PermGen problem jmap/jhat leads to dead end
Thomas.Rodriguez at Sun.COM
Mon Dec 14 16:10:14 PST 2009
Could you get the hotspot.log from a run with -XX:+UnlockDiagnosticVMOptions -XX:+LogCompilation and send it to me? It's likely to be big but it will include information about what's compiled and what's still in the code cache which might explain what's taking up all the space.
If you increase the size of the code cache using -XX:ReservedCodeCacheSize=128m do you still run out of space? I'm wondering if it's a problem with a large steady state size or if it's really a leak of some sort. I guess it's possible that's there something we missed in the fix for 4957990 which is causing code to be kept around.
On Dec 14, 2009, at 2:51 PM, Keith McNeill wrote:
> Yep, in test only.
> So using JDK 7 the Perm Gen problem seems to be fixed. PermGen went up & down with loading and unloading our our classes. Is that fix going to be back ported to 1.6?
> But, I'm now getting code cache problems (I assume). After a few rounds when I try to reload my test class I get OutOfMemoryErrors. Using jconsole: PermGen is fine, heap is fine, but CodeCache appears to be full:
> Used: 49,060 kbytes
> Committed: 49,152 kbytes
> Max: 49,152 kbytes
> Y. Srinivas Ramakrishna wrote:
>> On 12/14/09 14:21, Y. Srinivas Ramakrishna wrote:
>>> Keith, the simplest thing for you to do (if on Solaris
>>> or Linux) might be to simply pick up the relevant libjvm.so
>>> from JDK 7 and drop it into your 6uXX installation
>>> and run with it.
>> Just so there is no misunderstanding; i meant that only as a test.
>> Not for production loads!!
>> -- ramki
More information about the hotspot-dev