You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@airflow.apache.org by GitBox <gi...@apache.org> on 2022/11/02 21:01:52 UTC

[GitHub] [airflow] calebwoofenden commented on issue #11979: Support helm install --wait for the chart

calebwoofenden commented on issue #11979:
URL: https://github.com/apache/airflow/issues/11979#issuecomment-1301248754

   I am running into the same issue. My solution was to set this value in the helm chart:
   ```
     migrateDatabaseJob:
       useHelmHooks: false
   ```
   This solves the problem raised in this issue since it no longer uses the pre-install hook for the migrate db job.
   
   However, I now see an issue sometimes when running `helm upgrade --install` to patch an existing helm install where it's trying to patch the `Job` resource, which cannot be done since `Job`s are immutable. This causes the helm install to fail. What should be happening is either the `Job` gets a unique ID with each helm installation so that it's not trying to edit an existing `Job`, or the `Job` should be destroyed after it completes. One of the helm hooks that gets added to the job is
   ```
   "helm.sh/hook-delete-policy" "before-hook-creation,hook-succeeded"
   ```
   which I think will handle deleting the `Job` upon completion, but we can't use the helm hooks for the above reasons. Kube does offer a parameter called `.spec.ttlSecondsAfterFinished` that, when specified, will delete the `Job` after the specified number of seconds after completion (see https://kubernetes.io/docs/concepts/workloads/controllers/ttlafterfinished/). This is a pretty good solution, though installs done in a period of time shorter than this value might still run into the same problem.
   
   The solution I propose here is to set a helm pre-install hook on the database deployment as well as the db migrations, and set the hook weight on the database lower than that of the migrations, so the database will be created first, then the migrations will run, then the webserver and etc. pods will come up. Then I think we could safely remove the `wait-for-db-migrations` initContainer on the deployments.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: commits-unsubscribe@airflow.apache.org

For queries about this service, please contact Infrastructure at:
users@infra.apache.org