[
https://issues.apache.org/jira/browse/HBASE-17182?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15726090#comment-15726090
]
Phil Yang commented on HBASE-17182:
-----------------------------------
Oh, one concern is the close logic. I thought I had post it, sorry. If we close
a expired scanner in the main thread when it is removed by the guava cache, the
accessing may be blocked to wait the close RPC request. It is bad. But the
expired scanner must also be expired at region server (and the TTL in thrift
server should same as TTL at region server which is scanner timeout I think).
So the close RPC will only get an UnknownScannerException. In that case I am
not sure if we still need to close the scanner. Maybe let it be GCed is enough?
[[email protected]]'s patch also notice this issue. In the patch the close
logic is commented out.
[[email protected]] What do you think? Thanks.
> Memory leak from openScanner of HBase thrift2
> ---------------------------------------------
>
> Key: HBASE-17182
> URL: https://issues.apache.org/jira/browse/HBASE-17182
> Project: HBase
> Issue Type: Bug
> Components: Thrift
> Reporter: Jian Yi
> Assignee: Jian Yi
> Attachments: HBASE-17182-V1.patch, HBASE-17182-V2.patch
>
>
> Client call openScanner, but client (coredump or others) not closeScanner,
> the scanner will not be removed from scannerMap.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
