On Sat, Jun 5, 2010 at 1:52 PM, Cosimo Streppone <[email protected]> wrote:
> That sounds surprising to me. Can you explain? > Say that I restart varnish. > Then 10 clients hit the "/takes_long_time_to_load.html" page at > the same time. > > What I understand from your reply is that varnish will hold the > 9 remaining connections until the first one sees the backend > reply. At that point, the 9 remaining clients will get served > the cached object. Yes. There is a whole lot of work spent on the code that releases the 9 remaining clients. At some point (1.x) we had massive problems when thousands of clients where released at the same time. > Is that really how it works? If it where to work any other way it would be a bit silly, wouldn't you think? Load spikes could really kill the backend. > I'm specifically referring to the first time after restart, > so grace doesn't apply here. Yes. Grace is something else. Grace is a deliberate, this is not. You can't turn this behavior off. Evgeniy: I saw your reference to hit-for-pass. Hit-for-pass objects are created especially for the situation where we find that it doesn't make sense to queue up objects. A "hit for pass" object is created to cache this finding so incomming reqeusts are passed directly when there is a hit-for-pass object. -- Per Buer, Varnish Software Phone: +47 21 54 41 21 / Mobile: +47 958 39 117 / skype: per.buer _______________________________________________ varnish-misc mailing list [email protected] http://lists.varnish-cache.org/mailman/listinfo/varnish-misc
