On 31/05/11 00:17, Daniel Beschorner wrote:
>>> Connections with FIN_WAIT1 state on ICAP server side seem ESTABLISHED at
>> squid.
>>
>> ICAP-closed connection.
>>
>> Idle pconn in Squid have readers set listening for FIN to arrive and
>> close the FD. This is strange but not conclusive.
>>
>> Looks a bit like the FIN never arrived.
>>
>>> Squid connections in CLOSE_WAIT state are no longer visible at ICAP server
>> side.
>>
>> Squid-closed connection.
>>
>> FIN packet sent by both sides. FIN-ACK packet from ICAP server not
>> arriving at Squid box. This confirms the FIN are not flowing right.
>>
>>
>> Both cases are pointing to packets containing FIN not flowing from the
>> ICAP server to Squid. Though strangely seem fine going in the other
>> direction.
>
> In both cases Squid has a large Recv-Q. Does Squid no longer empty the queue and therefore misses the FIN or should the FIN work out-of-band?
>
> Daniel
>
No. Squid should be draining its queue, even if leaving the connections
idle. Being in the idle pool sets a read to abort/close the socket on
FIN or on excess data.
This does sound like those ICAP incomplete reply problems. Though how
its getting into Squids idle pool without the act of insertion killing
the socket with an excess data read is baffling me.
Amos
-- Please be using Current Stable Squid 2.7.STABLE9 or 3.1.12 Beta testers wanted for 3.2.0.7 and 3.1.12.1Received on Mon May 30 2011 - 12:58:56 MDT
This archive was generated by hypermail 2.2.0 : Tue May 31 2011 - 12:00:03 MDT