[
https://issues.apache.org/jira/browse/HBASE-2129?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=12829335#action_12829335
]
Jean-Daniel Cryans commented on HBASE-2129:
-------------------------------------------
Today I set up a replication stream between 2 clusters in 2 different
datacenters, each cluster has 20 nodes and it's replicating on average 50 000
rows per second. I will let it run for some time to see how it goes.
While setting it up I noticed the following:
- If a zoo.cfg is present in conf/, it completely overrides any value you may
have set directly on a HBC object so using add_peer.rb doesn't work. As a
workaround I moved all the configurations from that file to hbase-site.xml and
deleted the file.
- When using the add_peer.rb script, it's really important to give the exact
hbase.zookeeper.quorum specified in hbase-site.xml or the master cluster will
think he's a slave. It would be nice to add that notice in the script when it
confirms the addresses.
- When a region server is choosing peers it prints out a message like this:
"Considering 19 rs, with ratio 1.9". In this case we should read something more
like "Considering 2 rs, with ratio 0.1". Fix that.
- When setting up the second cluster, you either have to distcp the hbase
folder in HDFS from the first cluster or recreate all the tables empty. If
doing the latter, it can be a real pain if you have a lot of tables and you
don't keep the DDL scripts around so I wrote a jruby script that reads all the
HTableDescriptors from the master cluster and passes them to a HBaseAdmin
configured with the address of the second cluster. I could add it in the patch.
> Simple Master/Slave replication
> -------------------------------
>
> Key: HBASE-2129
> URL: https://issues.apache.org/jira/browse/HBASE-2129
> Project: Hadoop HBase
> Issue Type: Sub-task
> Reporter: Jean-Daniel Cryans
> Assignee: Jean-Daniel Cryans
> Fix For: 0.21.0
>
> Attachments: HBASE-2129-v2.patch, HBASE-2129.patch
>
>
> We should first build a very simple replication mechanism to validate our
> assumptions and get a feel of what replication is in this very distributed
> context.
--
This message is automatically generated by JIRA.
-
You can reply to this email to add a comment to the issue online.
