I wonder whether we should have an option in CopyTable that identifies the originating cluster and avoids that copying back.
When I implemented master-master replication I only added that to the replication source (so that edit do not bounce between the masters forever).
From: Himanshu Vashishtha <[EMAIL PROTECTED]>
To: [EMAIL PROTECTED]
Sent: Tuesday, November 13, 2012 1:45 PM
Subject: Re: Master Master replication
On Tue, Nov 13, 2012 at 1:36 PM, Varun Sharma <[EMAIL PROTECTED]> wrote:
> I want to setup a master-master replicated cluster using Hbase 0.94 - the
> two clusters will be in different availability zones. I was wondering if
> the following would work:
> 1) Setup cluster A and start reading/writing to it and run it for a week.
> 2) Now Setup cluster B and configure the two clusters as Master-Master.
> Is it possible to have such a setup, basically adding a master replica to
> an existing standalone hbase cluster ?
It will behave as master-master from the time you set that up (from
the current WALs).
Once the setup happens, would B
> inherit all of data that was written to cluster A prior to the replication
> setup ?
No, you need to run a copyTable job to do so. Note that you need to
run the copyTable before starting the replication otherwise, B will
send it back to A.
a) Run the copyTable job, (note the start timestamp of the job).
b) Start the replication.
c) Run the copyTable again with starttime equal to the starttime
noted in step 1.