On Tue, Feb 23, 2016 at 3:09 PM, Michael Paquier
<michael.paquier@gmail.com> wrote:
> On Tue, Feb 23, 2016 at 6:43 AM, John Wiencek <jwiencek3@comcast.net> wrote:
>
>> I have a three node cluster using streaming replication configured as
>> follows:
>> One synch node and one asynch node.
>> If my synch node is down the master node is hung until that node is brought
>> back on line.
>>
>> Two questions:
>> 1. Is this the expected action on my master? It makes sense if it is since
>> the master is waiting for an ack from the synch node.
>
> Yes, when synchronous_commit = on, the default values that you should
> have, the master will wait from standby the confirmation that the WAL
> record for the commit has been flushed;
> http://www.postgresql.org/docs/devel/static/runtime-config-replication.html#RUNTIME-CONFIG-REPLICATION-MASTER
>
>> 2. Is there anyway to configure my cluster so I have a synch node but not
>> have my master node “HANG” if the synch node is down?
>
> You can do that at transaction level for example by disabling
> synchronous_commit.
Or you could configure both of your standbys as synchronous standbys.
Only one of them will actually be a synchronous standby at a time, and
the other one will take over that role if the first one is down, so
your system won't hang but you'll still have the sync standby
guarantee.
--
Thomas Munro
http://www.enterprisedb.com