[
https://issues.apache.org/jira/browse/FLINK-9782?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16886109#comment-16886109
]
Kostas Kloudas commented on FLINK-9782:
---------------------------------------
The workflow of the test seems to be:
1) we launch a job with a finite source
2) we are waiting for the job to finish (because the source finished injecting
datas)
3) we check the output to see if all the expected data is there.
The BucketingSink is a "2-phase commit" sink that buffers data between
checkpoints
and commits them as soon as a checkpoint completes.
The above combination can lead to a race condition between the end-of-stream
event
propagating from sources to sinks and the checkpoint complete notification that
is sent from the checkpoint coordinator to the tasks.
I believe this is the source of instability BUT further investigation is
required.
If this assumption is correct, then we can change the workflow of the test to
be the same
as the {{test_streaming_file_sink.sh}} where we wait for the expected output to
be there
and then we cancel the job and verify the correctness of the output.
> BucketingSink E2E test fails on travis
> --------------------------------------
>
> Key: FLINK-9782
> URL: https://issues.apache.org/jira/browse/FLINK-9782
> Project: Flink
> Issue Type: Bug
> Components: Connectors / FileSystem, Tests
> Affects Versions: 1.5.0, 1.7.0
> Reporter: Chesnay Schepler
> Assignee: Kostas Kloudas
> Priority: Critical
> Labels: test-stability
> Fix For: 1.9.0
>
>
> The {{streaming_bucketing.sh}} end-to-end test fails consistently on the 1.5
> branch.
> https://travis-ci.org/zentol/flink-ci/builds/401708750
--
This message was sent by Atlassian JIRA
(v7.6.14#76016)
