On May 17, 2013, at 7:25 PM, Alex Rousskov <rousskov@xxxxxxxxxxxxxxxxxxxxxxx> wrote: > On 05/15/2013 09:12 AM, Guy Helmer wrote: > >> I'm seeing something odd with icap REQMOD requests for HTTP POST >> requests in Squid 3.3.4: the encapsulated body appears to not be >> terminated by \r\n0\r\n. This seems to occur consistently with bumped >> SSL requests to graph.facebook.com: > > The ICAP-encapsulated HTTP request body is not terminated because Squid > has not received the entire HTTP request body yet. Your ICAP server > timeout is too aggressive for this particular transaction. Thanks for the analysis. Since the issue was not in the adaptation layer, further testing indicates that building with --enable_kqueue causes squid not to read the remainder of the body in these bumped HTTPS transactions. > Here are the logged steps, FYI: > […] > >> 45:06.920 kid1| ModXact.cc(647) parseMore: >> ICAP/1.0 400 Bad Request >> ... > > 15 seconds later, the ICAP server gives up and returns a [bogus] error > to Squid. Thanks for catching that. There was a bug in my server that kept it from returning a 408 response. Any thoughts on appropriate timeouts for the ICAP protocol? I have not seen recommendations for timeouts in RFC 3507 or the ICAP Errata. Best regards, Guy
