[Dovecot] Is Dovecot SMP and thread-safe?
Josh´s problem report reminds me of other multi-CPU issues. Just a guess, Peter
Dovecot doesn't use threads itself anywhere. I don't know about OpenLDAP library though.
hmm. are you using login_process_per_connection = yes or no?
On Thu, 2005-03-03 at 09:14 -0800, Josh Burley wrote:
We *are* on an SMP machine...
Peter Franck wrote:
Josh´s problem report reminds me of other multi-CPU issues. Just a guess, Peter
Not to discount the SMP hypothesis, but do you possibly have a password database that uses uid's > 65535?
If there's some application in the chain that is assuming uid's should be a short, and another app in the chain assumes that uid's should be an int... Then when mapping int to short, you might lose the upper bits - which should only matter for uid's >= 65536.
On Thu, 2005-03-03 at 09:14 -0800, Josh Burley wrote:
We *are* on an SMP machine...
Peter Franck wrote:
Josh´s problem report reminds me of other multi-CPU issues. Just a guess, Peter
That would resolve the UID the same each time, even though incorrectly. So it would result in a more systematic error rather than the intermittent issue Josh is having I would say.
Steve
Dan Stromberg wrote:
Not to discount the SMP hypothesis, but do you possibly have a password database that uses uid's > 65535?
If there's some application in the chain that is assuming uid's should be a short, and another app in the chain assumes that uid's should be an int... Then when mapping int to short, you might lose the upper bits - which should only matter for uid's >= 65536.
On Thu, 2005-03-03 at 09:14 -0800, Josh Burley wrote:
We *are* on an SMP machine...
Peter Franck wrote:
Josh´s problem report reminds me of other multi-CPU issues. Just a guess, Peter
Ah, heya Dan. Haven't seen you in awhile.
Nope, no such luck. Most of our users are in the 3000's. I'm trying out the current setup, using ldap directly as the userdb. We'll see how it plays out.
.josh
Dan Stromberg wrote:
Not to discount the SMP hypothesis, but do you possibly have a password database that uses uid's > 65535?
If there's some application in the chain that is assuming uid's should be a short, and another app in the chain assumes that uid's should be an int... Then when mapping int to short, you might lose the upper bits - which should only matter for uid's >= 65536.
On Thu, 2005-03-03 at 09:14 -0800, Josh Burley wrote:
We *are* on an SMP machine...
Peter Franck wrote:
Josh´s problem report reminds me of other multi-CPU issues. Just a guess, Peter
participants (5)
-
Dan Stromberg
-
Josh Burley
-
Peter Franck
-
Stephen Kurzeja
-
Timo Sirainen