On 03.03.2010 14:35, Felix Meschberger wrote: > Hi, > > On 03.03.2010 14:20, Philipp Koch wrote: >> why not persisting the dump in the repository with timestamps, >> information about memory stats and other useful info when the dump got >> triggered. >> if we do so, the dump would be accessible within sling pretty easily over >> http. > > That would (theoretically) something to consider, but: > > * The VM writes to the filesystem anyway (not a stream, a file) > * In an OOME situation it might not be possible to write to the repo > > Plus, the dumps are available via http of course, because the plugin > provides them.
Plus, the dumps could be made available through Sling using the FS Resource Provider. So no real need to push the dumps into the repository. In addtion, if the system is running in a cluster, storing the dumps in the repository, might create a number of follow-up issues to solve. Regards Felix > > Regards > Felix > >> >> regards, >> philipp >> >> On Wed, Mar 3, 2010 at 2:14 PM, Felix Meschberger <fmesc...@gmail.com> wrote: >>> Hi, >>> >>> On 03.03.2010 13:59, Marcel Reutegger wrote: >>>> hi, >>>> >>>> On Tue, Mar 2, 2010 at 00:06, Felix Meschberger <fmesc...@gmail.com> wrote: >>>>> Hi all, >>>>> >>>>> As of Java 5, the platform provides a big number of information about >>>>> the memory state for the virtual machine. >>>>> >>>>> I have hacked together a Web Console plugin to display the overall >>>>> memory use as well as usage of the different memory pools. The plugin >>>>> also allows for creating heap dumps and for downloading these heap dumps. >>>> >>>> nice. >>>> >>>>> In addition, the plugin ensures, any existing heap dumps are also >>>>> included in the ZIP download of the Configuration Status page. >>>>> >>>>> Finally, the plugin allows for setting a percentage threshold (50-100%) >>>>> of memory usage (Perm Gen and Old Gen). As soon as memory consumption >>>>> crosses the threshold a heap dump is automatically generated (and yes, >>>>> if the threshold is repeatedly crossed, a dump is generated each time >>>>> the threshold is crossed, not just the first time). >>>> >>>> is that configurable? if things go wrong, an application will probably >>>> throw OOME multiple times. that would quickly fill the disk. >>> >>> Not yet. >>> >>> But I am thinking of adding a switch to prevent dumps from being written >>> in too close succession. For example setting this to "1d" would write a >>> dump only once a day. >>> >>> This could also be used to specify "only dump once and then ignore any >>> further events". >>> >>> Regards >>> Felix >>> >>>> >>>> regards >>>> marcel >>>> >>>>> The plugin requires a trunk build of the Apache Felix Web Console and is >>>>> available from my whiteboard at [1]. >>>>> >>>>> Enjoy and tell me what you think. >>>>> >>>>> Regards >>>>> Felix >>>>> >>>>> [1]http://svn.apache.org/repos/asf/sling/whiteboard/fmeschbe/memoryusage >>>>> >>>> >>> >>
