From: Pin007 <pin007@gmail.com>
Reply-To: pin007@gmail.com
To: Merlin Moncure <mmoncure@gmail.com>
Subject: Re: Streaming replication performance
Date: Tue, 16 Aug 2011 22:03:16 +0200
User-Agent: KMail/1.13.7 (Linux/3.0.0-1-686-pae; KDE/4.6.5; i686; ; )
Cc: pgsql-performance@postgresql.org
References: <201108111646.39321.pin007@gmail.com>
	<CAHyXU0zCVv24bzEeM8rANLK3ZyGo2VO43DS0ZwomNfkn9GsfTA@mail.gmail.com>
In-Reply-To: 
 <CAHyXU0zCVv24bzEeM8rANLK3ZyGo2VO43DS0ZwomNfkn9GsfTA@mail.gmail.com>
MIME-Version: 1.0
Content-Type: Text/Plain;
  charset="iso-8859-1"
Content-Transfer-Encoding: quoted-printable
Message-Id: <201108162203.16635.pin007@gmail.com>

Dne P=E1 12. srpna 2011 18:47:24 Merlin Moncure napsal(a):
> On Thu, Aug 11, 2011 at 9:46 AM, Antonin Faltynek <pin007@gmail.com> wrot=
e:
> > Hi all,
> >=20
> > I'm testing Streamin replication with one hot standby node and I'm
> > experiencing high delay of hot standby node.
> >=20
> > When I reach aprox. 50 transactions per second where every transaction
> > includes only simple "UPDATE status SET number =3D number + 1 WHERE nam=
e =3D
> > 'iterations'", the standby server falls behind master and the delay is
> > several minutes. In comparison, Slony1 have no issues with the same test
> > and its delay is only several seconds.
> >=20
> > Am I missing some configuration that affects speed of streaming
> > replicaiton?
>=20
> hm -- how are you determining the standby is behind?  are you running
> any other queries on the standby?
>=20
> merlin

Simply, during test I'm generating series of numbers that I'm periodicaly=20
checking on both master and stand by server.

Also I'm measuring several server metrics, like CPU usage, disk usage, netw=
ork=20
bandwidth.

When testing streaming replication network usage does not get over 1Mbps an=
d=20
catching master takes several tens of minutes, while during test of WAL fil=
e=20
shipping, network usage get to 3-5Mpbs and stand by server does not fall=20
behind master more than several seconds.

There is no other traffic on stand by server, only periodical check of test=
=20
table (SELECT * FROM status) every 2s. Extension of this check interval doe=
s=20
not affect behavior of streaming replication.

Tonda