Message-ID: <50C912AC.8070502@2ndquadrant.com>
Date: Thu, 13 Dec 2012 07:26:36 +0800
From: Craig Ringer <craig@2ndquadrant.com>
User-Agent: Mozilla/5.0 (Windows NT 6.1; WOW64;
	rv:16.0) Gecko/20121010 Thunderbird/16.0.1
MIME-Version: 1.0
To: =?windows-1252?Q?Niels_Kristian_Schj=F8dt?= <nielskristian@autouncle.com>
CC: Evgeny Shishkin <itparanoia@gmail.com>, Jeff Janes <jeff.janes@gmail.com>,
	"pgsql-performance@postgresql.org" <pgsql-performance@postgresql.org>
Subject: Re: Do I have a hardware or a software problem?
References: <AE27BD07-7A3A-449A-BD3F-A0DF4EF76C04@autouncle.com>
	<CAMkU=1zcXD5YWMD=QOAdoNG0BKoB-DAXKbzBKybE-3w-YzhSgw@mail.gmail.com>
	<8F7661E7-D6D4-4334-8C6A-A8B6C56A2C3F@autouncle.com>
	<50C7351C.30407@2ndQuadrant.com>
	<4507A68A-3CD5-4283-AFA5-1C0593427E14@autouncle.com>
	<67FBF3AF-5C6A-45D4-9206-7257C8C01C5E@gmail.com>
	<CADZnK0nmm=2WnVVsveu3j0sJ95Ct13SqKbjQ_9DeVCGAmrcggw@mail.gmail.com>
	<E2A43856-FE0A-44BC-9764-C3CA533D94D9@gmail.com>
	<CADZnK0=0q9Ti9jVpCX3VX=3X_Rs=AxONhxC9cejnbOd5GKjb4A@mail.gmail.com>
	<ED7DEE06-B822-43E0-AF49-D89BB40C4C83@gmail.com>
	<50C7D7D2.4070605@2ndQuadrant.com>
	<2BCB6843-1396-42C9-A5BF-21C7C978FF4B@gmail.com>
	<50C7DDDE.1000000@2ndQuadrant.com>
	<3ADBD9A3-966F-40E8-AF78-0F2093D79BEE@gmail.com>
	<50C7E598.2070902@2ndQuadrant.com>
	<8A08ED7E-CF78-4F40-ACF1-1C87CEBBD00A@gmail.com>
	<50C7F057.70309@2ndQuadrant.com>
	<C6A943C0-BD00-4CC2-9D37-43EF2D98027F@autouncle.com>
In-Reply-To: <C6A943C0-BD00-4CC2-9D37-43EF2D98027F@autouncle.com>
Content-Type: multipart/alternative;
	boundary="------------010901060601040903030900"
Precedence: bulk
Sender: pgsql-performance-owner@postgresql.org

This is a multi-part message in MIME format.
--------------010901060601040903030900
Content-Type: text/plain; charset=windows-1252
Content-Transfer-Encoding: 8bit

On 13/12/2012 12:22 AM, Niels Kristian Schjødt wrote:
> Well, In fact I do (as you can see from my configuration). I have a
> similar server running with hot standby replication - and it runs two
> 3T HDD in a RAID1 array.
>
> So - is it still very bad if I choose to put four intel 520 disks in a
> RAID10 array on the other production server?
So long as you have it recording to a synchronous replia on another
machine and you're fully prepared to accept the small risk that you'll
have total and unrecoverable data corruption on that server, with the
corresponding downtime while you rebuild it from the replica, it should
be OK.

Alternately, you could use PITR with a basebackup to ship WAL to another
machine or a reliable HDD, so you can recover all but the last
checkpoint_timeout minutes of data from the base backup + WAL. There's
small window of data loss that way, but you don't need a second machine
as a streaming replication follower. barman might is worth checking out
as a management tool for PITR backups.

If the data is fairly low-value you could even just take nightly backups
and accept the risk of losing some data.

--
Craig Ringer

--------------010901060601040903030900
Content-Type: text/html; charset=windows-1252
Content-Transfer-Encoding: 8bit

<html>
  <head>
    <meta content="text/html; charset=windows-1252"
      http-equiv="Content-Type">
  </head>
  <body bgcolor="#FFFFFF" text="#000000">
    <div class="moz-cite-prefix">On 13/12/2012 12:22 AM, Niels Kristian
      Schjødt wrote:<br>
    </div>
    <blockquote
      cite="mid:C6A943C0-BD00-4CC2-9D37-43EF2D98027F@autouncle.com"
      type="cite">
      <meta http-equiv="Content-Type" content="text/html;
        charset=windows-1252">
      Well, In fact I do (as you can see from my configuration). I have
      a similar server running with hot standby replication - and it
      runs two 3T HDD in a RAID1 array.
      <div><br>
      </div>
      <div text="#000000" bgcolor="#FFFFFF">So - is it still very bad if
        I choose to put four intel 520 disks in a RAID10 array on the
        other production server?</div>
    </blockquote>
    So long as you have it recording to a synchronous replia on another
    machine and you're fully prepared to accept the small risk that
    you'll have total and unrecoverable data corruption on that server,
    with the corresponding downtime while you rebuild it from the
    replica, it should be OK.<br>
    <br>
    Alternately, you could use PITR with a basebackup to ship WAL to
    another machine or a reliable HDD, so you can recover all but the
    last checkpoint_timeout minutes of data from the base backup + WAL.
    There's small window of data loss that way, but you don't need a
    second machine as a streaming replication follower. barman might is
    worth checking out as a management tool for PITR backups.<br>
    <br>
    If the data is fairly low-value you could even just take nightly
    backups and accept the risk of losing some data.<br>
    <br>
    --<br>
    Craig Ringer<br>
  </body>
</html>

--------------010901060601040903030900--