Redis with sentinel does not announce hostname because of regression #32573
Labels
in-progress
redis
stale
15 days without activity
tech-issues
The user has a technical issue about an application
Name and Version
bitnami/redis 20.11.4
What architecture are you using?
amd64
What steps will reproduce the bug?
bitnami/redis
version 20.11.0+ to a dual stack (IPv4+IPv6) cluster, withsentinel
enabled and the new external access service introduced in [bitnami/redis] feat: Add external access service for redis sentinel #32190 disabled (the default behaviour).See from the logs that sentinel pods are flapping their IPs rapidly between the IPv4 and IPv6 addresses of the pod.
Are you using any custom parameters or values?
No custom values are needed to reproduce the issue.
What is the expected behavior?
The generated sentinel configuration should set
sentinel announce-ip
with the hostname of the pod, routed via the headless service.What do you see instead?
bitnami/redis
chart versions lower than 20.11.0 setsentinel announce-ip
to the correct hostname as part of thestart-sentinel.sh
script. However, starting 20.11.0, introduced with PR #32190, this configuration is no longer set because it was wrapped inside the if-statement of the external service feature. Therefore, sentinels launched with chart version 20.11.0+ will announce their IPs to other sentinels.In clusters with a dual stack network configuration (IPv4+IPv6), the sentinals will start flapping between their IPv4 and IPv6 addresses, causing log spam. In my cluster, this also causes random restarts because the flapping is somehow slowing down readiness and liveness probes. This might be because of resource CPU/memory limitations.
Additional information
It seems to me that this configuration change is a regression of #32190. Even with the external cluster service disabled,
sentinel announce-ip
should still be set to the correct hostname to prevent described behaviour.The text was updated successfully, but these errors were encountered: