[
https://issues.apache.org/jira/browse/KUDU-2151?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16175673#comment-16175673
]
Todd Lipcon commented on KUDU-2151:
-----------------------------------
I think more generally it's all the open() and close() syscalls in env_posix.cc
which are missing RETRY_ON_EINTR. I managed to trigger this when I'd mounted
the consensus metadata dir on an sshfs mount, which I guess makes these
syscalls interruptible whereas perhaps in more normal environments EINTR is not
likely on them.
> ConsensusMetadata IO does not retry on EINTR
> --------------------------------------------
>
> Key: KUDU-2151
> URL: https://issues.apache.org/jira/browse/KUDU-2151
> Project: Kudu
> Issue Type: Improvement
> Components: consensus
> Affects Versions: 1.5.0
> Reporter: Mike Percy
>
> We saw the following errors when doing ConsensusMetadata IO on a volume
> mounted on sshfs:
> Read:
> {code}
> F0920 16:38:06.913570 16423 tablet_server_main.cc:80] Check failed: _s.ok()
> Bad status: IO error: Could not init Tablet Manager: Unable to load consensus
> metadata for tablet 4b2100922a8c4668baadc2fde100d8cf:
> /data0/ts-data/consensus-meta/4b2100922a8c4668baadc2fde100d8cf: Interrupted
> system call (error 4)
> {code}
> Write:
> {code}
> F0920 16:41:00.588088 5444 raft_consensus.cc:2604] Check failed: _s.ok() Bad
> status: IO error: Unable to write consensus meta file for tablet
> 8b7b2338ddec4690859830ef34713732 to path
> /data0/ts-data/consensus-meta/8b7b2338ddec4690859830ef34713732: Call to
> mkstemp() failed on name template
> /data0/ts-data/consensus-meta/8b7b2338ddec4690859830ef34713732.kudutmp.XXXXXX:
> Interrupted system call (error 4)
> {code}
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
