Mailing-List: contact bugs-help@httpd.apache.org; run by ezmlm
Precedence: bulk
Reply-To: "Apache HTTPD Bugs Notification List" <bugs@httpd.apache.org>
From: bugzilla@apache.org
To: bugs@httpd.apache.org
Subject: DO NOT REPLY [Bug 36636]  -
    blocking writev()/poll()-call causes mod_dav to stop responding
Message-Id: <20050914193244.4293B136@ajax.apache.org>
Date: Wed, 14 Sep 2005 21:32:44 +0200 (CEST)

DO NOT REPLY TO THIS EMAIL, BUT PLEASE POST YOUR BUG�
RELATED COMMENTS THROUGH THE WEB INTERFACE AVAILABLE AT
<http://issues.apache.org/bugzilla/show_bug.cgi?id=36636>.
ANY REPLY MADE TO THIS MESSAGE WILL NOT BE COLLECTED AND�
INSERTED IN THE BUG DATABASE.

http://issues.apache.org/bugzilla/show_bug.cgi?id=36636


------- Additional Comments From arne@arne.net  2005-09-14 21:32 -------
Thank you so much for your help! I put a check for f->c->aborted in
core.c:core_output_filter and thereby reduced the number of timed out writes to one.
But the core issue still remains. I am sending this prepared PROPFIND which
causes a large answer, and each request will just stop sending ACKs in the
middle of the transaction. In about 1 in 10 cases (or so), the behaviour changes
and the poll() locks, it is always with the same data to write, with the exact
same packets being sent by the client.

A "normal" situation looks like this:
20:52:09.407647 [ffffe410] writev(8, [...], 3) = 3686 <0.000013>
20:52:09.409004 [ffffe410] writev(8, [{"1f5e\r\n", 6}, ...], 4) = 5074 <0.000014>
20:52:09.409095 [ffffe410] poll([{fd=8, events=POLLOUT,
revents=POLLERR|POLLHUP}], 1, 300000) = 1 <0.246947>
20:52:09.656085 [ffffe410] writev(8, [...], 3) = -1 EPIPE (Broken pipe) <0.000013>
20:52:09.656163 [ffffe410] --- SIGPIPE (Broken pipe) @ 0 (0) ---

This I can strace all the time, but occasionally, it is this:

20:57:26.463493 [ffffe410] writev(8, [...], 4) = 1800 <0.000014>
20:57:26.463608 [ffffe410] poll([{fd=8, events=POLLOUT, revents=POLLOUT}], 1,
300000) = 1 <0.011306>
20:57:26.474956 [ffffe410] writev(8, [...], 3) = 6250 <0.000022>
20:57:26.476328 [ffffe410] writev(8, [{"1f5e\r\n", 6},...], 4) = 2510 <0.000012>
20:57:26.476424 [ffffe410] poll([{fd=8, events=POLLOUT}], 1, 300000) = 0
<299.851111>

In the latter case, poll does not return the POLLERR. The connection is then in
CLOSED_WAIT, after a while, it goes to LAST_ACK (while the send-Q is 20k).
I read something about the behaviour on 
http://www.ussg.iu.edu/hypermail/linux/kernel/0407.3/0495.html
but I'm not sure if this is intended in this case. 

Is it the kernel or apache that is loosing track of the FD? Why is the kernel
not signalling the SIGPIPE / POLLERR?

-- 
Configure bugmail: http://issues.apache.org/bugzilla/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.

---------------------------------------------------------------------
To unsubscribe, e-mail: bugs-unsubscribe@httpd.apache.org
For additional commands, e-mail: bugs-help@httpd.apache.org