-----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 Nicolas,
On 9/20/12 10:37 AM, Nicolas Sarazin wrote: > English is not my native language, please excuse typing errors. Welcome! > I met a problem which I do not manage to explain... > > My environment (one server) : > > Server version : Apache Tomcat/5.5.26 You have 10 days to upgrade, at which point you will be flying without a parachute: Apache support for Tomcat 5.5.x expires in 10 days. I highly recommend that you get Tomcat 7 or at least Tomcat 6 into your test environments ASAP. > JVM Version : 1.6.0_06-b02 You could afford to upgrade that, too. Oracle is on 1.6.0_35. > Web server version : Apache/2.2.21 (Unix) mod_jk/1.2.32 2.2.23 and 1.2.37 > My VirtualHost : > > [...] JkMount /* ajp13 [...] Do you have anything else? If not, why bother with Apache httpd? > [...] <!-- Define an AJP 1.3 Connector on port 8011 --> <Connector > port="8011" enableLookups="false" redirectPort="8443" debug="0" > maxThreads="600" keepAlive="true" backlog="8192" > minSpareThreads="25" maxSpareThreads="250" > connectionTimeout="600000" protocol="AJP/1.3" /> [...] That's a huge backlog, especially when you always expect a certain maximum number of connections coming from Apache httpd. > worker.properties : > > # define worker worker.list=ajp13 > > # Set properties for ajp13 => tomcat worker.ajp13.type=ajp13 > worker.ajp13.host=localhost worker.ajp13.port=8011 > worker.ajp13.connection_pool_size=1200 > worker.ajp13.connection_pool_timeout=600 > worker.ajp13.socket_keepalive=true worker.ajp13.socket_timeout=600 What MPM are you using? If you are using prefork, then your connection_pool_size is all wrong. Generally speaking, you should allow mod_jk to determine its own value for connection_pool_size when using Apache httpd. How many backend Tomcat servers do you have? Looks like one. Let's assume you are using threaded MPM in httpd (otherwise the value for 1200 is insane) and you are using only one backend Tomcat server. You have 1200 connections configured in httpd (connection_pool_size), but Tomcat can only accept 600 of them (maxThreads) at any given time. You have used backlog=8192 to cover this up so things become even more confusing. > Usually, it's work correctly, but sometime, only on certain pages, > woker can't connect to Tomcat. In my logs files, I have : I think it's only a coincidence that /page2 consistently gives you 500-response errors, here. Try looking at a wider section of your httpd access log to determine if there really is something special about /page2 (of course, /page2 could be returning 500-response itself: you might want to check on that). > At the beginning, I thought that the problem came from a bad > configuration here : In server.xml file : maxThreads="600" In > worker.properties : worker.ajp13.connection_pool_size=1200 > > But at the present time, there was no overload on the servers. You should really get your connection allocations straightened-out, even if you don't expect them to be a problem. If you expect 1200 connections from httpd, then set maxThreads=1200 in Tomcat. I see you have the same timeout values for both sides of the connection (good!). You don't need that insanely-high TCP backlog, so remove that, too. > How can we explain this behavior ? There are lots of explanations for what you are seeing. A few questions: 1. Do you really need Apache httpd at all? 2. Can you configure cping/cpong for connection liveness testing? 3. Have you tried disabling AJP connection re-use altogether? localhost communication is fast fast fast. - -chris -----BEGIN PGP SIGNATURE----- Version: GnuPG/MacGPG2 v2.0.17 (Darwin) Comment: GPGTools - http://gpgtools.org Comment: Using GnuPG with Mozilla - http://www.enigmail.net/ iEYEARECAAYFAlBbOikACgkQ9CaO5/Lv0PDhEwCfdEvr3VNpF/IEUvNlEzKD6qzN 1AsAoMQKM5BJXP/6etm2MsPbL3H32OOu =02al -----END PGP SIGNATURE----- --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org For additional commands, e-mail: users-h...@tomcat.apache.org
