Re: HA, failover and load balancing / howto?

Single master point subsets wasn't the plan but is doable. Each
geographic region should have a local read copy.

Ben wrote:
> Are those geographical copies, or geographical subsets? Multi-master
> replication is hard with postgres (read: probably not going to happen)
> but if you can partition your data up so that you have one writer for a
> subset of records, that could work quite well. Especially if you have
> rich clients that can afford fast links between your regional servers.
>
> On Sat, 4 Aug 2007, hanasaki wrote:
>
>> I have some web applications and rich clients that need to
>> geographically localized copies (for network latency reasons) of a
>> database (East Coast, Central, West Coast and Japan) These will be
>> mostly read however there will be full CRUD activities going on. I
>> think this means that there will be a cluster in each region to deal
>> with load and single failures and when a whole region perhaps dies,
>> clients will fall back to another region.
>>
>> ex:
>> 4 servers for load East Coast
>> - db and webservers
>> 4 servers for load Central USA Coast
>> - db and webservers
>> 4 servers for load West Coast
>> - db and webservers
>>
>> The web applications (Java, tomcat, ejb3, jboss4, php) If one one, or
>> more web or db servers die, the others in the region are still used (ie:
>> just 'degraded') If all the db servers die in a region, the web server
>> and applications will hit the db servers in another region)
>>
>> How can all of this be setup and configured and how can failed db
>> servers be brought back online and updated to sync into the clusters?
>>
>> Also looking at the pro/con of doing this in Postgres vs mysql
>>
>> ---------------------------(end of broadcast)---------------------------
>> TIP 4: Have you searched our list archives?
>>
>> http://archives.postgresql.org/
>>