[
https://issues.apache.org/jira/browse/HBASE-11887?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14120473#comment-14120473
]
stack commented on HBASE-11887:
-------------------------------
jmap -histo:
{code}
num #instances #bytes class name
----------------------------------------------
1: 3069883 780472352 [B
2: 3501587 112050784 org.apache.hadoop.hbase.KeyValue
3: 3441167 82588008
java.util.concurrent.ConcurrentSkipListMap$Node
4: 1077906 68985984
org.apache.hadoop.hbase.protobuf.generated.ClientProtos$MutationProto$ColumnValue$QualifierValue
5: 2372963 56951112 com.google.protobuf.LiteralByteString
6: 1720444 41290656
java.util.concurrent.ConcurrentSkipListMap$Index
7: 7024 18547960 [I
{code}
.... shows QualifierValue at 1/10th of that the numbers used be with no mention
of the LiteralByteString which used to have near equal pegging with [B
instances. Let me commit then since it makes branch-1 YCSB-able (Previous it
was not). Will be back to try and make more improvements but this is enough
for one issue.
> Memory retention in branch-1; millions of instances of LiteralByteString for
> column qualifier and value
> -------------------------------------------------------------------------------------------------------
>
> Key: HBASE-11887
> URL: https://issues.apache.org/jira/browse/HBASE-11887
> Project: HBase
> Issue Type: Bug
> Components: Protobufs
> Affects Versions: 0.99.0
> Reporter: stack
> Assignee: stack
> Priority: Critical
> Fix For: 0.99.0, 2.0.0
>
> Attachments: 11887.txt, Screen Shot 2014-09-03 at 10.18.58 AM.png
>
>
> Trying to test branch-1, I run out of mem pretty fast. Looking at dumps, I
> see too many instances of LiteralByteString. Seem to be 'qualifiers' and
> 'values' out of pb QualifierValue... and on up to the multi call into the
> server. Am having trouble finding how the retention is being done... Filing
> issue in meantime while work on it.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
