[
https://issues.apache.org/jira/browse/BEAM-11612?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17312598#comment-17312598
]
Kyle Weaver commented on BEAM-11612:
------------------------------------
[auto] can go two ways - it can either infer an environment from the context,
or it can create a new local environment from scratch.
# For the local case, I fixed that in
[https://github.com/apache/beam/pull/13722.]
# If it can infer an environment from context, that means the user already has
Flink set up somehow, and the Flink deployment will have its own Flink
configuration, which would usually be the same the one the user passed to Beam.
Beam only uses the Flink configuration for a few fields, and the configuration
passed to Beam was never expected to override the configuration of an existing
Flink cluster. So this is WAI.
[collection] I don't think collection environment is used much, and probably
never in prod (it's intended mostly for debugging), so I think p3 is
appropriate.
> [auto] and [collection] do not respect flink-conf.yaml.
> -------------------------------------------------------
>
> Key: BEAM-11612
> URL: https://issues.apache.org/jira/browse/BEAM-11612
> Project: Beam
> Issue Type: Bug
> Components: runner-flink
> Reporter: Kyle Weaver
> Priority: P3
>
> When the Flink master is set to [auto] and [collection], flink-conf.yaml
> (from the FLINK_CONF_DIR environment variable, or the --flink-conf-dir option
> of the job server) is disregarded.
> This can result in some confusing behavior. For example, when [auto] creates
> a local environment, an empty Configuration() object is used.
> https://github.com/apache/beam/blob/dd71c9ac472b62ff3598d3db47b0f26985bbc82a/runners/flink/src/main/java/org/apache/beam/runners/flink/FlinkExecutionEnvironments.java#L82-L85
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
