[
https://issues.apache.org/jira/browse/HBASE-1311?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12696730#action_12696730
]
Nitay Joffe commented on HBASE-1311:
------------------------------------
Hi Andrew,
I like the idea of the reinit() with the ephemeral node map.
However this patch fundamentally changes a lot of things in the design of
ZooKeeperWrapper. The initial idea was not to have any retries in
ZooKeeperWrapper so that each user of it can handle it differently. Each
ZooKeeper operation was supposed to either succeed or fail simply and the code
calling it would do what it needs.
Take a look at how I handled SessionExpired in the client, TableServers. I
think it is much cleaner to have each ZooKeeper user (TableServers,
HRegionServer, and HMaster) register itself as a watcher and handle
SessionExpired for itself. SessionExpired is not something that is particular
to a single operation, so littering every ZooKeeper call with it seems a bit
much?
Thoughts?
> ZooKeeperWrapper: Failed to set watcher on ZNode /hbase/master
> --------------------------------------------------------------
>
> Key: HBASE-1311
> URL: https://issues.apache.org/jira/browse/HBASE-1311
> Project: Hadoop HBase
> Issue Type: Bug
> Reporter: Andrew Purtell
> Assignee: Nitay Joffe
> Attachments: dumb-wrapper-hack.patch
>
>
> After about 12 hours of operation, this repeats over and over in the
> regionserver log:
> 2009-04-05 19:44:38,445 WARN
> org.apache.hadoop.hbase.zookeeper.ZooKeeperWrapper: Failed to set watcher on
> ZNode /hbase/master
> org.apache.zookeeper.KeeperException$SessionExpiredException: KeeperErrorCode
> = Session expired
> at org.apache.zookeeper.KeeperException.create(KeeperException.java:118)
> at org.apache.zookeeper.ZooKeeper.exists(ZooKeeper.java:709)
> at
> org.apache.hadoop.hbase.zookeeper.ZooKeeperWrapper.watchMasterAddress(ZooKeeperWrapper.java:235)
> at
> org.apache.hadoop.hbase.regionserver.HRegionServer.watchMasterAddress(HRegionServer.java:343)
> at
> org.apache.hadoop.hbase.regionserver.HRegionServer.process(HRegionServer.java:339)
> at org.apache.zookeeper.ClientCnxn$EventThread.run(ClientCnxn.java:366)
> 2009-04-05 19:44:38,445 WARN
> org.apache.hadoop.hbase.regionserver.HRegionServer: Unable to set watcher on
> ZooKeeper master address. Retrying.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
