Is it OK for users to use either of a pair of replicated servers ?

John cyrus at jelmail.com
Fri Feb 1 12:16:27 EST 2013


On 01/02/13 16:42, Adam Tauno Williams wrote:
>
> A modseq is very much like an etag or a ctag in HTTP/WebDAV.  It is a
> value that gets incremented with every change.  So the the modseq on the
> slave is greater than the modseq on the master... something is out of
> sync.
I guessed that's what it was. thanks for clearing that up for me.
>> On the master:
>> sync_client[5197]: MAILBOX received NO response: IMAP_MAILBOX_CRC
>> Checksum Failure
>> sync_client[5197]: CRC failure on sync for user.myuser, trying full update
>> sync_client[5197]: SYNCNOTICE: record mismatch with replica: user.myuser
>> more recent on master
> Aren't you connecting to the slave and making changes?  That would make
> sense then, the master and the replica are constantly getting
> out-of-sync.  Replication is one-way.
Yes. My original server became the master and the new one became the 
replica. I then switched off main retrieval on the master and enabled it 
on the replica so that would naturally become ahead.

If I understand correctly the replication isn't designed to work that 
way (although it does an admirable job of doing so). I need to configure 
the "ahead" server as the master and the other as the replica (e.g. 
switch the replication direction around). I'll do that over the weekend.
>> Despite these messages, my replication appears to be working but I can't
>> as yet be 100% sure. I'd like to understand the above and try and stop
>> them if I can...
> Because it is constantly recovering, as these messages indicate.  2.4.x
> replication is quite reliable and recovers from inconsistencies most of
> the time.
It works very well. If I understand from your earlier comments, when 2.5 
comes out will this support bi-directional replication along the lines 
of what I am trying to do ?

Meanwhile I'll stop updating the replica except via the sync client/server.



More information about the Info-cyrus mailing list