[
https://issues.apache.org/jira/browse/HDFS-1142?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12869424#action_12869424
]
Konstantin Shvachko commented on HDFS-1142:
-------------------------------------------
Todd,
Thanks for elaborating on the HBase / ZK interaction. Sounds like you are
trying to build HA there.
> Since this is a cluster-wide setting I don't think it's a good idea.
In my previous comment I explained what the soft limit means - the original
client is able to reclaim its lease by simply renewing it. Expiration of the
hard limit means that the original client does not own the file anymore, and
can claim the ownership only via reopening the file, same as everybody else.
After soft limit expires hdfs can either reassign it to a new client or keep
under the old ownership. Reassigning it to NN during lease recovery contradicts
the definition of the soft limit. The revoking ownership behavior you are
describing corresponds to expiration of the hard limit.
I do not understand how (b) file access revocation can be solved with lease
recovery, even with your modified semantics. The region server may wake up
before soft limit expiration but after the region reassignment.
What really needed is to first revoke file access, then start reassigning
regions. Revoking file access is achieved by just opening it with another
client. AFAU, regions reassignment should wait until the access is revoked.
> Since this is a cluster-wide setting I don't think it's a good idea. We only
> want this kind of lease-stealing behavior for hlog recovery
But you are talking about clusterS-wide change of semantics only for those hlog
recoveries.
I will be glad to continue discussing HBase problems. But "Lease recovery
doesn't reassign lease when triggered by append()" is not a bug and is not a
problem from HDFS point of view.
I propose to close this issue as "Not a Problem" (The described issue is not
actually a problem - it is as designed.)
> Lease recovery doesn't reassign lease when triggered by append()
> ----------------------------------------------------------------
>
> Key: HDFS-1142
> URL: https://issues.apache.org/jira/browse/HDFS-1142
> Project: Hadoop HDFS
> Issue Type: Bug
> Components: name-node
> Affects Versions: 0.21.0
> Reporter: Todd Lipcon
> Assignee: Todd Lipcon
> Attachments: hdfs-1142.txt, hdfs-1142.txt
>
>
> If a soft lease has expired and another writer calls append(), it triggers
> lease recovery but doesn't reassign the lease to a new owner. Therefore, the
> old writer can continue to allocate new blocks, try to steal back the lease,
> etc. This is for the testRecoveryOnBlockBoundary case of HDFS-1139
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
