Re: Broken pipe... PART 4...
I actually don't get 404's... the site just grinds to a halt, and jk_2 spews tonnes of messages like this: Fri Mar 12 15:57:47 2004] [error] ajp13.service() Error forwarding ajp13:localhost:8009 1 0 [Fri Mar 12 15:57:47 2004] [error] mod_jk.handler() Error connecting to tomcat 3 [Fri Mar 12 15:57:48 2004] [error] Error ajp_process_callback - write failed [Fri Mar 12 15:57:48 2004] [error] ajp13.service() ajpGetReply recoverable error 3 [Fri Mar 12 15:57:48 2004] [error] ajp13.service() Error forwarding ajp13:localhost:8009 1 0 [Fri Mar 12 15:57:48 2004] [error] mod_jk.handler() Error connecting to tomcat 3 [Fri Mar 12 15:57:48 2004] [error] Error ajp_process_callback - write failed The messages in catalina.log depend on whether I was recreating the problem in test, or watching it in production. In test, I would get many Broken Pipes, but in production I saw many Closed Sockets. Pepijnm, are you using Windows or a 'nix environment? In Unix, the CPU usage for the JVM goes through the roof. On Fri, 2004-03-19 at 09:52, Pepijn Palmans - Kangaroot wrote: Just to add some PART 3... it might be usefull to have the mod_jk connector give a clear message other than 404 to the browser, wouldn't it ? (of course we can setup a custom 404, but it's not the same as a real 404 !) Pepijn On Fri, 19 Mar 2004, Chris Boyce wrote: I was literally about to post on the same subject. If anyone knows how Tomcat (and in my case Apache 2.0.48+mod_jk2 on Solaris 8) handles interrupted/broken connections this would be helpful. Just to add my own observations... I can push over our test environment simply by hitting refresh (rapidly) for our front page, which does contain some SQL queries. By just one browser continuously interrupting the connections with refresh, I can watch the Java process in top climb over 80% CPU and the site becomes unresponsive. With the SAME test pointed at JRun 3.0, I cannot push JRun over the edge... Java reaches approximately 40-50% CPU and levels off. What could Tomcat (or at least Tomcat 4.1.30) be doing differently than JRun in handling these adverse threads? Does Tomcat 5 handle this any differently? Thanks On Fri, 2004-03-19 at 08:15, Pepijn Palmans - Kangaroot wrote: Hello there, When I use apache + mod_jk + tomcat I see something \strange\. When I connect my browser via apache to a servlet, hit the refresh button before I got a response this happens: - the tcp connection between browser and tomcat server is dropped. - apache knows that it's thread can be recycled and stops serving this useless request. - tomcat doesn't stop processing the request to the servlet. - a new processor is started. - the old one exits after a while with a 'Broken pipe' message. (jara.io.IOException) Is there any way to get the java/tomcat thread stopped when apache detects that a certain request is no longer needed ? At this moment, if I just keep refreshing some servlet my machine gets way to slow due to the fact that it has to process a lot of useless requests. As you can understand, suppose a servlet takes 2 seconds to process, the more I hit refresh, the slower the processing of a thread becomes, and the more people get fed up waiting for a response and hitting refresh, and the slower the tomcat server gets After a while, the tomcat doesn't even respond anymore. I found some solution in tuning the min/max processors in server.xml, tuning the acceptCount etc... but I think the main solution lies in the way tomcat handles the 'no longer needed' requests. Maybe another connector between apache and tomcat does tell tomcat it can stop the request ? As far as I know the ajp13(mod_jk) protocol is an asynchronous protocol. Is that correct ? Regards, Pepijn Palmans Managing Director Kangaroot Linux Solutions Grote Steenweg 91 2600 Berchem, Antwerpen Tel: +32 3/286.17.17 Fax: +32 3/281.23.49 Email: [EMAIL PROTECTED] - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: Broken pipe... PART 4...
