[GitHub] [airflow]: Workflow run "Build images for Fix running parallel k8s tests in "depth-first" mode in CI https://github.com/apache/airflow/pull/26904 " is working again!

[GitBox <gi...@apache.org>]

The GitHub Actions job "Build images for Fix running parallel k8s tests in "depth-first" mode in CI https://github.com/apache/airflow/pull/26904
" on airflow.git has succeeded.
Run started by GitHub user potiuk (triggered by potiuk).

Head commit for run:
0665e6ef65594bdd332fcff72d638c9096df8e34 / Jarek Potiuk <ja...@potiuk.com>
Fix running parallel k8s tests in "depth-first" mode in CI

The K8S tests when run in parallel for multiple versions require:

a) vast amount of memory to keep all the cluster running
b) extra time to startup for all k8s tests

This caused a problem when tests in CI were split into separate
steps - because we had to create all clusters before we run tests
even if we limited parallelism.

By using "depth-first" mode, we can run complete tests (with
cluster creation and deletion) in parallel with limited parallelism
without having to create all clusters upfront.

We now add the `run-complete-tests` command that does all the
steps in one function - which makes it much nicer to parallelise
without using too much resources.

There had to be a few fixes done to make it happen.

* Failure in CI for K8S jobs produced really long logs that
  were next to impossible to analyse any problems (for example
  problems involved with lack of memory or with overwhelming K8S
  with too many requests when there were multiple clusters.
  When we combined multiple steps in one output it made the logs
  absolutely unusable.

* Verbose moda has been disabled because it was creating too much
  noise (not that bad as resource dumps, but still lots of it).

* The resource dumps from K8S are now dumped to separate files.
  Those resource dumps can be downloaded separately as they are uploaded
  as artifacts in case of failure.

There were also few fixes to kubernetes tests to make them more stable
and debuggable:

* Assertion which was thrown in "setUp" of K8S test base did not give
  too much information about what the problem was so it has been now
  replaced with a much more user-friendly message showing the actual
  reason (could not check health of the webserver) and
  you can see the root cause alongside the error.

* Few tests in itest_kubernetes_pod_poperator were using fixed path for
  local temp file that caused intermittent failures of the tes when run
  in parallell.

Report URL: https://github.com/apache/airflow/actions/runs/3216960588

With regards,
GitHub Actions via GitBox


---------------------------------------------------------------------
To unsubscribe, e-mail: jobs-unsubscribe@airflow.apache.org
For additional commands, e-mail: jobs-help@airflow.apache.org