Home | About | Sematext search-lucene.com search-hadoop.com
 Search Hadoop and all its subprojects:

Switch to Threaded View
HBase >> mail # dev >> [Question] replication got impact by zoo.cfg (if put under ./hbase/conf)


Copy link to this message
-
[Question] replication got impact by zoo.cfg (if put under ./hbase/conf)
hi, folks,

I encountered a problem last week that my cluster was replicating to
itself. The problem was discussed and described here:
http://mail-archives.apache.org/mod_mbox/hbase-dev/201311.mbox/%3CCAOEq2C5%2ByqsewgzpkPO%3D8FRvQVv6tqOoo5wrdxDVN%2BzRzV1iWw%40mail.gmail.com%3E

Finally, we found the problem was introduced because someone copied
./zookeeper/conf/zoo.cfg to ./hbase/conf. While the problem was fix, we are
still wondering why replication code pick up the zoo.cfg? Can someone throw
some light here?

To briefly sum up about how to recreate the problem, and what's wrong (btw,
I am using hbase 0.94.9):
1) to recreate: copy the zoo.cfg under hbase/conf, and that is
2) the regionsever (of source Cluster) log will show something like:
2013-11-04 14:07:30,364 INFO
org.apache.hadoop.hbase.replication.regionserver.ReplicationSource:
Replicating *d59d578a-3276-474a-9f15-17fc08a4415e* ->
*d59d578a-3276-474a-9f15-17fc08a4415e*
the first value is correct (hbase.id of the source), but the 2nd should be
the hbase.id of peer. And it turns out, the replication logic didn't get
the correct one

I did some investigation, and was able to trace back to
ReplicationPeer.zkw.quorum was set incorrectly (with zoo.cfg under
./hbase/conf).

As odd as this is, I feel this may be a bug of replication logic. As we
should get the peer.zkw.quorum as the way 'add_peer' provides, instead of
retrieving it from another place, even in this case the zoo.cfg should not
be copied under ./hbase/conf.

Any ideas ? many thanks

Demai