Re: [squid-users] refresh.c/refreshIsCachable buglet?

>>>>> On Thu, 29 Jun 2006 17:02:19 +0200, Henrik Nordstrom <henrik@henriknordstrom.net> said:

> tor 2006-06-29 klockan 11:27 +0200 skrev Andreas J. Koenig:
>> We think this is a bug. Why should 0 byte replies be less cacheable?

> They are not less cacheable, but there is not much use in keeping them
> around after expiry for conditional cache validation which is what this
> test is about.

> A zero sized object with reasonably expiry time (at least 60 seconds
> from now) is still cached like anything else cacheable.

But according to these rules our objects have no "reasonable"
expiration time. Typically something between 20 and 60 seconds. But
that's what we really need to support, it's the nature of the
application. All our objects with more than 0 bytes Content-Length are
treated OK, but our objects with 0 byte Content-Length are dropped on
the floor. This makes no sense.

>> We're now planning to patch our squid to remove this chunk and hope
>> this will solve our problems. Are we missing something?

> You are welcome to if this helps you, but I think your real problem is
> elsewhere (lack of proper expiry information in the responses or
> similar).

Meanwhile we have removed the three lines and now squid delivers HITs
on our empty objects. Very nice. At least for us! We feel sorry for
our downstream squids who would also benefit from this fix. I think,
we have not any wrong header information, just an unusual short
expiration time (this is a news ticker, so we cannot go higher) and
frequent objects with zero content. We could write a 0x20 or some such
for empty content but in good old tradition we chose to send an empty
content response. That squid does not support such an application is
against the principle of least surprise.

I'd be very happy, if the squid team could reconceive this as a buggy
behaviour. Isn't it wonderful to improve software by just removing a
few lines?

-- 
andreas