HI John,
I would not mind to go for linux. Is there any guide to setup failover
safe postgres servers on linux/windows ?
I found few, but noe of them is complete/descriptive.
Regards,
Sanjay Rao
On 6/9/2011 11:30 PM, John R Pierce wrote:
> On 06/09/11 2:45 AM, Sanjay Rao wrote:
>> Hi,
>>
>> If anybody have setup this thing earlier or have any idea about how
>> to setup this, Please let me know. I am newbie in setting up database
>> servers .
>>
>
> frankly, for high availabilty first thing I'd do would be to take
> windows out of the picture, it just complicates everything.
>
> two general approaches to a high availability database server.
>
> 1) shared storage, with an active-standby pair of servers, managed by
> some form of cluster management software like Linux Heartbeat, Redhat
> or Veritas Cluster Service, etc etc. each server has its own IP,
> then the active server has the 'server' IP that clients connect
> to. active server has storage mounted, standby server is 'fenced'
> (usually done via a SAN switch) so it can't accidentally mount the
> storage thats in use. on failover, the formerly active server is
> fenced, the standby server is unfenced, mounts the storage, takes over
> the shared IP and starts the database service. clients see this as a
> service interruption, usually lasting a few seconds to a half minute,
> they reconnect when the former standby is ready and alive.
>
> 2) replication. active server is replicating data to the standby
> server. the rest of the failover scenario is similar minus the
> shared storage part. unless the replication is synchronous, there is
> a possibility of losing the last few transactions if the master hard
> fails.
>
>
>
> in general, high availability clusters are complex, delicate beasts,
> require careful attention to the details, and once working, should NOT
> be messed with except extremely carefully. an absolute bare minimum
> should be running on the database servers so as not to complicate things.