On Mon, 2009-10-05 at 14:20 -0400, Stephen Gallagher wrote:
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
On 10/05/2009 01:33 PM, Simo Sorce wrote:
> Should fix 218 (tested with non responding ldap server), and also fix
> other races and potential memleaks.
Nack.
I don't like that we're extending the timeout for the lookup operation.
It's there for a reason. If it doesn't respond to the first call within
one timeout period, then all we're doing by extending it with subsequent
calls is potentially trapping ourselves in an infinite wait with an
ever-growing number of requests.
Consider the case where our backend crashes and is restarted by the
monitor. If we have a busy system where one entry is being requested
continuously, say once per second, and the backend that NSS is talking
to segfaults for some reason, that request would never get a reply, and
the timeout would continue to be extended. Furthermore, all new requests
would just continue to add to the queue of callbacks, never to see the
light of day.
No, it's much better to just let the first timeout cause all of them to
return immediately, so that a new request can try again.
Yeah hadn't thought about it from this pov.
Will remove the extending timeout part.
Simo.