On 08/24/2016 11:56 PM, 曹士超 wrote: > I don`t know squid use smp , multi cpu process purge cache, each > process define cache_dir ,when purge cache Choose a different > worker,Lead to clear the cache fails Hello, Due to the language barrier, I am not sure what question you are asking, but I hope that the following facts give you enough information to find the right answer: The two SMP-aware caches are rock cache_dir and shared memory cache. All ufs-based cache_dirs and local memory cache are SMP-unaware. Shared memory caching is enabled by default when you use multiple workers. SMP caching is supported only when all your caches are SMP-aware. With SMP-aware caches all Squid workers should share the same set of caches. If you do not use SMP-aware caches, use a mixture of SMP-aware and SMP-unaware caches, or give different workers a different set of SMP-aware caches, then many caching-related things will fail to work in SMP mode (often without warnings or error messages). I am ignoring all those problematic cases in the text below! When receiving a DELETE transaction or otherwise purging a cached response, Squid must delete all copies of that response from all caches. This protocol requirement applies to SMP and non-SMP Squid alike, of course. There is a known Squid bug that leads to some copies of a being-purged response not deleted in SMP Squids. Factory is working on a patch to fix that protocol violation (with the exception of one corner case that goes beyond purging itself). Our fix is almost ready, and I hope to see the corresponding patch posted to squid-dev soon. HTH, Alex. _______________________________________________ squid-users mailing list squid-users@xxxxxxxxxxxxxxxxxxxxx http://lists.squid-cache.org/listinfo/squid-users