Oops. It appears the standard heap dump toosl don't really dump permgen, so that's not going to get us anywhere.
I usually do this in jprofiler, maybe someone else has a suggestion :) Kristian 2013/6/26 Kristian Rosenvold <[email protected]>: > Funny in a sort of ironic way, permgen is noticeably better in 3.1 for > my usecases :) > > The simplest way to get (my) attention to this issue is to create 2 > heap dumps of your maven process, one after "some time" and the other > just before it runs out of permgen. > > ("some time" is supposed to be well into the execution so I can avoid > tracking all the *correct* permgen usage that happens at the start of > a maven build as it loads the plugins) > > You can send them to me via something like dropbox or other means that handles > large files. > > Use jvisualvm or one of the other tools to get the heap dumps. > > Kristian > > > 2013/6/26 Jörg Schaible <[email protected]>: >> Hi Jason, >> >> Jason van Zyl wrote: >> >>> Doesn't see to be a whole lot of activity around the 3.1.0-alpha-1 so I >>> plan to cut the 3.1.0-beta-1 this weekend if there are no objections. >> >> Apart from the reported bogus build with snapshots (MNG-5207) it seems M31 >> has a major problem with PermGen space leakage. >> >> We have currently a build that contains 337 projects. With M221 I can build >> all of it in one run in ~11 minutes. >> >> With M305 the build runs faster, but I have to continue it two times with "- >> rf" option, because it runs out of PermGen space. >> >> With M31a the leakage is really worse. The build stops because of PermGen >> space 4 times, where I have to continue manually again. >> >> All plugins are locked down, so it has to be related with Maven core. >> MAVEN_OPTS is "-Xmx1g -XX:MaxPermSize=192m" and I am running "mvn clean >> install [-rf <name>]". >> >> Thanks, >> Jörg >> >> >> --------------------------------------------------------------------- >> 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]
