[
https://issues.apache.org/jira/browse/FLINK-25765?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17480941#comment-17480941
]
Niklas Semmler commented on FLINK-25765:
----------------------------------------
Okay, so then we basically have to add documentation along the lines of:
The configmap *does not* represent the complete configuration that is applied
to the TaskManager/JobManager. Additional configuration settings are/can be set
through environment variables and command-line arguments. Changes to the
configmap *will not affect* currently running pods, but will affect pods that
restart.
For the JobManager at least we can learn the configuration from the web
interface. It is to bad, that the TaskManager doesn't have this option. Maybe,
it would be a good addition. It seems to me there should be a single place for
the ground truth configuration.
> Kubernetes: flink's configmap and flink's actual config are out of sync
> -----------------------------------------------------------------------
>
> Key: FLINK-25765
> URL: https://issues.apache.org/jira/browse/FLINK-25765
> Project: Flink
> Issue Type: Bug
> Components: Deployment / Kubernetes
> Affects Versions: 1.15.0
> Reporter: Niklas Semmler
> Priority: Major
> Labels: usability
>
> For kubernetes setups, Flink's configmap does not reflect the actual config.
> Causes
> # Config values are overridden by the environment variables in the docker
> image (see FLINK-25764)
> # Flink reads the config on start-up, but does not subscribe to changes
> # Changes to the config map do not lead to restarts of the flink cluster
> Effects
> # Users cannot expect to understand Flink's config from the configmap
> # TaskManager/JobManager started at different times may start with different
> configs, if the user edits the configmap
> Related to FLINK-21383.
--
This message was sent by Atlassian Jira
(v8.20.1#820001)
