No activities for 9 secs (WL 8.1)

Question

We are running 8.1 on Solaris with 1.4.1_02, hotspot and -ms1024m -mx1024m. The
system gets a restart daily in the morning.
It happened at least once (this afternoon) that for a period of 9 secs, there
were no activities logged in the application log but we know there were many requests
coming from both JMS and RMI. Since requests come from very different sources,
we can pretty much conclude the problem is with the server itself. The system
appeared to be running fine afterwards.
Here is some more potentially useful info. We have about 300 SLSBs deployed, along
with ~120 MDBs. In config.xml, we have <ExecuteQueue Name="default" ThreadCount="50"/>.
I'm going to turn on -verbose:gc but since this doesn't seem to happen every day,
I hope someone can provide some pointers.
TIA,
Bill

Answer 1

Hi Rob, Caught one hang today and did a dump. However, as I posted in another thread
("Big full GC"), it seems to be pretty clear now that the "hanging" is caused
by GC. If you feel it would be worthwhile to continue along this (dump) direction,
I'll post the dump here. Thanks. /Bill
"Bill" <[email protected]> wrote:
The article is helpful. We'll try to catch it when it happens and get
a
thread dump. Thanks much!
"Rob Woollen" <[email protected]> wrote in message
news:[email protected]...
Bill wrote:
The server will then be down.A thread dump doesn't kill the server. It just has the jvm dump outthe
current stack of all the threads.
I can periodically query various parameters of
the server and system. Do you have any guesses that might provide
some
focus?Unfortunately without knowing more about your app, it's hard for meto
guess. It sounds like a deadlocked system (which is why a thread dump
would be so helpful.)
I wrote an article a while back that might be helpful:
http://dev2dev.bea.com/products/wlserver/articles/496.jsp
-- Rob
Thanks,
Bill
PS - Regards to thread dump. I can continue to provide service if
I put
it
in a cluster, assuming only one will get into that "mode". But thenthe
load
will be different and it's possible that it wont happen under thereduced
load. In fact it will be in a cluster but I'd like to resolve thisfirst.
"Rob Woollen" <[email protected]> wrote in message
news:[email protected]...
My best recommendation is to take a thread dump when the server appears
to be inactive or not responsive.
-- Rob
Bill wrote:
We are running 8.1 on Solaris with 1.4.1_02, hotspot
and -ms1024m -mx1024m. The
system gets a restart daily in the morning.
It happened at least once (this afternoon) that for a period of
9 secs,
there
were no activities logged in the application log but we know there
were
many requests
coming from both JMS and RMI. Since requests come from very differentsources,
we can pretty much conclude the problem is with the server itself.
The
system
appeared to be running fine afterwards.
Here is some more potentially useful info. We have about 300 SLSBsdeployed, along
with ~120 MDBs. In config.xml, we have <ExecuteQueue Name="default"ThreadCount="50"/>.
I'm going to turn on -verbose:gc but since this doesn't seem to
happen
every day,
I hope someone can provide some pointers.
TIA,
Bill