I'm using Linux I see a lot of 'no processors available' in de catalina_log files. processor usage is indeed eating whole cpu, but machine doesn't die at all. do note that I use mod_jk, not mod_jk2 ! I think there's a whole different way of logging etc. in jk2. also something I don't really like, but can understand is the fact that when I have my maxprocessors on 100, and I use 80 of them on a busy moment, tomcat keeps having 80 processors available in stead of closing them down when they're not needed anymore (cfr. the way apache does with it's threads) Any way to tune that one ? It whould make it easier to see the actual usage of the tomcat on a machine by for instance counting the processes. (as if that is a real solution for monitoring :)) Pepijn On Fri, 19 Mar 2004, Chris Boyce wrote: I actually don't get 404's... the site just grinds to a halt, and jk_2 spews tonnes of messages like this: Fri Mar 12 15:57:47 2004] [error] ajp13.service() Error forwarding ajp13:localhost:8009 1 0 [Fri Mar 12 15:57:47 2004] [error] mod_jk.handler() Error connecting to tomcat 3 [Fri Mar 12 15:57:48 2004] [error] Error ajp_process_callback - write failed [Fri Mar 12 15:57:48 2004] [error] ajp13.service() ajpGetReply recoverable error 3 [Fri Mar 12 15:57:48 2004] [error] ajp13.service() Error forwarding ajp13:localhost:8009 1 0 [Fri Mar 12 15:57:48 2004] [error] mod_jk.handler() Error connecting to tomcat 3 [Fri Mar 12 15:57:48 2004] [error] Error ajp_process_callback - write failed The messages in catalina.log depend on whether I was recreating the problem in test, or watching it in production. In test, I would get many Broken Pipes, but in production I saw many Closed Sockets. Pepijnm, are you using Windows or a 'nix environment? In Unix, the CPU usage for the JVM goes through the roof. On Fri, 2004-03-19 at 09:52, Pepijn Palmans - Kangaroot wrote: Just to add some PART 3... it might be usefull to have the mod_jk connector give a clear message other than 404 to the browser, wouldn't it ? (of course we can setup a custom 404, but it's not the same as a real 404 !) Pepijn On Fri, 19 Mar 2004, Chris Boyce wrote: I was literally about to post on the same subject. If anyone knows how Tomcat (and in my case Apache 2.0.48+mod_jk2 on Solaris 8) handles interrupted/broken connections this would be helpful. Just to add my own observations... I can push over our test environment simply by hitting refresh (rapidly) for our front page, which does contain some SQL queries. By just one browser continuously interrupting the connections with refresh, I can watch the Java process in top climb over 80% CPU and the site becomes unresponsive. With the SAME test pointed at JRun 3.0, I cannot push JRun over the edge... Java reaches approximately 40-50% CPU and levels off. What could Tomcat (or at least Tomcat 4.1.30) be doing differently than JRun in handling these adverse threads? Does Tomcat 5 handle this any differently? Thanks On Fri, 2004-03-19 at 08:15, Pepijn Palmans - Kangaroot wrote: Hello there, When I use apache + mod_jk + tomcat I see something \strange\. When I connect my browser via apache to a servlet, hit the refresh button before I got a response this happens: - the tcp connection between browser and tomcat server is dropped. - apache knows that it's thread can be recycled and stops serving this useless request. - tomcat doesn't stop processing the request to the servlet. - a new processor is started. - the old one exits after a while with a 'Broken pipe' message. (jara.io.IOException) Is there any way to get the java/tomcat thread stopped when apache detects that a certain request is no longer needed ? At this moment, if I just keep refreshing some servlet my machine gets way to slow due to the fact that it has to process a lot of useless requests. As you can understand, suppose a servlet takes 2 seconds to process, the more I hit refresh, the slower the processing of a thread becomes, and the more people get fed up waiting for a response and hitting refresh, and the slower the tomcat server gets After a while, the tomcat doesn't even respond anymore. I found some solution in tuning the min/max processors in server.xml, tuning the acceptCount etc... but I think the main solution lies in the way tomcat handles the 'no longer needed' requests. Maybe another connector between apache and tomcat does tell tomcat it can stop the request ? As far as I know the ajp13(mod_jk) protocol is an asynchronous protocol. Is that correct ? Regards, Pepijn Palmans Managing Director Kangaroot Linux Solutions Grote Steenweg 91 2600 Berchem, Antwerpen Tel: +32 3/286.17.17 Fax: +32 3/281.23.49 Email: [EMAIL
