[
https://issues.apache.org/jira/browse/HBASE-15091?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15098472#comment-15098472
]
stack commented on HBASE-15091:
-------------------------------
bq. In fast path, we use this.mvcc.completeAndWait under row lock, it will wait
to catch up write point, right?
Yes. This is actually wrong though. In the master branch patch, it should be
completeAndWait but here I have mistranslated this.mvcc.advanceMemstore... it
should have become complete only. Thanks for fingering this ('complete' tells
mvcc that this write can be marked as done, complete, whereas 'completeAndWait'
says mark the mvcc transaction as done AND wait until the read point catches up
to the write I just finished).
On your second comment, there is no functional change. We called
completeAndWait previous. Now I call completeAndWait or just complete if an
exception: i.e. if walKey != null (it will be null if processing got to the end
of the method without interrupt -- this is the pattern we use in fast path and
in the coming master patch).
Let me fix the above. [~chenheng] Thank you for the review.
> Forward-port HBASE-15031 "Fix merge of MVCC and SequenceID performance
> regression in branch-1.0"
> ------------------------------------------------------------------------------------------------
>
> Key: HBASE-15091
> URL: https://issues.apache.org/jira/browse/HBASE-15091
> Project: HBase
> Issue Type: Sub-task
> Components: Performance
> Reporter: stack
> Priority: Blocker
> Fix For: 1.2.0
>
> Attachments: 15091v2.branch-1.2.patch, HBASE-15091-branch-1.2.patch,
> HBASE-15091-branch-1.2_v1.patch, HBASE-15091.v1.branch-1.2.patch
>
>
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
