[
https://issues.apache.org/jira/browse/HBASE-16698?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15517721#comment-15517721
]
Yu Li commented on HBASE-16698:
-------------------------------
bq. So, why is this patch faster? In current implementation, contention is
farmed out to be per WALKey instance. Each has its own latch.
Yes, each WALKey has its own latch, but the contention is not on the latch
itself, but the sequential handling of ringbuffer event. The whole process is
like:
{noformat}
RingBufferEventHandler grab one append
-> FSHLog#append is called
-> FSWALEntry#stampRegionSequenceId is called
-> One CountDownLatch is released
-> RingBufferEventHandler grab another append
-> Another CountDownLatch is released
-> Repeat
{noformat}
So all CountDownLatch are released in sequential, no parallelism...
bq. I was thinking there a correctness issue but the numbering/mvcc is scoped
to the region so if you lock across the region append while getting the mvcc,
and this is only place mvcc is incremented, then all should be good
Yes, agree. And it seems our mighty [~eclark] has the same concern here. Hope
this answers your question also [~eclark] :-)
bq. Pity we have to lock. Could we be more radical and use the ringbuffer
bucket number? Then no locking needed. The change would be way more intrusive
though. You'd have to change a lot
Cannot agree more... Actually I ever tried to use multiple event handlers, but
too much logic to make sure if breaking sequential append, so I finally quit...
But I agree that we should revisit this sometime later, worth the efforts I
think.
> Performance issue: handlers stuck waiting for CountDownLatch inside
> WALKey#getWriteEntry under high writing workload
> --------------------------------------------------------------------------------------------------------------------
>
> Key: HBASE-16698
> URL: https://issues.apache.org/jira/browse/HBASE-16698
> Project: HBase
> Issue Type: Improvement
> Components: Performance
> Affects Versions: 1.1.6, 1.2.3
> Reporter: Yu Li
> Assignee: Yu Li
> Attachments: HBASE-16698.patch, hadoop0495.et2.jstack
>
>
> As titled, on our production environment we observed 98 out of 128 handlers
> get stuck waiting for the CountDownLatch {{seqNumAssignedLatch}} inside
> {{WALKey#getWriteEntry}} under a high writing workload.
> After digging into the problem, we found that the problem is mainly caused by
> advancing mvcc in the append logic. Below is some detailed analysis:
> Under current branch-1 code logic, all batch puts will call
> {{WALKey#getWriteEntry}} after appending edit to WAL, and
> {{seqNumAssignedLatch}} is only released when the relative append call is
> handled by RingBufferEventHandler (see {{FSWALEntry#stampRegionSequenceId}}).
> Because currently we're using a single event handler for the ringbuffer, the
> append calls are handled one by one (actually lot's of our current logic
> depending on this sequential dealing logic), and this becomes a bottleneck
> under high writing workload.
> The worst part is that by default we only use one WAL per RS, so appends on
> all regions are dealt with in sequential, which causes contention among
> different regions...
> To fix this, we could also take use of the "sequential appends" mechanism,
> that we could grab the WriteEntry before publishing append onto ringbuffer
> and use it as sequence id, only that we need to add a lock to make "grab
> WriteEntry" and "append edit" a transaction. This will still cause contention
> inside a region but could avoid contention between different regions. This
> solution is already verified in our online environment and proved to be
> effective.
> Notice that for master (2.0) branch since we already change the write
> pipeline to sync before writing memstore (HBASE-15158), this issue only
> exists for the ASYNC_WAL writes scenario.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
