[
https://issues.apache.org/jira/browse/FLINK-13500?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16908915#comment-16908915
]
Chesnay Schepler commented on FLINK-13500:
------------------------------------------
Currently this is expected behavior. This is because the high-availability
logic we use in the client is the same we use on the server, and thus has the
same capabilities and requirements (i.e., being able to access the blobstore,
and thus requiring access).
I've opened FLINK-13518 to separate client/server high-availability concerns,
which should remedy your problem.
> RestClusterClient requires S3 access when HA is configured
> ----------------------------------------------------------
>
> Key: FLINK-13500
> URL: https://issues.apache.org/jira/browse/FLINK-13500
> Project: Flink
> Issue Type: Bug
> Components: Runtime / Coordination, Runtime / REST
> Affects Versions: 1.8.1
> Reporter: David Judd
> Priority: Major
>
> RestClusterClient initialization calls ClusterClient initialization, which
> calls
> org.apache.flink.runtime.highavailability.HighAvailabilityServicesUtils.createHighAvailabilityServices
> In turn, createHighAvailabilityServices calls
> BlobUtils.createBlobStoreFromConfig, which in our case tries to talk to S3.
> It seems very surprising to me that (a) RestClusterClient needs any form of
> access other than to the REST API, and (b) that client initialization would
> attempt a write as a side effect. I do not see either of these surprising
> facts described in the documentation–are they intentional?
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
