Hello Everyone,

I thought it's a good time to share some news after yesterday's ASF
"Builds meeting". The latest news for our CI is that ~ 2 weeks ago
GitHub bumped the parallelism of jobs for Public runners for Apache
Software Foundation from 300 to 900.

Attached is the graph of queue size for Apache projects where you can
see how dramatic improvements it was (other projects of Apache do not
have self-hosted runners as we do so they are far bigger users).
Thanks to Tobiasz Kedzierski - one of the contributors to Airflow and
my friend who developed and maintains the graphs so that ASF can see
some stats.

This means that builds from forks of our contributors should
experience far, far less queuing. Some new things are coming as well -
we (ASF with our support) are talking to GitHub to get self-hosted
runners on Azure, as well as enabling bigger instances / ARM instances
for running our CI jobs.

This will likely allow us to optimize some of the build times as well.
I already have some ideas how we can make our builds leaner, faster
and with less number of failures. The goal is always to get the CI
feedback as fast as possible with as little as possible
false-negatives (and without burning too much money that our sponsors
give us). It's a moving target but we have some good possibilities now
on how to make those better.

Stay tuned.

Also I think this is the right moment to thank both Astronomer and
Amazon for the AWS credits (Amazon) and money (Astronomer). Again this
year we got USD 15.000 from Amazon (last year we got USD 10.000 and
Astronomer pays whatever we exceed above that and invested quite some
time (mostly of Ash) to get our "VM" infrastructure up and running.

There are also likely more news coming in the light of AIP-47 - where
we are working with Amazon, Google, Databricks, Snowflake (and others)
to make our "System" test infrastructure even stronger and be able to
automatically test the stability and regressions of integrations of
Airflow.

J.

Reply via email to