httpd-bugs mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From bugzi...@apache.org
Subject [Bug 53239] Httpd has problems delivering data on rate-limited line
Date Wed, 16 May 2012 20:19:13 GMT
https://issues.apache.org/bugzilla/show_bug.cgi?id=53239

--- Comment #4 from Stefan Fritsch <sf@sfritsch.de> ---
I don't think this is httpd's fault. The strace shows that at 08:14:49.102908,
httpd starts to send data. The socket buffer gets full a few times, httpd uses
poll() to wait each time and sends more data. This works as expected until
08:14:49.105963 when httpd has sent about 75k of data. There, the socket buffer
is full again but the poll() returns only 25 seconds later, at 08:15:15.686502.
This means either there was no data going through the network and the buffer
was really not writable for these 25 seconds, or there is a kernel bug.

Of course, it is rather strange that Apache httpd triggers this problem but it
seems lighttpd doesn't. Accordingly, I have no idea what we could do to avoid
the problem. If the kernel tells us we can't write to the socket, there is
nothing we can do about that.

You could try analyzing the network traffic with wireshark and see if there are
lost packets.

NB: There seems to be an unrelated bug somewhere because the mmap() of the read
file is done at unaligned offsets, which fails and makes httpd fall back to
writev(). I don't see this here on my system. This could be due to you using
apr 2.0 which is unreleased and not stable. I would recommend to use the newest
released versions of apr and apr-util instead (1.4.x).

-- 
You are receiving this mail because:
You are the assignee for the bug.

Mime
View raw message