Is there a node Waiting or Preparing phase when local-home data is different in confluence cluster?

[KUNG MIN PARK]

If files and directories other than confluence.cfg.xml, which is the core file of local-home, are different from other nodes (or if they are previous files), booting seems to be possible, but it is certain to request an index snapshot that exists in shared-home and take it. Shouldn't we disable the node while it's coming?

This seems to be an important part to apply to replicaCount of kubernetes deployment and Cloud (example AWS) Autoscaling service.

[Yevhen · Accepted solution]

@KUNG MIN PARK while the none is recovering index from shared home, it is responding 503 to /status calls from a readiness probe. Until readiness probe passes, the pod isn't marked as ready, and thus no endpoint is created. In other words, traffic does not go from service to such a pod.

[KUNG MIN PARK]

Thank you for quick response.

Is there an example to check the processing speed per GB that I can refer to when the index size is very large?

[Yevhen]

Unfortunately, there are no estimates. It also depends on  available resources. What Confluence does is uncompresses index snapshots, so it's cpu and io intensive operation. In my most recent tests, 26Gi of indexes were recovered in 5 to 6 minutes (gp2 storageclass)

[KUNG MIN PARK]

Thank you I'll refer to it

I forgot to check the health of cofluence and jira installed through the helm chart through the probe!

I can suggest a solution that applies health check to the load balancer.

Is Jira the same?

[Yevhen]

Yes, Jira pod won't be marked ready when it's recovering cache, so /status will report 503 or sometimes even 500

[KUNG MIN PARK]

Thank you!