Hi Alex With your explanation, I was able to reproduce the behaviour:
-> I opened a website within the browser -> Before the site was loaded completely, I closed the browser -> This causes squid to close the current icap-transaction (as you described) (-> ABORT-entries in the access.log) -> In the icap-log, I have the icap-error mentioned in my first post. So, this behaviour makes now sense. Probably the icap-daemon (third-party software) could be advised, not to throw an error but rather an informational message. Thanks a lot for your help and hints. Kind regards, Tom On Fri, Jun 21, 2013 at 5:13 PM, Alex Rousskov <rouss...@measurement-factory.com> wrote: > On 06/21/2013 02:05 AM, Tom Tom wrote: >> Hi >> >> Running squid 3.3.4 as an icap-client: >> >> Sometimes, I got on our ICAP-server (runs on the same box as squid) >> errors like: "ERROR write error: Connection reset by peer" > ... >> For me, the following lines seems curious: >> 2013/06/21 09:46:31.756 kid4| Xaction.cc(488) noteInitiatorAborted: >> Initiator gone before ICAP transaction ended >> 2013/06/21 09:46:31.756 kid4| Xaction.cc(503) setOutcome: ICAP_ERR_GONE > >> In the access.log, the error behaves like this: >> [Fri Jun 21 09:46:31 2013].756 0 10.xx.xx.xx NONE_ABORTED/000 0 >> GET http://www9.dict.cc/img/amtop1.gif USER HIER_NONE/- - >> >> The behaviour seems not to be reproducable. But it occurs n-times a >> hour. I also tried with set "icap_persistent_connections off". But >> this results in the same manner. > > Sounds like the HTTP client disconnected from Squid, forcing Squid to > abort related ICAP (and possibly server-side HTTP) transactions. This is > normal, but your ICAP server decided to complain about it. Ignore this > unless something does not appear to work for your HTTP clients. > > > HTH, > > Alex. >
