Here’s what happened on the MozMEAO SRE team from October 24th - October 31st.
Current work
SUMO
An initial infra project structure for SUMO, along with S3 + Cloudfront distributions for dev, stage and production environments has been created in this PR.
Future SUMO migration tasks will be tracked here, while infrastructure code will be stored here.
MDN
MDN attachments are now behind a Cloudfront CDN to help reduce load on the MDN web pods.
Here’s what happened in September in
Kuma,
the engine of
MDN Web Docs:
Ran Maintenance Mode Tests in AWS
Updated Article Styling
Continued Conversion to Browser Compat Data
Shipped Tweaks and Fixes
Here’s the plan for October:
Move MDN to AWS
Improve Performance of the Interactive Editor
Done in September
Ran Maintenance Mode Tests in AWS
Back in March 2017, we added Maintenance Mode to Kuma, which allows the site
content to be available when we can’t write to the database. This mode got
its first workout this month, as we put MDN into Maintenance Mode in SCL3,
and then sent an increasing percentage of public traffic to an MDN
deployment in AWS.
We ran 3 tests in September. In the first, we just tried Maintenance Mode with
production traffic in SCL3. In the second
test we sent 5% of traffic to
AWS, and in the third test we
ramped it up to 15%, then 50%, and finally 100%. The most recent test,
on October 3, included New Relic monitoring, which gave us useful data
and pretty charts.
Web Transactions Time shows how the average request is
handled by the different services. For the SCL3 side, you can see a steady
improvement in transaction time from 125 to 75 ms, as more traffic is handled
by AWS.
On the AWS side, the response time grows from 40 to 90 ms, as the DNS
configuration sends 100% of traffic to the new cluster.
The Web Transaction Percentiles chart shows useful statistics beyond the
average. For example, 99% of users see at least 375 ms response time, and the
median is at 50 ms.
On the AWS side, 99% of users see at least 350 ms response time (slightly better), and the
median is at 100 ms (slightly worse).
Finally, Throughput measures the requests handled per minute. SCL3
continued handling over 500 requests per minute during the test. This may be
due to clients using old DNS records, or because KumaScript continues making
requests to render out-of-date pages.
AWS ramped up to over 2000 requests per minute during the test, easily handing
the load of a US afternoon.
We consider this a successful test. Our AWS environment can easily handle
regular, read-only MDN traffic, with capacity to spare. We don’t expect
MDN users to notice much of a difference when we make the change.
Updated Article Styling
We’re working on the next phase of redesigning MDN. We’re looking at ways to
present MDN articles, to make them easier to read, to scan quickly, and to
emphasize the most useful information. We’re testing some ideas with users, and
some of the adjustments showed up on the site this month.
For example, MDN documents a lot of code in prose, such as HTML element and
attribute names. In PR 4400,
Stephanie Hobson added a highlight
background to make these stand out.
Before PR 4400, a fixed-width font was used to display literals:
After PR 4000, the literals stand out with a light grey background:
There’s a lot that goes into making text on the web readable (see
Stephanie’s slides
from her talk at #a11yTOConf for some suggestions).
One of the things we can do with the default style is to try to make lines
about 50-75 characters wide. On the other hand, code examples don’t wrap well,
and we want to make them stand out. We’re experimenting with style changes
for line length with beta testers, using some of the ideas from
blog.mozilla.org. For example,
PR 4402 expands the sample output,
making the examples stand out from the rest of the page.
Before PR 4402, the examples shared the text’s narrow width:
After PR 4402, the example is as wide as the code samples, and the buttons restyled:
We’ll test more adjustments with beta testers and in individual user tests.
Some of these we’ll ship immediately, and others will inform the article
redesign.
Continued Conversion to Browser Compat Data
The Browser Compat Data (BCD) project now includes all the HTML and JavaScript
compatibility data from MDN. 1,500 MDN pages now generate their compatibility
tables from this data. Only 4,500 more to go!
The BCD project was the most active MDN project in September. There were 159
commits over 90 pull requests. These PRs came from from 18 different
contributors, bringing the total to 50 contributors. There’s over 58,000
additional lines in the project. 13 of these PRs are from Daniel D.
Beck, who is joining the MDN team as a contractor.
This progress was made possible by
Florian Scholz,
Jean-Yves Perrier, and
wbamberg, who quickly and accurately reviewed
the PRs, working out issues and getting them merged. Florian has also started
a weekly release of the npm package, and we’re up to
mdn-browser-compat-data 0.0.8.
BCD PR 416:
Add Safari 11’s added support for the Intl JavaScript API, from
first-time contributor
Andy VanWagoner.
BCD PR 377
Firefox 57 will support Tab.discarded state, from first-time contributor
Joachim.
BCD PR 420:
Add ‘font-weight’ CSS properties, the first of 5 PRs from
first-time BCD contributor mfluehr.
mfluehr also contributed 4 PRs to KumaScript, and kept the mdn/data project
active with PR 107, documenting the
::placeholder pseudo-element.
Kuma PR 4434:
Wrap task completion call-to-action in gettext, so that we can start
getting better feedback from MDN users who see non-English pages, from
Stephanie Hobson.
Planned for October
Work will continue to migrate to Browser Compat Data, and to fix issues with the
redesign and the new interactive examples.
Move MDN to AWS
This week, we’ll complete our functional testing of MDN, making sure that
page editing and other read/write tests are working, and that the rarely used
features continue to work.
On Tuesday October 10, we’ll put SCL3 in Maintenance Mode again, move the
database, and come back with MDN in AWS.
We’ve done a lot of preparation, but we expect something to break, so we’re
planning on fixing AWS-related bugs in October. The AWS move will also allow
us to improve our deployment processes, helping us ship features faster.
If things go smoothly, we have plenty of other work lined up, such as style
improvements, SEO-related tweaks, updating to Django 1.11, and getting
KumaScript UI strings into Pontoon.
Improve Performance of the Interactive Editor
We’re continuing the
beta test for the interactive editor.
The feedback has been overwhelming positive, but we’re not happy with the page
speed impact. We’ll continue work in October to improve performance.
In the meantime, contractor Mark Boas is preparing
examples for the launch, such as 26 examples for JavaScript expressions and
operators (PR 286).
Here’s what happened on the MozMEAO SRE team from September 26th - October 3rd.
Current work
MDN Migration to AWS
We’ve successfully completed a series of tests against MDN hosted in AWS, but we have a few more to complete before moving to AWS.
Testing
A successful MDN maintenance mode test was performed on Tuesday October 3rd 2017, at 2pm eastern / 11 pacific.
Migration work
Restrict URLs for untrusted (files / samples) and CDN domains. PR 529
New Relic support has been added to the MDN Kubernetes deployments in these PRs: 549, 548, 547, 542
MDN K8s crontasks have been updated to change the process user:group to kuma, add Deadmanssnitch support, and some optimizations to prevent aws s3 sync from timing out. PR 533
Unused MDN S3 buckets have been deleted, with some manual cleanup due to versioning enabled on the buckets. PR 531
