On Tuesday 09 March 2010 20:49:34 Evan Daniel wrote:
On Tue, Mar 9, 2010 at 3:34 PM, Dennis Nezic denn...@dennisn.dyndns.org
wrote:
As I (and I'm sure others) mentioned before, my node is still going
down (crashing?) regularly -- roughly weekly. I currently allocate it
150MB of (precious) ram. Here are my last 2 (of infinity) heap reports
from the jvm dump:
I have filed a bug for this. Please register on the bug tracker and Monitor
Issue on this bug:
https://bugs.freenetproject.org/view.php?id=4030
The next build should be slightly improved on this, I have just fixed a bug
that was using at least 20MB of RAM for no good reason.
Heap
def new generation total 46080K, used 41323K
eden space 40960K, 100% used
from space 5120K, 7% used
to space 5120K, 0% used
tenured generation total 102400K, used 99586K
the space 102400K, 97% used
compacting perm gen total 12288K, used 11390K
the space 12288K, 92% used
ro space 10240K, 61% used
rw space 12288K, 60% used
Heap
def new generation total 46080K, used 41400K
eden space 40960K, 100% used
from space 5120K, 8% used
to space 5120K, 0% used
tenured generation total 102400K, used 102399K
the space 102400K, 99% used
compacting perm gen total 12288K, used 11214K
the space 12288K, 91% used
ro space 10240K, 61% used
rw space 12288K, 60% used
Is there NO way that freenet can do a better job cleaning up? (I
believe this happens even if I don't have (much) currently in the queues
-- Ie. if I did heavy activity days before -- or maybe even on it's own
without any manually-initiated activity, although I would have to test
this more :\.)
150M is a bit low if you have stuff queued - especially inserts. Do you still
get OOMs with *nothing* queued? (I suggest deleting or moving out of the way
your node.db4o to be sure; if you move it out of the way be careful to move
persistent-temp-* as well). Exactly what plugins are you running? Currently the
default is 192M. We're trying to get it to be auto-detected but that is not
deployed yet.
Maybe we can add more debugging info as to where all the memory is
allocated? Ie. which structures? (And hopefully decide that we can Let
Go of some of them :|.)
Here's part of what I do:
wrapper.java.additional.3=-Xloggc:freenet.loggc
(In wrapper.conf)
Then tail -f freenet.loggc. If it is constantly doing Full GC's, it is on the
edge. At which point, jmap can be very helpful (you might need the JDK):
$ jmap -histo -F pid of freenet
$ jmap -dump:live,format=b,file=dump.filename
The latter is particularly useful. You can then do:
$ jhat dump.filename
This opens a web server on 127.0.0.1:7000 on which you can investigate exactly
what is using memory. You could alternatively send me the dump, I'm not sure
whether the tools are backward compatible format-wise.
Another thing that is helpful is a stack dump (on the stats page, or run.sh
dump, or kill -QUIT pid of freenet, in any case it goes to stdout).
What plugins are you running (complete list)? Can you provide a copy
of your full stats page, in advanced mode?
Freetalk+WoT needs quite a bit of memory. XMLSpider needs absurd amounts. All
of these can be optimised substantially.
Evan Daniel
signature.asc
Description: This is a digitally signed message part.
___
Support mailing list
Support@freenetproject.org
http://news.gmane.org/gmane.network.freenet.support
Unsubscribe at http://emu.freenetproject.org/cgi-bin/mailman/listinfo/support
Or mailto:support-requ...@freenetproject.org?subject=unsubscribe
