It seems we could be hurting on cache efficiency due to MW commonly sending Vary:Accept-Encoding and us not sanitizing the client's Accept-Encoding per the example in https://www.varnish-cache.org/docs/3.0/tutorial/vary.html . On a quick check of varnish logs, I see examples of both MW backend sending that Vary header, and of various clients sending us all kinds of variations (e.g. "gzip, deflate" vs "gzip,deflate" vs "gzip, deflate, sdch" (chrome))...
Description
Description
Event Timeline
Comment Actions
Change 206387 had a related patch set uploaded (by BBlack):
sanitize Accept-Encoding for cache efficiency T97128
Comment Actions
I want to read and investigate a little further first, and also test this in isolated places before cluster-wide, but the patch above is the general idea...
Comment Actions
Varnish already handles this sanely, I was just reading the wrong outdated documentation apparently!