On 23/07/11 09:49, Dan Ford wrote:
We have our cache setup in the following manner: cache_dir ufs /cache_dir 100000 16 256 refresh_pattern . 10512000 100% 10512000 While testing everything looked good but after putting a bit of load on the cache we're seeing a lot of RELEASE entries in the log (non FFFFFFFF entries) and I don't understand why this would be the case, since our refresh pattern basically says to keep entries in as long as possible. The cache isn't being hit with any special expiry or no-cache headers, just the User-Agent, Connection, Keep-Alive and Host headers. Here are
The reply header are more relevant to the refresh_pattern. Check what they are and maybe enable debug level 22,3 to see what refreshCheck is deciding.
some examples of the RELEASE messages from our store.log. 1311359725.359 RELEASE 00 00019ACB 324E9C9538B7DE8EA5D8DCB7925E668B 200 1311225933 0 -1 audio/mpeg 80875/80875 GET http://10.0.21.15:8080/TranscodingInterface/preview/RABFaaTI 1311359737.598 RELEASE 00 00015642 312792582E333AD8185F95EDC0C04DCF 200 1311264767 0 -1 audio/aac 72865/72865 GET http://10.0.21.15:8080/TranscodingInterface/preview/RAC2iWNP 1311359737.747 RELEASE 00 000198C3 0A560353D2B2F834CE8225F7A429339E 200 1311357210 -1 1311457210 x-squid-internal/vary -1/0 GET http://10.0.21.15:8080/TranscodingInterface/preview/aHR0cDovL2NvbnRlbnQuOXNxdWFyZWQuY29tL21lZGlhL21wMy9wb2x5LzI3MDg3Ny5tcDM If the content is hit with a different User-Agent will this cause the cache to release the object as the result from the origin server will differ in size / content with a different User-Agent?
Depends on which user agents are involved with the cached version and the new request. Whether the new agents details match the Vary: details the cachesd request was generated from.
Amos -- Please be using Current Stable Squid 2.7.STABLE9 or 3.1.14 Beta testers wanted for 3.2.0.9
