At 05:45 PM 07/27/2002 -0400, Gianni Johansson wrote:

>I attached a thread dump from when the node was in the stuck state.

I've always wondered if there an easy way to make Sun's JVM do a thread 
dump into a file?  Is this easier with some other JVM?

>The most interesting follows below.  It's holding a lock on the
>routing table and the the directory while it syncs.   There are 16 other
>threads waiting on these two monitors.
>
>Do you think that this is a problem in itself, or the symptom of another
>problem? i.e. why would the sync call take very long? or why would it be hard
>to acquire the directory's lock?

I wonder if Gianni is seeing a less severe manifestation of a problem
I have been having recently. This problem happened a while back, went
away, and now it is back again. Unfortunately, I can't remember the last
Fred build it didn't happen on.

Anyway, I run my node on Linux with Sun's 1.4 JRE. Once in a while I get
anywhere from 1 to 3 threads that get stuck in state 'D' (according to
top and ps) which apparently is an uninterruptable sleep (appraently
this state is supposed to be used for disk io or something like that).
When they say uninterruptable, they mean it. Even if I kill the mother
thread of my node, these few threads survive all attempts at kill-ing
even with '-9'. These processes have my node's TCP port open according
to fuser. So if I want to start my node again, I have to reboot since I
know of no other way to kill these threads. These threads are so
uninterruptable, that the filesystem that they have files open on won't
unmount during the shutdown part of the reboot. So I always have to wait
for fsck during the startup part of the reboot to check the file system
where all my freenet files are.

This seems to be at least a bug in the JVM and/or kernel or something
that the JVM uses for file IO. But, like I said, it was there, it
disappeared and it's back again, so something apparently has changed to
trigger it. Any ideas? I'm glad Gianni was able to provide stack traces
for his situation. I'm not so lucky.

Ed


_______________________________________________
devl mailing list
[EMAIL PROTECTED]
http://hawk.freenetproject.org/cgi-bin/mailman/listinfo/devl

Reply via email to