On Monday 28 July 2008 11:04:36 Kristian Rink wrote: > Folks; > > dealing with this issue again; yet not really being capable of resolving > it. Maybe someone does have some more ideas on that. Scenario: > > - apache 2.2 (Ubuntu 8.04) as reverse proxy in front of a glassfish v2 > application server (seems to be unrelated to the latter one, though, same > issue appearing with tomcat 6.x). Reverse proxying is done using mod_proxy > and ProxyPass / ProxyPassReverse directives, but the same behaviour also > appears using tomcat+mod_jk. Likewise, the same behaviour used to be around > using apache 2.0, both in mod_proxy- and mod_rewrite setup. > > - Generally, access to proxied resources works for all clients. However in > some cases (involving downloading larger files - > 10 megs, up to 250 megs > in some situations), connection seems to prematurely end, resulting in data > not being completely transferred to the client. On the java backend > servers, this is being noticed as a ClientAbortException - "connection > reset by peer". The clients to complain about this behaviour however > haven't consciously terminated the transmission / reset the connection. > This behaviour _seems_ to be limited to virtually all versions of MS > Internet Explorer, while in some cases on the same client host things work > well with Firefox or Opera (on others, people don't have a choice...). Most > common explanation to be found around the web is that "MSIE sometimes > doesn't cleanly end connections" and so these exceptions are caused by this > behaviour, thus not to worry about them - in our case we actually _have_ to > worry about them anyway as they _do_ end the connection / transmission. :/ > > - After discussing this on the glassfish mailing list [1], I have been > hinted to disable keep-alive connections in the backend system which did > provide a minor improvement though not resolve all of the issue. > > - Messing with KeepAlive configuration on apache2.2, same way, had as > little effect as applying the "Protocol Adjustments" outlined in [2]. > > - Thoroughly browsing through the apache2.2 error log, I don't see any > things hinting me what possibly could be wrong here... > > So, overally, I am pretty clueless now, lacking even a good starting point > for further investigations and debugging. Has anyone out here seen similar > things and, eventually, got past them? Kristian, I've observerd similar problems with large POST requests which sometimes time out on the backend servers while the client is sending the request body (like you with mod_proxy/apache 2.2 on Ubuntu 8.04). But the whole issue might actually not be caused by apache at all, but by a bug in the TCP implementation of the linux kernel 2.6.24. (See this thread: http://marc.info/?t=121518334200003&r=1&w=2) For me setting /proc/sys/net/ipv4/tcp_frto to 0 helped to solve the problem for a coulple of users, but not for everyone. I'm currently compiling 2.6.25 to see if this issue is still present there. Best regards, -Rainer --------------------------------------------------------------------- The official User-To-User support forum of the Apache HTTP Server Project. See <URL:http://httpd.apache.org/userslist.html> for more info. To unsubscribe, e-mail: users-unsubscribe@xxxxxxxxxxxxxxxx " from the digest: users-digest-unsubscribe@xxxxxxxxxxxxxxxx For additional commands, e-mail: users-help@xxxxxxxxxxxxxxxx
|