[
https://issues.apache.org/jira/browse/HDFS-10701?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16162792#comment-16162792
]
SammiChen commented on HDFS-10701:
----------------------------------
I have noticed 2 frequently happened {{testBlockTokenExpired}} failure cases.
1.expected:<1001> but was:<1002>
2.java.lang.AssertionError: If an internal block is smaller than parity block,
then its last cell should be small than last parity cell
(1) it expects there is no block timestamp bump when no Datanode failure.
But actuallly there is write streamer failure due to token expiration even
there is no Datanode failure. It's caused by the cell size change, from 64K to
1M, while the block token life time is still 1s.
(2) when there is more than parity block streamer failure, assertion will be
triggered.
It's also because the token expiration period is too short, so there is more
than expected write streamer failures.
There is maybe other failure scenarios not covered here.
Upload a patch to address case (1) and (2)
> TestDFSStripedOutputStreamWithFailure#testBlockTokenExpired occasionally fails
> ------------------------------------------------------------------------------
>
> Key: HDFS-10701
> URL: https://issues.apache.org/jira/browse/HDFS-10701
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: erasure-coding
> Affects Versions: 3.0.0-alpha2
> Reporter: Wei-Chiu Chuang
>
> I noticed this test failure in a recent precommit build, and I also found
> this test had failed for a few times in Hadoop-Hdfs-trunk build in the past.
> But I do not have sufficient knowledge to tell if it's a flaky test or a bug
> in the code.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
