On 01/07/2015 11:28 AM, deans wrote:
> Hi Guys.
>
> First of all, BDR is cool, should have tried it earlier.
>
> Environment: CentOS 6.5, PostgreSQL 9.4.0 with BDR from yum repository
>
> Done the PostgreSQL 9.4 with BDR setup successfully by following the User
> Guide and Admin Doc, but got a issue when tried to do postgresql service
> restart on any of the cluster nodes, it always lost items in the relation
> "pg_stat_replication" after the service restarted
It'd be weird if it didn't. That's not a table, it's a view over a
function that reports server status information.
It's like pg_stat_activity for replication.
How exactly do you restart the node?
> means lost replication
> connections, like normally we have 26 items(13 databases and 3 servers in
> cluster, so 13x2 items in pg_stat_replication on each node, yes, lots dbs on
> one cluster), but after the restart, got some random count of item in the
> relation, like 20, or 16 or 4, but not 26 as expected, i.e. not all the
> replication connections running well.
That's normal, in that it can take time for peers to notice and
re-connect. This is true whether it's BDR, other logical replication, or
conventional streaming replication.
> And the logs showing replication is
> "wait until the node has caught up", but they never catch up again.
Please show the full log line, not just an excerpt, along with the lines
around it.
If possible run with log_error_verbosity=verbose and show the extra
line(s) printed after each log line too, please.
> BDR Settings(replaced the real db name here):
> 1. on node 01, the replication src one:
>
> 2. on node 02, mostly like 03
You forgot these.
> Error logs:
and these.
--
Craig Ringer http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Training & Services