_failure metric (like all the other bindings) (#1039).
    • Python default upgraded to Python 3.8 (#995)
  • iOS
    • BUGFIX: Make LabeledMetric subscript public, so consuming applications can actually access it (#1027)

v31.2.3 (2020-06-29)

Full changelog

  • General
    • Move debug view tag management to the Rust core. (1640575, #998)
    • BUGFIX: Fix mismatch in events keys and values by using glean_parser version 1.23.0.

v31.2.2 (2020-06-26)

Full changelog

  • Android
    • BUGFIX: Compile dependencies with NDEBUG to avoid linking unavailable symbols. This fixes a crash due to a missing stderr symbol on older Android (#1020)

v31.2.1 (2020-06-25)

Full changelog

  • Python
    • BUGFIX: Core metrics are now present in every ping, even if submit is called before initialize has a chance to complete. (#1012)

v31.2.0 (2020-06-24)

Full changelog

  • General
    • Add rate limiting capabilities to the upload manager. (1543612, #974)
  • Android
    • BUGFIX: baseline pings with reason "dirty startup" are no longer sent if Glean did not full initialize in the previous run (#996).
  • Python
    • Support for Python 3.5 was dropped (#987).
    • Python wheels are now shipped with Glean release builds, resulting in much smaller libraries (#1002)
    • The Python bindings now use locale.getdefaultlocale() rather than locale.getlocale() to determine the locale (#1004).

v31.1.2 (2020-06-23)

Full changelog

  • General
    • BUGFIX: Correctly format the date and time in the Date header (#993).
  • Python
    • BUGFIX: Additional time is taken at shutdown to make sure pings are sent and telemetry is recorded. (1646173, #983)
    • BUGFIX: Glean will run on the main thread when running in a multiprocessing subprocess (#986).

v31.1.1 (2020-06-12)

Full changelog

  • Android
    • Dropping the version requirement for lifecycle extensions down again. Upping the required version caused problems in A-C.

v31.1.0 (2020-06-11)

Full changelog

  • General:
    • The regex crate is no longer required, making the Glean binary smaller (#949)
    • Record upload failures into a new metric (#967)
    • Log FFI errors as actual errors (#935)
    • Limit the number of upload retries in all implementations (#953, #968)
  • Python
    • Additional safety guarantees for applications that use Python threading (#962)

v31.0.2 (2020-05-29)

Full changelog

  • Rust
    • Fix list of included files in published crates

v31.0.1 (2020-05-29)

Full changelog

  • Rust
    • Relax version requirement for flate2 for compatibility reasons

v31.0.0 (2020-05-28)

Full changelog

  • General:

    • The version of glean_parser has been upgraded to v1.22.0
      • A maximum of 10 extra_keys is now enforced for event metric types.
      • Breaking change: (Swift only) Combine all metrics and pings into a single generated file Metrics.swift.
        • For Swift users this requires to change the list of output files for the sdk_generator.sh script. It now only needs to include the single file Generated/Metrics.swift.
  • Python:

    • BUGFIX: lifetime: application metrics are no longer recorded as lifetime: user.
    • BUGFIX: glean-core is no longer crashing when calling uuid.set with invalid UUIDs.
    • Refactor the ping uploader to use the new upload mechanism and add gzip compression.
    • Most of the work in Glean.initialize happens on a worker thread and no longer blocks the main thread.
  • Rust:

    • Expose Datetime types to Rust consumers.

v30.1.0 (2020-05-22)

Full changelog

  • Android & iOS
    • Ping payloads are now compressed using gzip.
  • iOS
    • Glean.initialize is now a no-op if called from an embedded extension. This means that Glean will only run in the base application process in order to prevent extensions from behaving like separate applications with different client ids from the base application. Applications are responsible for ensuring that extension metrics are only collected within the base application.
  • Python
    • lifetime: application metrics are now cleared after the Glean-owned pings are sent, after the product starts.
    • Glean Python bindings now build in a native Windows environment.
    • BUGFIX: MemoryDistributionMetric now parses correctly in metrics.yaml files.
    • BUGFIX: Glean will no longer crash if run as part of another library's coverage testing.

v30.0.0 (2020-05-13)

Full changelog

  • General:
    • We completely replaced how the upload mechanism works. glean-core (the Rust part) now controls all upload and coordinates the platform side with its own internals. All language bindings implement ping uploading around a common API and protocol. There is no change for users of Glean, the language bindings for Android and iOS have been adopted to the new mechanism already.
    • Expose RecordedEvent and DistributionData types to Rust consumers (#876)
    • Log crate version at initialize (#873)
  • Android:
    • Refactor the ping uploader to use the new upload mechanism.
  • iOS:
    • Refactor the ping uploader to use the new upload mechanism.

