On 19.11.2009 10:10, conrad-tomcat.users.2...@tivano.de wrote: > Hi, > > On Thu, Nov 19, 2009 at 12:50:44AM +0100, Rainer Jung wrote: >> >> On 18.11.2009 17:01, conrad-tomcat.users.2...@tivano.de wrote: >>> >>> As you can see, 24552 (=3 * 8184) bytes are received almost immediately, >> >> 8184 looks like the body size of one full AJP packet (protocol used by >> mod_jk and Tomcat). > > yep, that's what I thought, too. It looks like the last, partially filled > AJP packet from the Tomcat response is not making it through the SSL > layer, somehow. Or whatever signals "end of response" to the SSL layer. > >>> while the rest is only transferred after 5 seconds. Leaving "-0" away >>> from the curl command line, the complete result is received immediately. >>> Requesting the same page via http instead of https, the complete result >>> is received immediately. The 5-second-delay can be seen using wget >>> instead of curl, too, so this is probably not a client problem. >>> >>> So far, the problem has only been seen on the production system. >>> Due to the load conditions, it is infeasible to run mod_jk with significant >>> logging output. >> >> To bad. >> >>> mod_jk configuration is straightforward, timeouts are not defined (i. e. >>> we use default values). >> >> That's not so nice but also likely not the cause of the problem. Can you >> run a network sniff (Wireshark et.al.) between Apache and Tomcat? > > No, that's infeasible due to the high traffic volume.
If the problem is easily reproducible, you can add another Apache instance which is not used by the main traffic in front of the same tomcat (simply clone your existing Apache, and when using different directories and ports, you can do that on the same machine), then switch that one to JkLogLevel trace and run your reproduction against that Apache. > The webapp behaviour (for this page) depends neither on the HTTP protocol > version nor on the presence of SSL. So I'm certain that the webapp delivers > the complete response immediately. Your webapp might, the Tomcat connector might not. We need to find out, and the above way is an easy way to tell. Regards, Rainer --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org For additional commands, e-mail: users-h...@tomcat.apache.org
