On Thu, May 24, 2012 at 3:03 AM, Winkelmann, Thomas (Radio/Tele FFH - Online) <[email protected]> wrote: > Hello, > > we have a setup with two varnish servers (version 3.0.2) and two backend > servers. This setup works fine, if we split the requests with an > loadbalancer to the two varnish servers. Around 500req/s 20MBit/s bandwidth > on each server. > > In Addition we have defined a backend probe which will tests the backend > each 10s. This also works fine. > > Now, if we switch all requests to one varnish server, after some time > (around 20 seconds) the backends will be marked as sick, but they are both > up and answering all requests. Also without any backend probe defined in > VCL, varnish gets in trouble serving content. Some months ago, we had no > problems with this and served more than 400Mbit/s from one single server.
I don't really know exactly what could be, but sometimes when I get backend errors (missconfiguration or overload on servers), and begin debugging with varnishlog: varnishlog -i Backend_health ( | "grep Backend_health" to show only the probes) This usually gives me the return code from the probes and any problems with it. If the server begins to return a different code than HTTP 200, you can see exactly when this happens... It's a start :) -- []'s Hugo www.devin.com.br _______________________________________________ varnish-misc mailing list [email protected] https://www.varnish-cache.org/lists/mailman/listinfo/varnish-misc