v29.1.1 (2020-05-22)

Full changelog

  • Android
    • BUGFIX: Fix a race condition that leads to a ConcurrentModificationException. Bug 1635865

v29.1.0 (2020-05-11)

Full changelog

  • General:
    • The version of glean_parser has been upgraded to v1.20.4
      • BUGFIX: yamllint errors are now reported using the correct file name.
    • The minimum and maximum values of a timing distribution can now be controlled by the time_unit parameter. See bug 1630997 for more details.

v29.0.0 (2020-05-05)

Full changelog

  • General:
    • The version of glean_parser has been upgraded to v1.20.2 (#827):
      • Breaking change: glinter errors found during code generation will now return an error code.
      • glean_parser now produces a linter warning when user lifetime metrics are set to expire. See bug 1604854 for additional context.
  • Android:
    • The PingType.submit() can now be called without a null by Java consumers (#853).
  • Python:
    • BUGFIX: Fixed a race condition in the atexit handler, that would have resulted in the message "No database found" (#854).
    • The Glean FFI header is now parsed at build time rather than runtime. Relevant for packaging in PyInstaller, the wheel no longer includes glean.h and adds _glean_ffi.py (#852).
    • The minimum versions of many secondary dependencies have been lowered to make the Glean SDK compatible with more environments.
    • Dependencies that depend on the version of Python being used are now specified using the Declaring platform specific dependencies syntax in setuptools. This means that more recent versions of dependencies are likely to be installed on Python 3.6 and later, and unnecessary backport libraries won't be installed on more recent Python versions.
  • iOS:
    • Glean for iOS is now being built with Xcode 11.4.1 (#856)

v28.0.0 (2020-04-23)

Full changelog

  • General:
    • The baseline ping is now sent when the application goes to foreground, in addition to background and dirty-startup.
  • Python:
    • BUGFIX: The ping uploader will no longer display a trace back when the upload fails due to a failed DNS lookup, network outage, or related issues that prevent communication with the telemetry endpoint.
    • The dependency on inflection has been removed.
    • The Python bindings now use subprocess rather than multiprocessing to perform ping uploading in a separate process. This should be more compatible on all of the platforms Glean supports.

v27.1.0 (2020-04-09)

Full changelog

  • General:
    • BUGFIX: baseline pings sent at startup with the dirty_startup reason will now include application lifetime metrics (#810)
  • iOS:
    • Breaking change: Change Glean iOS to use Application Support directory #815. No migration code is included. This will reset collected data if integrated without migration. Please contact the Glean SDK team if this affects you.
  • Python
    • BUGFIX: Fixed a race condition between uploading pings and deleting the temporary directory on shutdown of the process.

v27.0.0 (2020-04-08)

Full changelog

  • General
    • Glean will now detect when the upload enabled flag changes outside of the application, for example due to a change in a config file. This means that if upload is disabled while the application wasn't running (e.g. between the runs of a Python command using the Glean SDK), the database is correctly cleared and a deletion request ping is sent. See #791.
    • The events ping now includes a reason code: startup, background or max_capacity.
  • iOS:
    • BUGFIX: A bug where the metrics ping is sent immediately at startup on the last day of the month has been fixed.
    • Glean for iOS is now being built with Xcode 11.4.0
    • The measure convenience function on timing distributions and time spans will now cancel the timing if the measured function throws, then rethrow the exception (#808)
    • Broken doc generation has been fixed (#805).
  • Kotlin
    • The measure convenience function on timing distributions and time spans will now cancel the timing if the measured function throws, then rethrow the exception (#808)
  • Python:
    • Glean will now wait at application exit for up to one second to let its worker thread complete.
    • Ping uploading now happens in a separate child process by default. This can be disabled with the allow_multiprocessing configuration option.

v26.0.0 (2020-03-27)

Full changelog

  • General:
    • The version of glean_parser has been updated to 1.19.0:
      • Breaking change: The regular expression used to validate labels is stricter and more correct.
      • Add more information about pings to markdown documentation:
        • State whether the ping includes client id;
        • Add list of data review links;
        • Add list of related bugs links.
      • glean_parser now makes it easier to write external translation functions for different language targets.
      • BUGFIX: glean_parser now works on 32-bit Windows.
  • Android:
    • gradlew clean will no longer remove the Miniconda installation in ~/.gradle/glean. Therefore clean can be used without reinstalling Miniconda afterward every time.
  • Python:
    • Breaking Change: The glean.util and glean.hardware modules, which were unintentionally public, have been made private.
    • Most Glean work and I/O is now done on its own worker thread. This brings the parallelism Python in line with the other platforms.
    • The timing distribution, memory distribution, string list, labeled boolean and labeled string metric types are now supported in Python (#762, #763, #765, #766)

v25.1.0 (2020-02-26)

