[
https://issues.apache.org/jira/browse/HBASE-17179?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15701093#comment-15701093
]
Anoop Sam John commented on HBASE-17179:
----------------------------------------
How many regions in ur RS?
One thing to note - Each of the region work with one MSLAB object and that ask
for pooled chunks (each of size 2 MB). There is no sharing btw MSLABs in
regions. Means if there is one region with very less data size it it, we still
occupy the 2 MB chunk size fully. But that has to get flushed any way at some
point in time.. (We dont keep data for longer time even if flush size not
met).. Just said FYI.
But this way of 3 GB - 10 GB seems not reasonable. Will see branch-1 code.
> the chunkpool are not reclamed correctly
> ----------------------------------------
>
> Key: HBASE-17179
> URL: https://issues.apache.org/jira/browse/HBASE-17179
> Project: HBase
> Issue Type: Bug
> Affects Versions: 1.0.1.1, 1.1.2
> Reporter: chenxu
>
> in our cluster, the RS's config is as below
> {quote}
> heap size(25G)
> hbase.regionserver.global.memstore.size(0.5)
> hbase.hregion.memstore.chunkpool.maxsize(0.8)
> hbase.hregion.memstore.chunkpool.initialsize(1.0)
> {quote}
> so the chunkpool's size is 25*0.5*0.8 = 10GB
> in our monitor system, the _memStoreSize_ never up to 3G，so the _reuseRatio_
> of the chunkpool should always 100%, but it's not, see the logs below:
> {quote}
> 2016-11-23 11:47:34,302 DEBUG
> [StoreOpener-33a77b971aff15e5e6be3cc614870e43-1-MemStoreChunkPool Statistics]
> regionserver.MemStoreChunkPool: Stats: current pool size=5009,created chunk
> count=0,reused chunk count=81,reuseRatio=100.00%
> 2016-11-28 11:47:34,302 DEBUG
> [StoreOpener-33a77b971aff15e5e6be3cc614870e43-1-MemStoreChunkPool Statistics]
> regionserver.MemStoreChunkPool: Stats: current pool size=3,created chunk
> count=1325,reused chunk count=67209,reuseRatio=98.07%
> {quote}
> 5 days later, the _reuseRatio_ is below 100%
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
