[Pgpool-general] SQL sessions die when a node fails

[Jaume Sabater]

On Mon, Jul 13, 2009 at 2:28 AM, Tatsuo Ishii<ishii at sraoss.co.jp> wrote:

>> In case described, we do not want to RESTART the session but rather continue
>> the session with the functioning (not failed)  postgress nodes.
>>
>> So unless we miss something, we are not clear as to what you refer to by
>> "restart the session" - why restart, and which session? (client to pgpool?
>> pgpool to active nodes? pgpool to failed nodes?)
>
> 1) PostgreSQL fails and stops working while returning a query result
> 2) pgpool detects #1 and tries to continue the session between client
>   and pgpool
> 3) #2 is difficult because of the reason I mentioned

I think that Tatsuo means that it's easier and safer to rollback the
current transaction (in all nodes) and restart (or start again, come
to that) than find out what was already processed by the surviving
node and try to apply the rest.

Nevertheless, we all agree that the end result would be commiting the
current transaction to the remaining node(s) and returning an okay to
the client, so that the service can continue (this is a feature I'd
like to have, too).

-- 
Jaume Sabater
http://linuxsilo.net/

"Ubi sapientas ibi libertas"