On Aug 6, 2009, at 3:42 PM, Dan Cook (dancook) wrote:
> The rub comes from connections that have not expired in the pool
> after a
> server restart - the next grab will fail (I just tested this using
> gdb).
> I don't think there is anything a programmer can do about this except
> retry.
Yes. Even ping-on-grab() isn't 100% foolproof, as the DB server or
network link could bounce between a successful ping and the query.
So, if you either need a bulletproof system or these are common cases
for you, you still need connection retry logic.
> But then how do you purge (expire) all the connections in the
> pool to get them to reset and do a fresh reconnect?
Ping-on-grab() would handle that. I don't see a compelling reason to
special-case this, dumping all the connections at once. Just handle
it as part of your normal connection robustness assurance code.
