[
https://issues.apache.org/jira/browse/HDFS-10913?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15706142#comment-15706142
]
Arpit Agarwal commented on HDFS-10913:
--------------------------------------
Thanks for this testability improvement [~xiaobingo]. Few comments:
# delayWritingPacketToMirror looks unnecessary as we already have
stopSendingPacketDownstream.
# Can you please add a hook for delayWritingChecksumToDisk also here?
{code}
checksumOut.write(buf);
{code}
# delaySyncingWriterOsCache should not introduce delay when
syncBehindWritesInBackground is enabled. The delay should be in the else clause
only in manageWriterOsCache:
{code}
if (syncBehindWritesInBackground) {
...
} else {
NativeIO.POSIX.syncFileRangeIfPossible(outFd,
...
}
{code}
Also you may have to rebase the patch once HDFS-10930 is committed. Still
reviewing the unit test.
> Introduce fault injectors to simulate slow mirrors
> --------------------------------------------------
>
> Key: HDFS-10913
> URL: https://issues.apache.org/jira/browse/HDFS-10913
> Project: Hadoop HDFS
> Issue Type: Improvement
> Components: datanode
> Affects Versions: 2.8.0
> Reporter: Xiaobing Zhou
> Assignee: Xiaobing Zhou
> Attachments: HDFS-10913.000.patch, HDFS-10913.001.patch,
> HDFS-10913.002.patch, HDFS-10913.003.patch, HDFS-10913.004.patch
>
>
> BlockReceiver#datanodeSlowLogThresholdMs is used as threshold to detect slow
> mirrors. BlockReceiver only writes some warning logs. In order to better test
> behaviors of slow mirrors, it necessitates introducing fault injectors.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
