[
https://issues.apache.org/jira/browse/HBASE-14460?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15111221#comment-15111221
]
stack commented on HBASE-14460:
-------------------------------
Added release note on approach going on in here. (Unfortunately,) We have three
different changes made under the rubric of this issue. In the branch-1.0 (and
branch-1.1) fix, HBASE-15031, and in the branch-1.2+ fix, HBASE-15091, we
address the Increment slow down only; you set a flag to take a 'fast path' for
Increments only. The patches for branch-1.0+branch-1.1 and that for branch-1.2+
differ pretty substantially mostly because HBASE-12751 changed the theatre of
operation. The fix for master branch is a general fix. It builds on the changes
in branch-1 taking the 'fast path' only for increments but then changes the
write path for all updates in hbase -- appends, checkAnd* AND the ho-hum
batchMutate, doMiniBatchMutate, etc., -- to do similar, making all changes
under row lock, including sync and mvcc catchup, so append and checkAnd* can
take the same 'fast path' and we can continue to honor our consistency-on-a-row
tenet. Doing all under a lock was intentionally avoided in the past for
performance reasons but HBASE-12751 (row locks are read/write rather than
xclusive) plus all that has changed in the write path since changes the lay of
the land; HBASE-15046 seems to show the default path runs a little faster
though all now done under lock (more perf testing to follow).
> [Perf Regression] Merge of MVCC and SequenceId (HBASE-8763) slowed
> Increments, CheckAndPuts, batch operations
> -------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-14460
> URL: https://issues.apache.org/jira/browse/HBASE-14460
> Project: HBase
> Issue Type: Bug
> Components: Performance
> Reporter: stack
> Assignee: stack
> Priority: Critical
> Attachments: 0.94.test.patch, 0.98.test.patch,
> 1.0.80.flamegraph-7932.svg, 14460.txt, 14460.v0.branch-1.0.patch,
> 98.80.flamegraph-11428.svg, HBASE-14460-discussion.patch, client.test.patch,
> flamegraph-13120.svg.master.singlecell.svg, flamegraph-26636.094.100.svg,
> flamegraph-28066.098.singlecell.svg, flamegraph-28767.098.100.svg,
> flamegraph-31647.master.100.svg, flamegraph-9466.094.singlecell.svg,
> hack.flamegraph-16593.svg, hack.uncommitted.patch, m.test.patch,
> region_lock.png, testincrement.094.patch, testincrement.098.patch,
> testincrement.master.patch
>
>
> As reported by 鈴木俊裕 up on the mailing list -- see "Performance degradation
> between CDH5.3.1(HBase0.98.6) and CDH5.4.5(HBase1.0.0)" -- our unification of
> sequenceid and MVCC slows Increments (and other ops) as the mvcc needs to
> 'catch up' to our current point before we can read the last Increment value
> that we need to update.
> We can say that our Increment is just done wrong, we should just be writing
> Increments and summing on read, but checkAndPut as well as batching
> operations have the same issue. Fix.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