Full changelog

  • Python:
    • The Boolean, Datetime and Timespan metric types are now supported in Python (#731, #732, #737)
    • Make public, document and test the debugging features (#733)

v25.0.0 (2020-02-17)

Full changelog

  • General:
    • ping_type is not included in the ping_info any more (#653), the pipeline takes the value from the submission URL.
    • The version of glean_parser has been upgraded to 1.18.2:
      • Breaking Change (Java API) Have the metrics names in Java match the names in Kotlin. See Bug 1588060.
      • The reasons a ping are sent are now included in the generated markdown documentation.
  • Android:
    • The Glean.initialize method runs mostly off the main thread (#672).
    • Labels in labeled metrics now have a correct, and slightly stricter, regular expression. See label format for more information.
  • iOS:
    • The baseline ping will now include reason codes that indicate why it was submitted. If an unclean shutdown is detected (e.g. due to force-close), this ping will be sent at startup with reason: dirty_startup.
    • Per Bug 1614785, the clearing of application lifetime metrics now occurs after the metrics ping is sent in order to preserve values meant to be included in the startup metrics ping.
    • initialize() now performs most of its work in a background thread.
  • Python:
    • When the pre-init task queue overruns, this is now recorded in the metric glean.error.preinit_tasks_overflow.
    • glinter warnings are printed to stderr when loading metrics.yaml and pings.yaml files.

v24.2.0 (2020-02-11)

Full changelog

  • General:
    • Add locale to client_info section.
    • Deprecation Warning Since locale is now in the client_info section, the one in the baseline ping (glean.baseline.locale) is redundant and will be removed by the end of the quarter.
    • Drop the Glean handle and move state into glean-core (#664)
    • If an experiment includes no extra fields, it will no longer include {"extra": null} in the JSON payload.
    • Support for ping reason codes was added.
    • The metrics ping will now include reason codes that indicate why it was submitted.
    • The version of glean_parser has been upgraded to 1.17.3
  • Android:
    • Collections performed before initialization (preinit tasks) are now dispatched off the main thread during initialization.
    • The baseline ping will now include reason codes that indicate why it was submitted. If an unclean shutdown is detected (e.g. due to force-close), this ping will be sent at startup with reason: dirty_startup.
  • iOS:
    • Collections performed before initialization (preinit tasks) are now dispatched off the main thread and not awaited during initialization.
    • Added recording of glean.error.preinit_tasks_overflow to report when the preinit task queue overruns, leading to data loss. See bug 1609734

v24.1.0 (2020-01-16)

Full changelog

  • General:
    • Stopping a non started measurement in a timing distribution will now be reported as an invalid_state error.
  • Android:
    • A new metric glean.error.preinit_tasks_overflow was added to report when the preinit task queue overruns, leading to data loss. See bug 1609482

v24.0.0 (2020-01-14)

Full changelog

  • General:
    • Breaking Change An enableUpload parameter has been added to the initialize() function. This removes the requirement to call setUploadEnabled() prior to calling the initialize() function.
  • Android:
    • The metrics ping scheduler will now only send metrics pings while the application is running. The application will no longer "wake up" at 4am using the Work Manager.
    • The code for migrating data from Glean SDK before version 19 was removed.
    • When using the GleanTestLocalServer rule in instrumented tests, pings are immediately flushed by the WorkManager and will reach the test endpoint as soon as possible.
  • Python:
    • The Python bindings now support Python 3.5 - 3.7.
    • The Python bindings are now distributed as a wheel on Linux, macOS and Windows.

v23.0.1 (2020-01-08)

Full changelog

  • Android:
    • BUGFIX: The Glean Gradle plugin will now work if an app or library doesn't have a metrics.yaml or pings.yaml file.
  • iOS:
    • The released iOS binaries are now built with Xcode 11.3.

v23.0.0 (2020-01-07)

Full changelog

  • Python bindings:
    • Support for events and UUID metrics was added.
  • Android:
    • The Glean Gradle Plugin correctly triggers docs and API updates when registry files change, without requiring them to be deleted.
    • parseISOTimeString has been made 4x faster. This had an impact on Glean migration and initialization.
    • Metrics with lifetime: application are now cleared when the application is started, after startup Glean SDK pings are generated.
  • All platforms:
    • The public method PingType.send() (in all platforms) have been deprecated and renamed to PingType.submit().
    • Rename deletion_request ping to deletion-request ping after glean_parser update

v22.1.0 (2019-12-17)

Full changelog

  • Add InvalidOverflow error to TimingDistributions (#583)

v22.0.0 (2019-12-05)

