| problems in WAS 6.1 [message #119990] |
Wed, 02 January 2008 04:16  |
Eclipse User |
|
|
|
Originally posted by: kensabir.au1.ibm.com
Hi,
I'm trying to profile on Websphere Application Server (WAS) 6.1 and
having trouble when starting up.
WAS is using JRE 1.5 (IBM version). I have had no problems profiling
earlier WAS versions using JRE 1.4x and the old -XrunpiAgent.
When i try starting up using these
arguments:" -agentlib:JPIBootLoader=JPIAgent:server=enabled;CGProf:execd etails=true ",
it takes a very long time to startup even though I haven't connected to
it yet. It takes 2-3 hours to startup with profiling enabled, instead of
15 minutes without profiling.
By doing multiple thread dumps, it appears that a thread is always stuck
in the CGProxy.MethodEnter or MethodLeave.
org/eclipse/tptp/martini/CGProxy.MethodEnter(Native Method)
org/eclipse/tptp/martini/CGProxy.EarlyMethodEnter(CGProxy.ja va:50(Compiled
Code))
java/util/zip/ZipFile.ensureOpen(ZipFile.java(Compiled Code))
java/util/zip/ZipFile.getEntry(ZipFile.java(Compiled Code))
is this expected behaviour? Since I haven't connect to the agent
controller using the workbench yet (as it is still starting up WAS), I
would have thought this MethodEnter should not be called yet.
When it does finally start, when I connect to it, it connects OK but
says 0 events get processed (even though i've set my filters up
correctly). I then try to pause the profiling and it crashes the WAS
server. In my native std err, I see "[Error: Attach failed] Unhandled
exception" and some registry data. The JVM produces a System Dump.
I am running linux 64 bit using AgentController version 4.4.1.
thanks
Kenny
|
|
|
| Re: problems in WAS 6.1 [message #120946 is a reply to message #119990] |
Wed, 16 January 2008 12:57  |
 Asaf Yaffe
Messages: 333
Registered: July 2009 |
Senior Member |
|
|
Kenny,
This is a known issue. It is related to the fact that the new JVMTI
Profiler attach/detach capabilities are not yet supported with the IBM
1.5 JVM, due to some JVM limitations (see
https://bugs.eclipse.org/202550 for a detailed discussion about this issue).
In your case, the aforementioned limitation is causing data collection
to begin immediately instead of being delayed until you connect/attach
from the workbench. This is why WAS startup takes so much time. It is
also the reason why you see
org/eclipse/tptp/martini/CGProxy.MethodEnter(Native Method) in your
thread dumps (this method is called to dispatch "method enter" events to
the profiler runtime).
When you try to attach, the system crashes because of Bug 202550
mentioned above.
There are 2 possible workarounds that you can try:
1. Launch WAS and the profiler in "controlled" mode (using
server=controlled instead of server=enabled in the -agentlib: argument).
When you do that, WAS startup will be "stuck" until you attach from the
workbench (note that Agent Controller must be running). When you attach,
make sure to define a filter that includes only your web application and
excludes everything else. This will greatly speedup WAS startup times.
2. Launch WAS and the profiler in "standalone" mode (using
server=standalone instead of server=enabled in the -agentlib: argument).
The "standalone" mode causes the profiler to work in "offline" mode (no
need for Agent Controller) and to collect data into a file, which you
can import to the workbench. Again, make sure to define a filter which
includes only your web application. Check out the online documentation
for the exact procedure of defining filters in standalone mode
( http://help.eclipse.org/help33/topic/org.eclipse.tptp.platfo rm.doc.user/tasks/teprofsa.htm).
HTH,
Asaf
--
Asaf Yaffe
Eclipse TPTP Committer, JVMTI Profiler
> Hi,
>
> I'm trying to profile on Websphere Application Server (WAS) 6.1 and
> having trouble when starting up.
>
> WAS is using JRE 1.5 (IBM version). I have had no problems profiling
> earlier WAS versions using JRE 1.4x and the old -XrunpiAgent.
>
>
> When i try starting up using these
> arguments:" -agentlib:JPIBootLoader=JPIAgent:server=enabled;CGProf:execd etails=true ",
> it takes a very long time to startup even though I haven't connected to
> it yet. It takes 2-3 hours to startup with profiling enabled, instead of
> 15 minutes without profiling.
>
> By doing multiple thread dumps, it appears that a thread is always stuck
> in the CGProxy.MethodEnter or MethodLeave.
>
> org/eclipse/tptp/martini/CGProxy.MethodEnter(Native Method)
> org/eclipse/tptp/martini/CGProxy.EarlyMethodEnter(CGProxy.ja va:50(Compiled
> Code))
> java/util/zip/ZipFile.ensureOpen(ZipFile.java(Compiled Code))
> java/util/zip/ZipFile.getEntry(ZipFile.java(Compiled Code))
>
> is this expected behaviour? Since I haven't connect to the agent
> controller using the workbench yet (as it is still starting up WAS), I
> would have thought this MethodEnter should not be called yet.
>
> When it does finally start, when I connect to it, it connects OK but
> says 0 events get processed (even though i've set my filters up
> correctly). I then try to pause the profiling and it crashes the WAS
> server. In my native std err, I see "[Error: Attach failed] Unhandled
> exception" and some registry data. The JVM produces a System Dump.
>
> I am running linux 64 bit using AgentController version 4.4.1.
>
> thanks
> Kenny
>
|
|
|
Powered by
FUDForum. Page generated in 0.03064 seconds
] 
Search
Help
Register
Login
Home
