On Tuesday, April 10, 2001, at 03:42 PM, Graham Leggett wrote:
20-30 sec "dead time" and suddenly, a bunch of connections to the browser complete, probably a timeout close flush. IE 5.1b1 OS X is the browser.Chuck Murcko wrote:
It's those experimental unc.edu servers. 8^)
Aside from some child exit sluggishness when shutting down, I'm seeing what looks like timeouts every so often in http proxy. Probably related to keepalives.
What do you mean by timeouts?
Also, the odd group of
[error] (32)Broken pipe: core_output_filter: writing data to the network
What platform is this? I've been hunting down this bug for almost a week - it's been driving me nuts.
SuSE 6.4, 2.2.14 kernel with 3.1.25 PCMCIA support.
./configure --prefix=/usr/local/www --with-mpm=prefork --enable-proxy --enable-maintainer-mode
Running a LinkSys wireless card WPC11.
I'll see if I can make this recur with ab more often.
Ryan Bloom says the most likely cause is the socket being allocated from the wrong pool, but I've been checking them over and over and so far nothing. The "keepalive mismatches" are worrying me too - there should be absolutely no way for this message to ever appear when upstream keepalives are not used (such as when an HTTP/1.0 browser like Netscape v4.7 is used) and I'm seeing it now and again in the logfile.
The single only time a socket should live past the end of ap_proxy_http_handler() is when the request was successful, the upstream browser was HTTP/1.1 or higher, and there was no Connection:close.
The search continues...
Time to look for race conditions?
Chuck
