X-Original-To: pgsql-admin-postgresql.org@localhost.postgresql.org Received: from localhost (unknown [200.46.204.144]) by svr1.postgresql.org (Postfix) with ESMTP id AF01C32A056 for ; Thu, 21 Oct 2004 11:33:23 +0100 (BST) Received: from svr1.postgresql.org ([200.46.204.71]) by localhost (av.hub.org [200.46.204.144]) (amavisd-new, port 10024) with ESMTP id 59304-03 for ; Thu, 21 Oct 2004 10:33:10 +0000 (GMT) Received: from millenium.mst.co.jp (unknown [210.230.185.241]) by svr1.postgresql.org (Postfix) with ESMTP id 5B21732A154 for ; Thu, 21 Oct 2004 11:33:11 +0100 (BST) Received: from mst1x5r347kymb (lc12114 [192.168.1.114]) by millenium.mst.co.jp (8.11.6p2/3.7W) with SMTP id i9LAX9q04122 for ; Thu, 21 Oct 2004 19:33:09 +0900 Message-ID: <000901c4b759$55874fc0$7201a8c0@mst1x5r347kymb> From: "Iain" To: Subject: high (or at least improved) availability and failover Date: Thu, 21 Oct 2004 19:32:53 +0900 MIME-Version: 1.0 Content-Type: multipart/alternative; boundary="----=_NextPart_000_0005_01C4B7A4.C24926B0" X-Priority: 3 X-MSMail-Priority: Normal X-Mailer: Microsoft Outlook Express 6.00.2900.2180 X-MimeOLE: Produced By Microsoft MimeOLE V6.00.2900.2180 X-Virus-Scanned: by amavisd-new at hub.org X-Spam-Status: No, hits=0.5 tagged_above=0.0 required=5.0 tests=HTML_20_30, HTML_MESSAGE X-Spam-Level: X-Archive-Number: 200410/276 X-Sequence-Number: 15269 This is a multi-part message in MIME format. ------=_NextPart_000_0005_01C4B7A4.C24926B0 Content-Type: text/plain; charset="iso-2022-jp" Content-Transfer-Encoding: 7bit HI, I've been looking into options for improving the availability of postgres servers beyond what you get with a monolithic server and RAID. These are, as far as possible "cheap" options and therefore don't include use of shared storage and all the good options those systems have. Here is a very rough summary of my thoughts so far. I havn't actually tried any of this yet, so if anyone has any thoughts, comments, additions I'd be glad to hear from you. regards iain Configurations for HA ===================== Pgpool Only ----------- Pgpool can detect the failure of a server and failover automatically. Re-syncing the DBs after a failure will require the databases to be offline for a file system copy. Should pgpool run on the db servers or a separate machine(s)? Performance impact of using the replication facility? Deadlock possibility? If the machine that pgpool is running on fails, need manual intervention to start pgpool on another machine. Can only manage 2 servers at most. Pgpool and Heartbeat ----------------- Similar to pgpool only, with the addition that pgpool itself can be failed over automatically. Heartbeat and DRDB --------------- Replication is done at the block level by DRDB. The database cannot be running on the standby server while it is being replicated from the master. Automatic failover is possible as Heartbeat will start the database server and switch to the standby if the master fails. A full re-sync of the DBs after a failure doesn't require the master to be offline. Failure ofthe standby server typically doesn't require a full re-sync. Can't do load balancing. No particular need for pgpool as heartbeat will manage the aliased IP address Slony-I only ------------ Relies on triggers in the database, and some administration of the replication configuration. Monitors replicated databases and performs automatic failover. When a failed server is brought back online it will automatically start catching up with the other servers. When the master fails, clients can reconnect to the new master without changing the IP address? Slony-I and pgpool --------------- All the features of Slony-I Some load balancing of SELECT SQL is possible Need manual intervention if pgpool fails. Slony-I and pgpool and Heartbeat ---------------------------- All the features of Slony-I with pgpool Use heartbeat to monitor the machines running pgpool and switch over automatically if one fails. ------=_NextPart_000_0005_01C4B7A4.C24926B0 Content-Type: text/html; charset="iso-2022-jp" Content-Transfer-Encoding: quoted-printable
HI,
 
I've been looking into options fo= r=20 improving the availability of postgres servers beyond what you get with a= =20 monolithic server and RAID. These are, as far as possible "cheap"=20 options and therefore don't include use of shared storage and all the = good=20 options those systems have.
 
Here is a very rough summary of m= y=20 thoughts so far. I havn't actually tried any of this yet, so if anyone has = any=20 thoughts, comments, additions I'd be glad to hear from you.
 
regards
iain
 
Configurations for=20 HA
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
 
Pgpool Only
-----------
Pgp= ool can=20 detect the failure of a server and failover automatically.
 
Re-syncing the DBs after a failur= e will=20 require the databases to be offline for a file system copy.
 
Should pgpool run on the db serve= rs or a=20 separate machine(s)?
 
Performance impact of using the= =20 replication facility? Deadlock possibility?
 
If the machine that pgpool is run= ning on=20 fails, need manual intervention to start pgpool on another machine.<= /DIV>
 
Can only manage 2 servers at=20 most.
 

Pgpool and Heartbeat
-----------------
Similar to pgpool onl= y,=20 with the addition that pgpool itself can be failed over automatically.
 

Heartbeat and DRDB
---------------
Replication is done at th= e=20 block level by DRDB.
 
The database cannot be running on the standby server while it is being= =20 replicated from the master.
 
Automatic failover is possible as Heartbeat will start the database se= rver=20 and switch to the standby if the master fails.
 
A full re-sync of the DBs after a failure doesn't require the master t= o be=20 offline.
 
Failure ofthe standby server typically doesn't require a full=20 re-sync.
 
Can't do load balancing.
 
No particular need for pgpool as heartbeat will manage the aliased IP= =20 address
 

Slony-I only
------------
Relies on triggers in the database= , and=20 some administration of the replication configuration.
 
Monitors replicated databases and performs automatic failover.
 
When a failed server is brought back online it will automatically star= t=20 catching up with the other servers.
 
When the master fails, clients can reconnect to the new master without= =20 changing the IP address?
 

Slony-I and pgpool
---------------
All the features of=20 Slony-I
 
Some load balancing of SELECT SQL is possible
 
Need manual= =20 intervention if pgpool fails.
 

Slony-I and pgpool=20 and Heartbeat
----------------------------
All the features of= =20 Slony-I with pgpool
 
Use heartbeat to monitor the machines running pgpool and switch over= =20 automatically if one fails.
 
------=_NextPart_000_0005_01C4B7A4.C24926B0--