There is no "race condition" within qmail simply because of load, so
it might benefit from some elaboration from your collegues as to what "race
condition" they are referring to.
qmail goes non-linear if the unprocessed queue gets large (> 20-30 is a
sign of trouble), and it goes non-linear for the same reasons that sendmail
does with a single directory for a queue. Unix directory operations are
typically non-linear as the directory grows.
One possible reason for local deliveries slowing is that qmail-send isn't
scheduling them as quickly due to the aforementioned problem or they are contending
for the same disk.
As always. Why a program is slowing down is total speculation unless an analysis of
resources is performed at the time. What did your analysis show?
Regards.
On Mon, Jan 24, 2000 at 07:28:56PM +0100, Andras Tudos - Computronic, C3 wrote:
> Hi,
>
> I have an operational theory question: when the load in a qmail setup
> reaches certain level the queue starts to grow and after a short time the
> number of unprocessed messages starts to grow as well. This is OK, but: at
> the same time the speed of local-deliveries slows down tenfold and the
> queue is filling up more and more. The only way to stop this is to stop the
> incoming flow of messages, then the local deliveries are fast again and the
> server recovers. Of course the whole story is about an overloaded i/o
> subsystem which has to be upgraded, but still I want to know why do the
> local deliveries slow down so much in a race condition to give an exact
> answer for the collegues who blame qmail and say this would not occur in a
> sendmail or other MTA based system.
>
> Andras Tudos
> C3, Budapest
>
