nimbus/stateful/

dbcache.rs

/* This Source Code Form is subject to the terms of the Mozilla Public
 * License, v. 2.0. If a copy of the MPL was not distributed with this
 * file, You can obtain one at https://mozilla.org/MPL/2.0/. */

use std::collections::{HashMap, HashSet};
use std::sync::{Arc, RwLock};

use crate::enrollment::{
    EnrolledFeature, EnrolledFeatureConfig, ExperimentEnrollment, map_features_by_feature_id,
};
use crate::error::{NimbusError, Result, warn};
use crate::stateful::enrollment::get_enrollments;
use crate::stateful::gecko_prefs::GeckoPrefStore;
use crate::stateful::persistence::{Database, StoreId, Writer};
use crate::{EnrolledExperiment, Experiment};

// This module manages an in-memory cache of the database, so that some
// functions exposed by nimbus can return results without blocking on any
// IO. Consumers are expected to call our public `update()` function whenever
// the database might have changed.

// This struct is the cached data. This is never mutated, but instead
// recreated every time the cache is updated.
struct CachedData {
    pub experiments: Vec<Experiment>,
    pub enrollments: Vec<ExperimentEnrollment>,
    pub experiments_by_slug: HashMap<String, EnrolledExperiment>,
    pub features_by_feature_id: HashMap<String, EnrolledFeatureConfig>,
    pub gecko_pref_to_enrollment_slugs: Option<HashMap<String, HashSet<String>>>,
}

// This is the public cache API. Each NimbusClient can create one of these and
// it lives as long as the client - it encapsulates the synchronization needed
// to allow the cache to work correctly.
#[derive(Default)]
pub struct DatabaseCache {
    data: RwLock<Option<CachedData>>,
}

impl DatabaseCache {
    // Call this function whenever it's possible that anything cached by this
    // struct (eg, our enrollments) might have changed.
    //
    // This function must be passed a `&Database` and a `Writer`, which it
    // will commit before updating the in-memory cache. This is a slightly weird
    // API but it helps enforce two important properties:
    //
    //  * By requiring a `Writer`, we ensure mutual exclusion of other db writers
    //    and thus prevent the possibility of caching stale data.
    //  * By taking ownership of the `Writer`, we ensure that the calling code
    //    updates the cache after all of its writes have been performed.
    //  * `update_gecko_prefs` - Pass true for regular enrollment changes. Pass false
    //     when the Gecko prefs do not need to be synced with Gecko.
    pub fn commit_and_update(
        &self,
        db: &Database,
        writer: Writer,
        coenrolling_ids: &HashSet<&str>,
        gecko_pref_store: Option<Arc<GeckoPrefStore>>,
        update_gecko_prefs: bool,
    ) -> Result<()> {
        // By passing in the active `writer` we read the state of enrollments
        // as written by the calling code, before it's committed to the db.
        let enrollments = get_enrollments(db, &writer)?;

        // Build a lookup table for experiments by experiment slug.
        // This will be used for get_experiment_branch() and get_active_experiments()
        let mut experiments_by_slug = HashMap::with_capacity(enrollments.len());
        for e in enrollments {
            experiments_by_slug.insert(e.slug.clone(), e);
        }

        let enrollments: Vec<ExperimentEnrollment> =
            db.get_store(StoreId::Enrollments).collect_all(&writer)?;
        let experiments: Vec<Experiment> =
            db.get_store(StoreId::Experiments).collect_all(&writer)?;

        let features_by_feature_id =
            map_features_by_feature_id(&enrollments, &experiments, coenrolling_ids);

        let gecko_pref_to_enrollment_slugs = gecko_pref_store.map(|store| {
            store.map_gecko_prefs_to_enrollment_slugs_and_update_store(
                &experiments,
                &enrollments,
                &experiments_by_slug,
                update_gecko_prefs,
            )
        });

        // This is where testing tools would override i.e. replace experimental feature configurations.
        // i.e. testing tools would cause custom feature configs to be stored in a Store.
        // Here, we get those overrides out of the store, and merge it with this map.

        // This is where rollouts (promoted experiments on a given feature) will be merged in to the feature variables.

        let data = CachedData {
            experiments,
            enrollments,
            experiments_by_slug,
            features_by_feature_id,
            gecko_pref_to_enrollment_slugs,
        };

        // Try to commit the change to disk and update the cache as close
        // together in time as possible. This leaves a small window where another
        // thread could read new data from disk but see old data in the cache,
        // but that seems benign in practice given the way we use the cache.
        // The alternative would be to lock the cache while we commit to disk,
        // and we don't want to risk blocking the main thread.
        writer.commit()?;
        let mut cached = self.data.write().unwrap();
        cached.replace(data);
        Ok(())
    }

    // Abstracts safely referencing our cached data.
    //
    // WARNING: because this manages locking, the callers of this need to be
    // careful regarding deadlocks - if the callback takes other own locks then
    // there's a risk of locks being taken in an inconsistent order. However,
    // there's nothing this code specifically can do about that.
    fn get_data<T, F>(&self, func: F) -> Result<T>
    where
        F: FnOnce(&CachedData) -> T,
    {
        match *self.data.read().unwrap() {
            None => {
                warn!("DatabaseCache attempting to read data before initialization is completed");
                Err(NimbusError::DatabaseNotReady)
            }
            Some(ref data) => Ok(func(data)),
        }
    }

    pub fn get_experiment_branch(&self, id: &str) -> Result<Option<String>> {
        self.get_data(|data| -> Option<String> {
            data.experiments_by_slug
                .get(id)
                .map(|experiment| experiment.branch_slug.clone())
        })
    }

    // This gives access to the feature JSON. We pass it as a string because uniffi doesn't
    // support JSON yet.
    pub fn get_feature_config_variables(&self, feature_id: &str) -> Result<Option<String>> {
        self.get_data(|data| {
            let enrolled_feature = data.features_by_feature_id.get(feature_id)?;
            let string = serde_json::to_string(&enrolled_feature.feature.value).unwrap();
            Some(string)
        })
    }

    pub fn get_enrollment_by_feature(&self, feature_id: &str) -> Result<Option<EnrolledFeature>> {
        self.get_data(|data| {
            data.features_by_feature_id
                .get(feature_id)
                .map(|feature| feature.into())
        })
    }

    pub fn get_active_experiments(&self) -> Result<Vec<EnrolledExperiment>> {
        self.get_data(|data| {
            data.experiments_by_slug
                .values()
                .map(|e| e.to_owned())
                .collect::<Vec<EnrolledExperiment>>()
        })
    }

    pub fn get_experiments(&self) -> Result<Vec<Experiment>> {
        self.get_data(|data| data.experiments.to_vec())
    }

    pub fn get_enrollments(&self) -> Result<Vec<ExperimentEnrollment>> {
        self.get_data(|data| data.enrollments.to_owned())
    }

    pub fn get_enrollments_for_pref(&self, pref: &str) -> Result<Option<HashSet<String>>> {
        self.get_data(|data| {
            if let Some(a) = &data.gecko_pref_to_enrollment_slugs {
                Ok(a.get(pref).cloned())
            } else {
                Ok(None)
            }
        })?
    }
}