Full changelog

  • Add option to defer ping lifetime metric persistence (#530)
  • Add a crate for the nice control API (#542)
  • Pending deletion_request pings are resent on start (#545)

v21.3.0 (2019-12-03)

Full changelog

  • Timers are reset when disabled. That avoids recording timespans across disabled/enabled toggling (#495).
  • Add a new flag to pings: send_if_empty (#528)
  • Upgrade glean_parser to v1.12.0
  • Implement the deletion request ping in Glean (#526)

v21.2.0 (2019-11-21)

Full changelog

  • All platforms

    • The experiments API is no longer ignored before the Glean SDK initialized. Calls are recorded and played back once the Glean SDK is initialized.

    • String list items were being truncated to 20, rather than 50, bytes when using .set() (rather than .add()). This has been corrected, but it may result in changes in the sent data if using string list items longer than 20 bytes.

v21.1.1 (2019-11-20)

Full changelog

  • Android:

    • Use the LifecycleEventObserver interface, rather than the DefaultLifecycleObserver interface, since the latter isn't compatible with old SDK targets.

v21.1.0 (2019-11-20)

Full changelog

  • Android:

    • Two new metrics were added to investigate sending of metrics and baseline pings. See bug 1597980 for more information.

    • Glean's two lifecycle observers were refactored to avoid the use of reflection.

  • All platforms:

    • Timespans will now not record an error if stopping after setting upload enabled to false.

v21.0.0 (2019-11-18)

Full changelog

  • Android:

    • The GleanTimerId can now be accessed in Java and is no longer a typealias.

    • Fixed a bug where the metrics ping was getting scheduled twice on startup.

  • All platforms

    • Bumped glean_parser to version 1.11.0.

v20.2.0 (2019-11-11)

Full changelog

  • In earlier 20.x.x releases, the version of glean-ffi was incorrectly built against the wrong version of glean-core.

v20.1.0 (2019-11-11)

Full changelog

  • The version of Glean is included in the Glean Gradle plugin.

  • When constructing a ping, events are now sorted by their timestamp. In practice, it rarely happens that event timestamps are unsorted to begin with, but this guards against a potential race condition and incorrect usage of the lower-level API.

v20.0.0 (2019-11-11)

Full changelog

  • Glean users should now use a Gradle plugin rather than a Gradle script. (#421) See integrating with the build system docs for more information.

  • In Kotlin, metrics that can record errors now have a new testing method, testGetNumRecordedErrors. (#401)

v19.1.0 (2019-10-29)

Full changelog

  • Fixed a crash calling start on a timing distribution metric before Glean is initialized. Timings are always measured, but only recorded when upload is enabled (#400)
  • BUGFIX: When the Debug Activity is used to log pings, each ping is now logged only once (#407)
  • New invalid state error, used in timespan recording (#230)
  • Add an Android crash instrumentation walk-through (#399)
  • Fix crashing bug by avoiding assert-printing in LMDB (#422)
  • Upgrade dependencies, including rkv (#416)

v19.0.0 (2019-10-22)

Full changelog

First stable release of Glean in Rust (aka glean-core). This is a major milestone in using a cross-platform implementation of Glean on the Android platform.

  • Fix round-tripping of timezone offsets in dates (#392)
  • Handle dynamic labels in coroutine tasks (#394)

v0.0.1-TESTING6 (2019-10-18)

Full changelog

  • Ignore dynamically stored labels if Glean is not initialized (#374)
  • Make sure ProGuard doesn't remove Glean classes from the app (#380)
  • Keep track of pings in all modes (#378)
  • Add jnaTest dependencies to the forUnitTest JAR (#382)

v0.0.1-TESTING5 (2019-10-10)

Full changelog

  • Upgrade to NDK r20 (#365)

v0.0.1-TESTING4 (2019-10-09)

Full changelog

  • Take DST into account when converting a calendar into its items (#359)
  • Include a macOS library in the forUnitTests builds (#358)
  • Keep track of all registered pings in test mode (#363)

v0.0.1-TESTING3 (2019-10-08)

Full changelog

  • Allow configuration of Glean through the GleanTestRule
  • Bump glean_parser version to 1.9.2

v0.0.1-TESTING2 (2019-10-07)

Full changelog

  • Include a Windows library in the forUnitTests builds

v0.0.1-TESTING1 (2019-10-02)

Full changelog

General

First testing release.

This Week in Glean (TWiG)

“This Week in Glean” is a series of blog posts that the Glean Team at Mozilla is using to try to communicate better about our work. They could be release notes, documentation, hopes, dreams, or whatever: so long as it is inspired by Glean.

Blog posts