On Wednesday, July 14, 2010, Poul-Henning Kamp wrote: > The the powers that be, that the author of Varnish said that it is by > far the least hackish way to get something working reliably.
Believe it or not, your and Artur's comments have helped to convince the powers. :-) While discussing the solution, one engineer brought up the following concern. Let's say Varnish gets hammered with user requests and is busy serving them. Now comes the driver HTTP client (the one that forces the retrieval of a file from the backend). If the system is very busy, the driver HTTP request will take a long time to be processed. The goal is to populate the cache as fast as possible, so that incoming user requests will always get the cached data. (This might be a dubious requirement, so tell me if it is so. ;-) Is there a way to give the driver request a higher priority? I have investigated two possibilities without any luck: 1. Run two Varnish instances against the same workspace/shared memory using identical "-n" settings but different frontend ports. This failed with an error message telling me I could not do that. The idea here is that one Varnish instance is "reserved" for the driver client. 2. Add VCL directives to prefer one request over another. I have read carefully through the VCL documentation but could not a way to give a particular request precedence over the rest. Does VCL support something like this? Thanks, Stephan -- Entrepreneur and Software Geek Google me. "Zope Stephan Richter" _______________________________________________ varnish-dev mailing list [email protected] http://lists.varnish-cache.org/mailman/listinfo/varnish-dev
