[ale] load issue

[Christopher Fowler]

I had a problem like this on FC2 and ran into an interesting situation. 
My load would be at 2+ and could rise but top showed nothing.  I was
perplexed.  Here the kernel said some processes were using up CPU and
top could not list those.  The culprit was a few threads in tomcat.

Apparently we had some code written that did not understand that when
read() returns 0 it means the other end has terminated the connection. 
Therefor Java IO was looping as fast as possible because it thought it
had something to read on a socket.  read() would return 0 and it would
keep doing this.  After a few of these threads the system would go to a
crawl.  Strace was the _only_ way I found this.  I killed a few
processes trying to see which one was a problem and then I attached to
java and saw what was going on.  Restart tomecat and load went to 0.00. 
It may be the way the FC2 kernel handled threads that confused top?  I
do not know.  The problem was that the JVM running tomcat was eating all
my CPU and top was useless in telling me this information.


On Thu, 2006-06-22 at 06:02, David Corbin wrote:
> On Thursday 22 June 2006 12:53 am, Benjie wrote:
> > do you have any zombie, or otherwise dead processes?  I have seen my
> > load average go up mysteriously ( no real cpu load) when there was a
> > dead process that had a file locked.
> 
> What's the best way to tell?
> 
> David
> _______________________________________________
> Ale mailing list
> Ale at ale.org
> http://www.ale.org/mailman/listinfo/ale