There is something amiss. The client has seen a ZK transaction ID far ahead of
what the ZK server thinks is the current epoch. Usually this happens if you
blow away ZK storage and restart it, ie you are creating transient ZK quorums.
A client that continues to run will remember the zxid of the old ZK quorum
while the new one will start again with a zxid of 0. Another way this can
happen is if you've some how mixed two distinct ZK quorums in your HBase
configuration: it contains the names of servers in two distinct quorums. Those
two ZK clusters will have uncorrelated zxids.
Check your ZK configuration.
Best regards,
- Andy
On Feb 15, 2012, at 6:34 AM, "Kaluskar, Sanjay" <[email protected]>
wrote:
> I am new to hbase, I can't get the Hive handler working. I downloaded
> the latest Hive (0.8.1) which has a handler for 0.89, and based on the
> instructions on
> https://cwiki.apache.org/confluence/display/Hive/HBaseIntegration I
> recompiled hive after updating the hbase, zookeeper and guava versions
> in src/ivy/library.properties to match what hbase 0.92.0 has. The build
> is successful, but a simple create table (from the above wiki) hangs.
>
>
>
> I checked the logs, and looks like the client connection gets refused by
> zookeeper; log repeats the following:
>
>
>
> 2012-02-15 05:33:22,729 INFO
> org.apache.zookeeper.server.NIOServerCnxnFactory: Accepted socket
> connection from /10.1.42.203:60625
>
> 2012-02-15 05:33:22,730 INFO
> org.apache.zookeeper.server.ZooKeeperServer: Refusing session request
> for client /10.1.42.203:60625 as it has seen zxid 0x30000002c our last
> zxid is 0x2000000f1 client must try another server
>
> 2012-02-15 05:33:22,730 INFO org.apache.zookeeper.server.NIOServerCnxn:
> Closed socket connection for client /10.1.42.203:60625 (no session
> established for client)
>
>
>
> BTW everything basic is working on my Hbase cluster (shell commands,
> Java programs to put/get, map-reduce). Am I doing something wrong? Has
> anybody else seen this?
>
> Thanks,
>
> -sanjay
>
