[squid-users] understanding behaviour with collapsed_forwarding patch

Hi,

I patched squid-2.5.3 with the collapsed_forwarding-2_5.patch from Hendrik
Nordstrom. Now I have some behaviour I don't understand. I have set up the
patched Squid as a reverse proxy (127.0.0.1:80) and turned on
collapsed_forwarding in the squid configuration file. The website(127.0.0.1:81)
which it accelerates has a nice perl script which gives _no_ output for 10
seconds and after this it returns a nice HTTP conform header with an expire of
now() + 120 seconds. ok now let me try to describe a situation I don't
understand:

* The object index.pl was purged from the cache

Client 1 requests http://127.0.0.1/index.pl (I turned on debug loglevel 33,7):

2003/08/27 13:18:58| clientProcessRequest2: storeGet() MISS
2003/08/27 13:18:58| clientProcessRequest: TCP_MISS for 'http://127.0.0.1:81/index.pl'
2003/08/27 13:18:58| clientProcessMiss: 'GET http://127.0.0.1:81/index.pl'

I added some debug messages just to see if the new object is made immediately
public, allowing new requests to attach to the pending request. As I understand
correctly from the documentation
(http://devel.squid-cache.org/collapsed_forwarding/)

2003/08/27 13:18:58| storeSetPublicKey(http->entry)

It does!

During the 10 seconds it needs to get the content of index.pl I request 1 or
several other requests for the same object (index.pl). I expect the request to
be attached to the pending request and only 1 total request should arrive on
the backend server. This doesn't happen. All the requests send to squid for the
same object during the 10 seconds index.pl needs to send output is forwarded to
the backendserver. A more detailed debug output for such a request:

2003/08/27 13:30:13| clientRedirectDone: 'http://127.0.0.1:81/index.pl' result=NULL
2003/08/27 13:30:13| clientInterpretRequestHeaders: REQ_NOCACHE = NOT SET
2003/08/27 13:30:13| clientInterpretRequestHeaders: REQ_CACHABLE = SET
2003/08/27 13:30:13| clientInterpretRequestHeaders: REQ_HIERARCHICAL = SET
2003/08/27 13:30:13| clientProcessRequest: GET 'http://127.0.0.1:81/index.pl'
2003/08/27 13:30:13| clientProcessRequest2: !storeEntryValidToSend MISS
2003/08/27 13:30:13| clientProcessRequest: TCP_MISS for 'http://127.0.0.1:81/index.pl'
2003/08/27 13:30:13| clientProcessMiss: 'GET http://127.0.0.1:81/index.pl'
2003/08/27 13:30:13| storeSetPublicKey(http->entry)

To shortely describe my problem again. If I eg. concurrently make 50 requests
for the same url (which is not in the cache) with a Squid with the
collapsed_forwarding patch. When the request in my test setup hangs for 10
seconds all the 50 concurrent requests are send to the backend server as I
understand from the documentation this is just the thing that the patch should
prevent it from doing.

Could anyone help me in understanding why this behaviour occures? Or what I
should investigate more to solve the problem.

Thanks,
Brendan
Received on Wed Aug 27 2003 - 05:45:41 MDT