Nikolaos Milas:
> On 23/5/2011 9:26 ??, Nikolaos Milas wrote:
>
> > With some googling I found this rather old message:
> > http://archives.neohapsis.com/archives/postfix/2004-03/2663.html where
> > Wietse suggested to increase the var_flock_tries undocumented
> > parameter in main.cf (from 20 to 40).
> >
> > Would this suggestion be applicable in the case too?
> >
> >
>
> Although no one has responded on this, I would like to mention that I
> have not seen this error again for more than a week, since I upgraded to
> latest version (2.8.3) - built from source.
>
> I can't tell if it's a coincidence, but things seem to be running
> smoothly now.
FYI, the code around that error message has not changed since it
was put there in 2000.
Also FYI, virtual machine monitors have problems with accurately
delivering timer interrupts when they manage many virtual guests
(or virtual CPUs). Virtual machine monitors play with the guest's
clock to compensate. This makes Postfix's time-based defenses less
effective than they could be, because time is not smooth as it
should be.
A while ago, someone tried to run Postfix on eight virtual CPUs on
ESX. Postfix would randomly hang until they went down to two virtual
CPUs. That's not a bug in Postfix. It's incomplete virtual hardware
emulation that loses timer events that Postfix relies on.
Wietse
