[Pgcluster-general] Recovery after one node failure

[Pavel Shaydo]

Hello all,

first I'm forget to say that I'm using pgcluster-1.7.0rc12.

It seems that I've found cause of the problem. Pgreplicate processes
periodically sends pings to dbcluster processes. If dbcluster process
becomes unavailable, then pgreplicate marks it as DB_TBL_ERROR and do
not pings anymore. When dbcluster starting again it can synchronize
itself using one of pgreplicate servers. This server marks it as
DB_TBL_USE after recovery, but other pgreplicates knows nothing about
this and not replicate queries to restored dbcluster. 

Patch that solves this problem for me attached.

-- 
Best regards,
Pavel Shaydo


-----Original Message-----
From: Pavel Shaydo <pshaydo at elverils.com>
Reply-To: pgcluster-general at pgfoundry.org
To: Pgcluster-general <pgcluster-general at pgfoundry.org>
Subject: [Pgcluster-general] Recovery after one node failure
Date: Fri, 21 Mar 2008 15:52:30 +0300

Hello all,

we have following configuration:

Node1:            Node2:
ClusterDB1        ClusterDB2
Pgreplicate1      Pgreplicate2

It works, if we inserting record into table at one node it appears
at another, same for deletion. If we put Node1 down, insert some 
records into table at Node2 and when turn on Node1 again then Node1
gets synchronized with Node2. After this if we inserting record into
table at Node1 it appears at Node2, but not vice versa -- if we
inserting record into table at Node2 it not replicated to Node1.
After restarting pgreplicate on Node2 all works again.


-------------- next part --------------
A non-text attachment was scrubbed...
Name: pgcluster-1.7.0rc12-notify.patch
Type: text/x-patch
Size: 6918 bytes
Desc: not available
Url : http://pgfoundry.org/pipermail/pgcluster-general/attachments/20080325/eada6f3c/attachment.bin