https://issues.apache.org/bugzilla/show_bug.cgi?id=44592
Spil <[EMAIL PROTECTED]> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|NEEDINFO |ASSIGNED
--- Comment #17 from Spil <[EMAIL PROTECTED]> 2008-06-30 12:38:58 PST ---
It seems I have a problem with the Keep-Alive in conjunction with a changing
order in sending and receiving of the TCP/IP frames.
We noticed that the last successful request from the client (as captured on the
client) near to exactly 5 seconds before the first part of the next POST was
sent by the client. Neither KeepAlive nor KeepAliveTimeout were set on Apache
leaving it default 'on' and 5 seconds.
(numbering of frame differs from previous post in this bugreport)
Trace on client
30 10:05:39.241734 S->C HTTP/1.1 200 OK [PSH,ACK]
31 10:05:39.390101 C->S [ACK]
32 10:05:44.269173 C->S First frame of POST
33 10:05:44.269313 C->S First frame of POST
34 10:05:44.275964 S->C [FIN,ACK]
Trace on proxy (clocks NOT synchronized, frame matching on Seq en Ack in tcp)
1800 11:05:55.007933 S->C SSL Continuation (client trace frame 30)
1801 11:05:55.010493 C->S ACK
1802 11:05:55.013463 C->S ACK
1803 11:05:55.163385 C->S ACK
1804 10:05:59.981127 [FIN,ACK] (client trace frame 34)
1805 10:06:00.048579 SSL Unreassembled packet (client trace frame 32)
1807 10:06:00.049030 Continuation (client trace frame 33)
Due to the latency of the network the sending and receiving order on client and
server-side have switched around.
My workaround was to set a specific KeepAliveTimeout in the Virtual Hosts
affected (37 seconds) on the productive instance. After 2 hours I had 2 more
errors:
- one where the previous request from the client was around 37 seconds before
the failing POST request
- the other had 2 requests in short succesion from which the last one failed
and the request before that was around 37 seconds earlier
The default Keep-Alive Time-Out of the Internet Explorer clients is 60 seconds
(source http://support.microsoft.com/kb/813827). I am now switching the
KeepAliveTimeout on the Apache servers to 67 seconds so it is longer than the
Keep-Alive timeout on the client.
Additionally I found my bug described in
http://support.microsoft.com/kb/921090, the PSH,ACK and FIN,ACK are listed in
the client trace. I am told that wininet.dll builds after the one mentioned in
this article can implement the fix by applying the registry setting
FEATURE_HANDLE_SSL_CLOSE_NOTIFY_KB921090 from the MSKB article. Hopefully I can
have this fix rolled out through the landscape (thousands of PC's) but that
still won't fix issues with clients not under our control.
Some questions remain open
- What are the downsides of increasing the KeepAliveTimeout > 12 times?
- Is apache behaving as should be expected?
--
Configure bugmail: https://issues.apache.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
