gauravkm commented on code in PR #3854:
URL: https://github.com/apache/celeborn/pull/3854#discussion_r4056238915
##########
charts/celeborn/values.yaml:
##########
@@ -126,6 +131,36 @@ master:
- name: celeborn-ratis
mountPath: /mnt/celeborn_ratis
+ # -- Startup probe for Celeborn master containers. It holds off the liveness
probe while the
+ # master waits for the DNS records of its peers, which can take a while if a
replica is slow to
+ # be scheduled. Set to `null` to disable.
+ startupProbe:
Review Comment:
**[P1] Avoid deadlocking the default master startup**
With the default three replicas and `OrderedReady`, master-0 waits for every
master DNS record before starting HTTP. This startup probe prevents it becoming
Ready until `/healthz` succeeds, even with no readiness probe ([kubelet
behavior](https://github.com/kubernetes/kubernetes/blob/v1.34.0/pkg/kubelet/prober/prober_manager.go#L270-L316)).
Master-1/2 therefore never get created and master-0 keeps restarting.
`publishNotReadyAddresses` cannot publish nonexistent pods, and increasing the
timeout does not resolve the cycle. Please reconcile the bootstrap/probe
defaults while preserving StatefulSet upgrade compatibility, and cover a fresh
installation with the default three replicas.
--
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: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
