[
https://issues.apache.org/jira/browse/FLINK-33598?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Yun Tang resolved FLINK-33598.
------------------------------
Resolution: Fixed
Merged
master: 608546e090f5d41c6a8b9af2c264467279181027 ...
b7e8b792c086c3c445ee8429fbcfe035097a878c
release-1.18: 6f30c6e427251dd4b2e4ad03f89bed06a519b05f
release-1.17: 18d5a4696eccac3b5e7fe1d579547feef4537c08
> Watch HA configmap via name instead of lables to reduce pressure on APIserver
> ------------------------------------------------------------------------------
>
> Key: FLINK-33598
> URL: https://issues.apache.org/jira/browse/FLINK-33598
> Project: Flink
> Issue Type: Improvement
> Components: Deployment / Kubernetes
> Affects Versions: 1.18.0, 1.17.1
> Reporter: Yun Tang
> Assignee: Yun Tang
> Priority: Critical
> Labels: pull-request-available
> Fix For: 1.19.0, 1.18.1, 1.17.3
>
>
> As FLINK-24819 described, the k8s API server would receive more pressure when
> HA is enabled, due to the configmap watching being achieved via filter with
> labels instead of just querying the configmap name. This could be done after
> FLINK-24038, which reduced the number of configmaps to only one as
> {{<clusterId>-cluster-config-map}}.
> This ticket would not touch {{<clusterId>-<jobId>-config-map}}, which stores
> the checkpoint information, as that configmap is directly accessed by JM and
> not watched by taskmanagers.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
