"Christopher Schultz" <ch...@christopherschultz.net> schrieb:
>-----BEGIN PGP SIGNED MESSAGE----- >Hash: SHA1 > >Mark, > >On 5/3/2011 10:52 AM, Mark Thomas wrote: >> On 03/05/2011 15:46, André Warnier wrote: >>> Mark Thomas wrote: >>>> On 03/05/2011 15:22, Christopher Schultz wrote: >>>>> All, >>>>> >>>>> Moments ago in our development environment, our webapp suffered an >OOME >>>>> after many re-reployments (we know we have an undeploy-related >leak). >>>>> When attempting to bounce Tomcat, the shutdown failed and I took a >>>>> thread dump which included the one non-trivial thread shown below. >>>>> >>>>> I haven't looked at the code yet to see if there is some kind of >loop >>>>> around this code, but top reported 100% CPU usage from this >process so I >>>>> suspect something like that was happening. >>>>> >>>>> We are using TC 6.0.32 on Debian Lenny and Sun/Oracle 32-bit >>>>> 1.6.0_22-b04 Server VM. >>>>> >>>>> This is less of a "help" request as it is an observation of an >>>>> unfortunate situation: the OOME is clearly the real problem and >has >>>>> nothing to do with Tomcat itself. Unfortunately, after the OOME, >Tomcat >>>>> was unable to shut down gracefully and that could be a problem in >and of >>>>> itself. >>>> >>>> After an OOME there are no guarantees as to the state of the JVM. >That >>>> Tomcat is unable to shut down cleanly in that case is not >suprising. > >Chuck's assertion is that an OOME isn't fatal to the JVM but almost >always fatal to the application since it's unlikely equipped to handle >the bizarre situations that arise from it's occurrence. A bit of a >hair-split if you ask me, but it's always worth examining any options >TC >has to at least be able to stop itself gracefully. > >>> Do I not remember seeing somewhere, at some point, a parameter named >>> "OOMParachute" or similar ? >> >> Yes, in the NIO connector. It may, or may not, provide the JVM with >> enough memory to exit gracefully. > >Since this is a PegmGen issue, it's unlikely that the NIO connector's >OOM parachute will help, here. Also, I'm not using the NIO connector :) > >Any reason the parachute is in the /connector/ and not somewhere more >universal? > >I was mostly interested in why I was getting 100% CPU usage... checking It could've been the garbage collector itself. Just yesterday I had an oom on permgen, which lead to a major-gc loop. I could see the loop nicely as I had enabled logging of gc activity with -Xloggc:$CATALINA_BASE/logs/gc.log -XX:+PrintGCDetails -XX:PrintGCTimeStamps Regards Felix >the code, I *do* see a loop in >WebappClassLoader.clearReferencesStaticFinal but it's over an iterator >that will presumably run out of entries. The CPU spike seems to be >happening within the native code in java.lang.Class.getDeclaredFields, >so there's nothing TC can do to prevent it. :( > >- -chris >-----BEGIN PGP SIGNATURE----- >Version: GnuPG v1.4.10 (MingW32) >Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ > >iEYEARECAAYFAk3AWvEACgkQ9CaO5/Lv0PAodwCfUxrRVV0TZihjiS1d6QwF1LCo >8PgAn3jOrCy4IvUTe4diJ68gGrqIyQXZ >=Ys5e >-----END PGP SIGNATURE----- > >--------------------------------------------------------------------- >To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org >For additional commands, e-mail: users-h...@tomcat.apache.org --------------------------------------------------------------------- To unsubscribe, e-mail: users-unsubscr...@tomcat.apache.org For additional commands, e-mail: users-h...@tomcat.apache.org
