(Apologies if this has been previously discussed; I wasn't able to form
a search on the list archive that yielded anything.)
We use Squid as a reverse proxy cache to single upstream host.
We have an issue when a cache entry expires; because refreshing that
cache entry from the upstream host frequently takes multiple seconds, if
we have a high rate of identical requests coming in (which we usually
do), *all* of the requests become upstream requests until one of them
finally ends up in cache.
So we end up with a large number of upstream requests that are redundant
and a put a lot of stress on the back-end systems that we'd like to avoid.
I'm thinking that one solution to this would be to have a cache miss
request have a 'refreshing' state, which indicates it is in process of
being refreshed, and squid should hold other identical requests until
the cache entry gets refreshed.
Has this been considered or discussed? Are there basic problems with
this approach that I haven't considered?
Thanks,
dwh
