On Thu, Apr 28, 2011 at 06:34:11PM -0700, Rich Rauenzahn wrote: > I just ran a test where I requested (via curl) the same resource in > parallel from varnish, and I got multiple backend requests (I checked > my apache log). > > Could it be that there is a small race condition that is possible to > hit such that two backend requests get initiated if they are close > enough in time? (my test was a curl ... & curl ... & -- if I do it by > hand, which is slower, they seem to get bundled.)
That sounds unlikely. It's more likely that the object isn't being cached at all. Can you provide varnishlog output from a test like that? Also, setting obj.ttl to 1d in vcl_hit means never expiring objects, as you reset the ttl every time they are accessed... So the only way they are expired is if they aren't requested for a day. - Kristian _______________________________________________ varnish-misc mailing list [email protected] http://www.varnish-cache.org/lists/mailman/listinfo/varnish-misc
