Re: [orion-dev] runaway server

[John Arthorne <John_Arthorne@xxxxxxxxxx>]

> "John J. Barton" <johnjbarton@xxxxxxxxxxxxxxx>

> I am again seeing the pattern I noticed yesterday:
8-10 secs of full 
> CPU, 22-25 secs off.  I did not see it yesterday after restarting
the 
> server; I saw it first thing to day before I did anything with the
> server. I wonder if the problem is related to suspending the processor
> while the server is up? (Windows "sleep" mode"). That
would explain why 
> this was not noticed by server developers who might be focused on
> machines which don't suspend often.

It could be. I use sleep all the time when a local
server is running and haven't seen this yet though. The best bet is to
launch the server using Java 6 using -Dcom.sun.management.jmxremote. This
adds minimal overhead to the server, but allows launching jvisualvm to
capture thread dumps and/or profile the running app. For more details see:

http://wiki.eclipse.org/How_to_report_a_deadlock#Using_jvisualvm

The only thing that comes to mind that fits the CPU
usage pattern is Lucene indexing, but getting a trace at the time of heavy
CPU would be a huge help in narrowing that down. I presume there is nothing
in your server error log at the time of heavy CPU (eclipse/serverworkspace/.log)?

John