This post is a placeholder for a more in-depth tutorial about ArgoCD, Sentry, and Helm post-install hooks. I am publishing it early to capture the workaround, and to permit closing of a bug report I filed upstream in the adfinis-sygroup/helm-charts community-provided Helm chart repository (on a self-imposed deadline.)

The issue is explained in greater depth at the report linked above, but in summary: there is a subtle difference in the way that ArgoCD manages Helm’s “Chart Hooks”, specifically a post-install hook. As a result, there is a chicken-and-egg problem when installing Sentry On-Premise this way.

Helm’s post-install hook (when activated by Helm itself) starts executing “after all resources are loaded into Kubernetes.” ArgoCD’s Helm support does not use Helm itself, does not generate Helm release configmaps or leave a release footprint in the form of a Helm release; instead it basically runs helm template and captures the manifest yaml output, which it then syncs to the cluster at sync time. Argo maps its own pre/post sync hooks to the Helm hooks, which is unfortunately not a clean mapping, in this case PostSync waits until all resources are healthy before executing, which will never happen as the deploy/sentry-relay artifact cannot become healthy until post-install hooks executed, running (among other things) postgres database initialization and migrations.

Unfortunately this is unlikely to be fixed in ArgoCD, and maintainers recommend Helm Operator when perfect helm installations are needed.

Fortunately this is easy to work around! (Placeholder: Screencast is forthcoming)

In tl;dw: switch your ArgoCD Application UI’s filter to show resources with a Health status of Progressing or Degraded only. Sync the ArgoCD Application resource that was created via:

helm repo add adfinis https://charts.adfinis.com
helm upgrade --install sentry-apps adfinis/sentry-apps --values values.yaml

When sentry-relay is started but not ready, this prevents ArgoCD’s PostSync hook from executing. Three other deployments (sessions-consumer, snuba-outcomes-consumer, and snuba-replacer) also don’t become healthy, but rather than using Readiness check, they terminate (and will land in CrashLoopBackOff status if this continues long enough.) This all means that sentry-relay’s failure prevents PostSync, but the others do not.

So, with the filter engaged, wait until everything is healthy except sentry-relay is not ready, and the three others mentioned are starting, but also stuck in Terminating loop. You can trick ArgoCD into proceeding easily at that point by scaling down sentry-relay for a brief moment.

kubectl -n sentry scale --replicas=0 sentry-relay
# ... wait for Jobs to start firing
kubectl -n sentry scale --replicas=1 sentry-relay

If it worked, you can tell PostSync has begun because Jobs start firing (and hopefully eventually completing.) When the sentry-relay deploy has been scaled back up and the post-install Jobs have all been completed, you should find ArgoCD reporting Sync succeeded and everything is Healthy.

I hope this helps someone! It took me a whole afternoon to figure this out myself. 😅