[
https://issues.apache.org/jira/browse/HBASE-2149?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12802957#action_12802957
]
stack commented on HBASE-2149:
------------------------------
Chatting w/ J-D, it was set by me -- I think -- at 0.4 and 0.25 because of
PE-itis, a condition that comes of loading only using the PE tool. Symptoms
are inability at being able to comprehend any other type of loading pattern.
This issue is prompted by a fella loading hbase with UUID keys would likely
spread writes nice and evenly across the cluster requiring the flushing of many
regions to go from 0.4 to 0.25. 0.4/.38? We could try it. Or 0.4 and 0.35?
> hbase.regionserver.global.memstore.lowerLimit is too low
> --------------------------------------------------------
>
> Key: HBASE-2149
> URL: https://issues.apache.org/jira/browse/HBASE-2149
> Project: Hadoop HBase
> Issue Type: Bug
> Reporter: Jean-Daniel Cryans
> Assignee: Jean-Daniel Cryans
> Fix For: 0.20.4, 0.21.0
>
>
> The default value of hbase.regionserver.global.memstore.lowerLimit of 25% is
> very wrong and in almost all cases was problematic (I've seen this in at
> least 3 occurrences). The cost of flushing a memstore is fairly high and when
> the global size reaches 40% then ALL inserts are blocked. This means that
> with a heap of 1GB you could be flushing for 10-20 seconds or worse.
> I suggest a default setting of 38% or even 40% so that only a region or two
> will be flushed (the biggest ones) for maximum availability.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
