[squid-users] smp purge

[Alex Rousskov]

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.