[
https://issues.apache.org/jira/browse/HBASE-24578?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Reid Chan updated HBASE-24578:
------------------------------
Release Note: Introduce a new parameter
"hbase.regionserver.wal.sync.batch.count" to control the wal sync batch size
which is equals to "hbase.regionserver.handler.count" by default. It should
work well if you use default wal provider---one wal per regionserver. But if
you use read/write separated handlers, you can set
"hbase.regionserver.wal.sync.batch.count" to the number of write handlers. And
if you use wal-per-groups or wal-per-region, you can consider lower
"hbase.regionserver.wal.sync.batch.count", the default number will be too big
and consume more memories as the number of wals grows. (was: Introduce a new
parameter "hbase.regionserver.wal.sync.batch.count" to control the wal sync
batch size which is equals to "hbase.regionserver.handler.count" by default. It
should work well if you use default wal provider---one wal per regionserver.
But if you use Read/Write separated handlers, you can set
"hbase.regionserver.wal.sync.batch.count" to the number of write handlers. And
if you use wal-per-groups or wal-per-region, you can consider lower
"hbase.regionserver.wal.sync.batch.count", the default number will be too big
and consume more memories as the number of wals grows.)
> [WAL] Add a parameter to config RingBufferEventHandler's SyncFuture count
> -------------------------------------------------------------------------
>
> Key: HBASE-24578
> URL: https://issues.apache.org/jira/browse/HBASE-24578
> Project: HBase
> Issue Type: Improvement
> Components: wal
> Affects Versions: 1.4.13, 2.2.5
> Reporter: Reid Chan
> Assignee: wenfeiyi666
> Priority: Major
> Fix For: 3.0.0-alpha-1, 2.3.1, 1.7.0, 2.2.6
>
>
> The current value of RingBufferEventHandler's handler is the value of
> {{hbase.regionserver.handler.count}}, which works good in default wal
> provider --- one WAL per regionserver.
> When trying to use WAL group provider, either by group or wal per region, the
> default value is bad. If rs has 100 regions and wal per region strategy is
> used, then rs will allocate 100 *
> SyncFuture[$hbase.regionserver.handler.count] array
> {code}
> int maxHandlersCount = conf.getInt(HConstants.REGION_SERVER_HANDLER_COUNT,
> 200);
> this.ringBufferEventHandler = new RingBufferEventHandler(
> conf.getInt("hbase.regionserver.hlog.syncer.count", 5),
> maxHandlersCount);
> ...
> ----
> RingBufferEventHandler(final int syncRunnerCount, final int maxHandlersCount)
> {
> this.syncFutures = new SyncFuture[maxHandlersCount];
> ...
> }
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
