[
https://issues.apache.org/jira/browse/HDFS-15220?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17058428#comment-17058428
]
Ayush Saxena commented on HDFS-15220:
-------------------------------------
FSCK is used to fetch replica details and even conclude whether the state of
the filesystem is healthy or not. There are couple of reasons where Observer
may not be having the correct replica state.
The observer is like standby only, it won't process certain IBR's immediately,
it will enqueue them for processing later({{Check PendingDataNodeMessages}}),
So corrupt replica count won't be correct.
There would be multiple Observers, One is having stale storages, Others are not
having, So result on different Observers would be different. I tackled bunch of
issues regarding this, HDFS-15187 & HDFS-15200.
The Actual Filesystem state can only be judged by the state at the active,
since the block deletion call is also taken by Active NN, That is replica
getting marked as (EXCESS), That ONN won't be knowing
And the most prominent reason as Surendra already Mentioned, FSCK being a REST
call, you can't do msync too
> FSCK calls are redirecting to Active NN
> ---------------------------------------
>
> Key: HDFS-15220
> URL: https://issues.apache.org/jira/browse/HDFS-15220
> Project: Hadoop HDFS
> Issue Type: Bug
> Reporter: krishna reddy
> Assignee: Ravuri Sushma sree
> Priority: Major
> Attachments: screenshot-1.png
>
>
> Run any fsck except (-delete & - move) should go to ONN as it is read
> operation
> In below image spikes indicates when it ran fsck / -storagepolicies
> !screenshot-1.png!
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
