Re: [squid-users] garbage before reply

ok, i've looked where you told me, added debugging and it never gets
to that point (i assume it thinks the stream is never corrupted). Like
i said, this happens only with squid -> dansguardian, and not with
either individually, I'm at a loss trying to figure this out.. to add
to the confusion if i add debug_options ALL,9 it happens much less
often leading me to believe it's a race condition or timing error somewhere,

I'm not sure what is happening to my cache either, there are zero hits
for some reason, although cache manager shows objects in the cache,
the store log shows many many releases (which i think is normal for
uncacheable objects and when stuff expires) but everything else
comes up as a TCP_MISS or a CP_CLIENT_REFRESH_MISS even
if i do not use the refresh button on the browser (i tried to open the
same page in another brower on another computer and got that)

Is this all because of squid-3's maturity or am I hitting problems
due to something else (perhaps the way i'm implementing this?)

I'll do whatever i can to get these bugs squashed and get the
patches upstream but I'm really stuck right now, any suggestions?

Thanks.

Joshua Brindle

>>> Robert Collins <robertc@squid-cache.org> 08/22/03 04:45PM >>>
On Sat, 2003-08-23 at 07:08, Joshua Brindle wrote:
> I've started to isolate it. I'm using dansguardian so I don't know if it
> is there or in squid, I can't reproduce it with only dansguardian, and
> I can't reproduce it with just squid, but it happens every time with both.
>
> I've turned debug_options ALL,9 on and am going line by line when it
> happens to something and am coming up with nothing right now
> (lots of logs to look through though), and i'm also not sure if this
> is the cause of the broken pictures either... Where do you think the
> cause of this is so I can take a look?

I suspect that a cached file is corrupt, and that the re-retrieval code
has a buglet. We ran into a couple of segfaults recently due to
unparsable replies - which correlates with this. There was a buglet in
handling those replies - but there is still the unexplained cause of
them...

The place to look, is client_side_reply.cc: 1445 - where the reply
parsing has failed.

Add some debug fluff there, or run under gdb and breakpoint it.

Then, if this code block is invoked every time you see the bug, we need
to work backwards to identify it.

A bt full would be a good place to start.

Re: the images, lets squash this bug first and see if they get magically
better.

Cheers,
Rob

-- 
GPG key available at: <http://members.aardvark.net.au/lifeless/keys.txt>.