[
https://issues.apache.org/jira/browse/HBASE-8394?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13638745#comment-13638745
]
Ted Yu edited comment on HBASE-8394 at 4/23/13 4:19 AM:
--------------------------------------------------------
Here are some related comments from HBASE-8389:
https://issues.apache.org/jira/browse/HBASE-8389?focusedCommentId=13638534&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13638534
bq. You are right that client calls recoverLease to Namenode and then Namenode
will choose a live datanode as the primary for recovery. However, if the
primary datanode indeed is dead, a second lease recovery won't be started until
the lease expires. This will take a long time unless we have HDFS-4721.
https://issues.apache.org/jira/browse/HBASE-8389?focusedCommentId=13638543&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13638543
bq. I forgot to say that the second lease recovery will be started
automatically by the namenode again but it has to wait for
LEASE_HARDLIMIT_PERIOD, which is one hour by default. If HBase wants to speed
up the second recovery, it may wait and check file close for some reasonable
time period, say 1 minute, and then call recoverLease again.
https://issues.apache.org/jira/browse/HBASE-8389?focusedCommentId=13638689&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13638689
was (Author: [email protected]):
Here are some related comments from HBASE-8389:
https://issues.apache.org/jira/browse/HBASE-8389?focusedCommentId=13638534&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13638534
https://issues.apache.org/jira/browse/HBASE-8389?focusedCommentId=13638543&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13638543
https://issues.apache.org/jira/browse/HBASE-8389?focusedCommentId=13638689&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-13638689
> Utilize isFileClosed() so that the wait for lease recovery can be optimized
> ---------------------------------------------------------------------------
>
> Key: HBASE-8394
> URL: https://issues.apache.org/jira/browse/HBASE-8394
> Project: HBase
> Issue Type: Improvement
> Reporter: Ted Yu
>
> In HBASE-8389, there was discussion for utilizing isFileClosed() to optimize
> the wait time for lease recovery.
> Subsequent lease recovery request for the same Path would preempt outstanding
> recovery. But if we wait for one particular recovery without limit on
> duration, we may incur penalty due to various failure scenarios such as stale
> primary Data Node, etc.
> This JIRA continues the work for 0.95 / trunk.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators
For more information on JIRA, see: http://www.atlassian.com/software/jira
