Date: Thu, 26 Feb 2015 11:00:07 -0500
From: "David F. Skoll" <dfs@roaringpenguin.com>
To: pgsql-admin@postgresql.org
Cc: John Scalia <jayknowsunix@gmail.com>
Subject: Re: Weird spikes in delay for async streaming replication
 on 9.1
Message-ID: <20150226110007.09bb1551@hydrogen.roaringpenguin.com>
In-Reply-To: <54EF411B.80202@gmail.com>
References: <20150214110050.2f032f05@ollie.roaringpenguin.com>
 <20150226094428.34531376@hydrogen.roaringpenguin.com>
 <54EF3E6F.60403@gmail.com>
 <20150226104433.1b0df1db@hydrogen.roaringpenguin.com>
 <54EF411B.80202@gmail.com>
Organization: Roaring Penguin Software Inc.
Mime-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Precedence: bulk
Sender: pgsql-admin-owner@postgresql.org

On Thu, 26 Feb 2015 10:51:55 -0500
John Scalia <jayknowsunix@gmail.com> wrote:

> OK, if it's asynchronous, is your script checking that primary isn't
> holding up closing out and transmitting the latest WAL segment during
> these times?

No.  The checking script is dead simple.  It's a perl script that does
more-or-less the equivalent of this.  $master_dbi is a DBI handle
pointing to the master database server and $standby_dbi is one pointing
to the standby.  Pseudocode is shown below (the real code actually measures
the duration in much finer increments than 1s, but you get the idea...)

   my $time = time();

   # There's one row in the table with key column 'Timestamp'
   $master_dbi->do("UPDATE table SET value = ? WHERE key = 'Timestamp'", undef, $time);

   my $start = time();
   while(..) {
       $row = $standby_dbi->fetchrow_arrayref("SELECT * FROM table WHERE key = 'Timestamp'");
       # Check that we got back the same $time we put in and if so,
       # break out of the while loop; otherwise pause for a bit.
   }
   # See how long that took
   my $duration = time() - $start;

> And if the Standby really needs to be up to date, why
> not try synchronous replication?

It doesn't need to be totally up-to-date.  It can lag by up to 30s without
impacting our application.  And we don't use synchronous replication because
then a network problem would stall all the write transactions on the primary
and that would be fatal.

Regards,

David.


-- 
Sent via pgsql-admin mailing list (pgsql-admin@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-admin