scheduling off the waiting list
slink at schokola.de
Tue Dec 29 07:58:08 CET 2009
thank you for your quick response.
>> So basically there are two different scenarios when hsh_rush is called.
>> * Trigger delivery of an object which just got unbusied
>> * and trigger delivery of more sessions which did not fire in the first round
> The basic point here is that we do not want to unleash 500 waiting sessions
> when the object is unbusied, so we release a few, and when they are done
> they each release a few etc.
Sure, and I think this is a good idea.
I should have mentioned that the reason why I care about this is that in
HSH_Prepare, memory is allocated on sp->http->ws each time a session fires, so
when it doesn't get to delivery, we can run out of session memory. And I have a
dozen coredumps showing the workspace filled up with hashptrs.
The symptoms look very much like those in
http://varnish.projects.linpro.no/ticket/551, and reports of these symptoms
might be related to this issue.
> Grace mode helps a lot, if you can use it.
More information about the varnish-dev