[
https://issues.apache.org/jira/browse/HBASE-5153?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13185601#comment-13185601
]
Jieshan Bean commented on HBASE-5153:
-------------------------------------
Thanks, Ted.
Without HBASE-3065, 0.90 doesn't handle the ConnectionLossException correctly.
Consider the below case:
1. Somewhere trigger a HConnection#abort.
2. Suppose the check of "if (t instanceof
KeeperException.SessionExpiredException)" is true. Then called the
resetZooKeeperTrackers().
3. A ConnectionLossException occur during ZookeeperNodeTracker#start. then
trigger a new HConnection#abort. At this scenario, the previous abort may print
a log of
"Reconnected successfully. This disconnect could have been caused by a network
partition or a long-running GC pause......"
4. The new abort carry a Throwable with a type which is not
KeeperException.SessionExpiredException. so this time abort directly.
It seems a recursion here.
Either re-use the old connection by resetZooKeeperTrackers, or re-create the
connection, the ZookeeperWatcher will be a new one. So I still think the patch
for 0.90 is reasonable.
Trunk patch will be made big changes.
So any other good suggestions? Thanks.
> HConnection re-creation in HTable after HConnection abort
> ---------------------------------------------------------
>
> Key: HBASE-5153
> URL: https://issues.apache.org/jira/browse/HBASE-5153
> Project: HBase
> Issue Type: Bug
> Components: client
> Affects Versions: 0.90.4
> Reporter: Jieshan Bean
> Assignee: Jieshan Bean
> Fix For: 0.90.6
>
> Attachments: 5153-trunk.txt, HBASE-5153-V2.patch,
> HBASE-5153-V3.patch, HBASE-5153-V4-90.patch, HBASE-5153-trunk-v2.patch,
> HBASE-5153-trunk.patch, HBASE-5153.patch
>
>
> HBASE-4893 is related to this issue. In that issue, we know, if multi-threads
> share a same connection, once this connection got abort in one thread, the
> other threads will got a
> "HConnectionManager$HConnectionImplementation@18fb1f7 closed" exception.
> It solve the problem of "stale connection can't removed". But the orignal
> HTable instance cann't be continue to use. The connection in HTable should be
> recreated.
> Actually, there's two aproach to solve this:
> 1. In user code, once catch an IOE, close connection and re-create HTable
> instance. We can use this as a workaround.
> 2. In HBase Client side, catch this exception, and re-create connection.
--
This message is automatically generated by JIRA.
If you think it was sent incorrectly, please contact your JIRA administrators:
https://issues.apache.org/jira/secure/ContactAdministrators!default.jspa
For more information on JIRA, see: http://www.atlassian.com/software/jira
