On 01.08.2012 09:53, alex fiddler wrote:
In my configuration, my ICAP server returns to Squid either an HTTP
403 or an HTTP 302 as an adapted response, before Squid consumes any
of the "virgin body". Consequently, the ICAP is destroyed before
Squid has a chance to consume any of the data in
"ModXact::virgin::body_pipe".
In my case, this is a valid use case. The adapted response should be
forwarded to the HTTP client, but it is not. Instead, squid is
waiting for "ServerStateData::virginBodyDestination" (which is a
reference to "ModXact::virgin::body_pipe") to be destroyed.
In Squid's code, in the function "BodyPipe::clearConsumer" I found
the comment saying
// do not abort if we have not consumed so that HTTP or ICAP
can retry
// benign xaction failures due to persistent connection race
conditions
My understanding is that this is controlled by icap_retry? Is the
condition described above considered by Squid as a failure, or is
this
a bug in Squid (I already opened a Squid bug 3595 on this topic)?
In my squid config file I put the directive icap_retry allow all, to
no avail. Or is the condition above handled by another configuration
directive?
I am using squid 3.1.15.
The dev who know most about adaptation implementation do not follow
this group closely. The bug report is the best avenue for this. If you
want a faster poke, try posting to squid-dev.
Amos
