Some Eclipse Foundation services are deprecated, or will be soon. Please ensure you've read this important communication.

Bug 159604

Summary:

piagent seg faults when running in application server

Product:

z_Archived

Reporter:

Ashish Patel <ashishp>

Component:

TPTP

Assignee:

Samson Wai <samwai>

Status:

CLOSED WONTFIX

QA Contact:

Severity:

critical

Priority:

CC:

jkubasta

Version:

unspecified

Keywords:

plan

Target Milestone:

---

Hardware:

OS:

Windows XP

Whiteboard:

Attachments:

Description	Flags
Core dump 1	none
Core dump 2	none
The regular binary of the piagent.	none
A debug version of the piagent	none

Description Ashish Patel

2006-10-03 11:13:37 EDT

I have a simple scenario where there are two jvmpi agents, one of which is the pi agent, that runs in WAS 6.  When running a large load test against the application server, the piagent seg faults and causes the application server to crash.  I have attached the core dump files.  

This only seems to happen when running on linux. Running the two agents on Windows seems to be ok.

This defect is needed to be fixed for a consuming product.

Comment 1 Ashish Patel

2006-10-03 11:15:20 EDT

Created attachment 51341 [details]
Core dump 1

Comment 2 Ashish Patel

2006-10-03 11:15:44 EDT

Created attachment 51342 [details]
Core dump 2

Comment 3 Bob Duncan

2006-10-03 12:08:20 EDT

This problem involves a pass-thru agent that sits on the JVMPI interface and broadcasts JVMPI events to two actual JVMPI agents, one of which is the TPTP piAgent. The scenario works fine (including the pass-thru behavior) as long as piAgent is the only jvmpi agent being serviced by the pass-thru agent.

Comment 4 Bob Duncan

2006-10-27 12:30:35 EDT

Re-targeting for 4.2.1 patch (consuming product will not be picking up 4.3).

Comment 5 Harm Sluiman

2006-11-20 14:57:53 EST

Please retarget this defect asap. 4.2.1 was clsoed a long time ago

Comment 6 Bob Duncan

2006-11-23 17:17:55 EST

Retargetting to 4.4 until I can diagnose and assess a fix for another, open target. Working on this currently.

Comment 7 Bob Duncan

2006-12-05 19:55:45 EST

Ahish,

From what I am seeing so far, I would've expected the following message to have appeared in the WAS console or the native error log or some such place (wherever WAS captures stderr for native code) just before the Segv occurred:
   "FATAL ERROR:  Out of memory...JVM Exiting"

Can you find that message anywhere after this failure? If so, I have an idea of what the problem might be (ie: why you get a Segv instead of a graceful exit for an OutOfMemory condition).  If not, I have more digging to do.

Comment 8 Ashish Patel

2006-12-14 18:22:08 EST

Created attachment 55728 [details]
The regular binary of the piagent.

Comment 9 Ashish Patel

2006-12-14 18:22:35 EST

Created attachment 55729 [details]
A debug version of the piagent

Comment 10 Ashish Patel

2006-12-14 18:23:25 EST

Attached two binaries from the piagent based on the following note:

Hi Ashish,

Thanks for your patience.  I am hitting some road-blocks with a couple of things I've been trying on this (confirming the exact point of failure, repro'ing with a smaller testcase, testing the jvmpi ProfilerExit call that may be implicated, etc). Let's see if we can at least get a proper traceback with a debug version and gdb.  I've attached two versions of the piAgent lib: one vanilla/reg and the other debug. 



Let's try two runs:

Replace the libpiAgent.so in the <rac_dir>/lib with the attached libpiAgent.so.reg (minus the ".reg" suffix) and reproduce the problem
Replace the libpiAgent.so in the <rac_dir>/lib with the attached libpiAgent.so.dbg (minus the ".dbg" suffix), attach WAS process from gdb, and reproduce the problem


Thanks,
Bob

Comment 11 Bob Duncan

2006-12-14 18:38:55 EST

Note that the second run (with the debug piAgent and attaching gdb to the AppServer/profiled process) is expected to produce the precise traceback we need to nail the point of the Segv.  That's what we need at this point.

Comment 12 Samson Wai

2007-01-31 16:09:41 EST

Set priority to P1 for 4.4 plan closure.

Comment 13 jkubasta

2007-02-04 09:14:32 EST

Please indicate the Java version being used.

Comment 14 Ashish Patel

2007-02-05 07:41:11 EST

Java version being used is IBM JDK 1.4.2 (as bundled with WAS 6).

Comment 15 jkubasta

2007-10-18 14:10:08 EDT

No plan to fix

Comment 16 Ashish Patel

2007-10-25 16:43:01 EDT

cls