COM+ application package hangs - Production Crash
Srikanth Nellore
Ours is a 3 tier web application.
Tier 1 - 2 load balanced IIS/win2k servers
Tier 2 - 1 application server hosting components in COM+
environment. Components are developed using VB 6.0 Biztalk
is also installed on this BOX
Tier 3 - Oracel 8.1.6 server on sun solaris box
Problem
The application stops responding all of a sudden. When the
users try to access this application through URL the
application doesn't respond.
Whenever this situation happens we see some COMponets
hanging in the COM+ package. Tha application call time for
some of the components becomes to high.
We need to shut down and restart the COM package to make
the application work.
We have checked the code for infinite loops etc but
haven't found any.
How to debug this situation? We are not able to reproduce
this in development or test environment.
Additional information
1) We have installed two application proxies that point to
the components on application server
2) msodbc for oracle is used
3) Windows SP3 and MDAC 2.7 installed on all machines
4) We keep seeing these two error messages in the
application logs.
2002-10-12
15:25:06|SeverityError|VossCommon.clsReport.vntGetHubContra
ct|[Microsoft][ODBC driver for Oracle]Failure enlisting in
Resource Manager (0x8004D00A)|
2002-10-03
09:20:49|SeverityError|VossJob.clsKPI.vntJobDetailsPerUser|
[Microsoft][ODBC driver for Oracle][Oracle]ORA-00000:
normal, successful completion (0x80004005)
Any idea on how to debug this situation? Our production
servers has hanged more than 30 times in last 4/5 weeks.
Any help here is highly appreciated.
Rgds,
Srik
Ryan Lu [MS]
The error -2147168246/8004D00A right means the failure of MSDTC enlisting
in resource manager - Oracle in your situation.
Please try following steps for troubleshooting:
1. Q193893 INFO: Using Oracle with Microsoft Transaction Server and COM+
http://support.microsoft.com/default.aspx?scid=KB;EN-US;Q193893
2. Apply the latest patches for Oracle.
If the problem still exists, please gererate dump files and contact your
local PSS for help.
- Download Debugging Tools for Windows from
http://www.microsoft.com/ddk/debugging/ and install it on the machine.
- Run "adplus -hang -iis -o c:\adplusout" in command prompt.
- Generate dump files three times with 5 minutes interval and send all the
files.
Best regards,
Ryan Lu
Microsoft
This posting is provided "AS IS" with no warranties, and confers no rights.
Michael D. Long
The ORA-00000 is a false message; the driver layer is failing
to decode the real error. Analyze your trace files on the COM+
server. The files have the extension .trc, and they contain the
actual error detected by the Oracle Client layer.
Also, you need to involve your Oracle DBA. Tell him he needs
to monitor v$session for "Inactive" sessions. He should know
what an increasing number of these sessions indicates.
If this is load related (and I assume it is), have your DBA look
at tuning the Oracle Server configuration. In addition, a review
for appropriate indexing and poorly written SQL is in order.
FYI - sometimes the "poorly written" SQL works OK when
the Oracle Server is configured for cost-based optimizations,
but poorly for rule-based (or vice-versa). I diagnosed an issue
for one site that switched that setting on a Sunday afternoon,
and performance went from 10ms to 12,000+ on a particular
query. Obviously, this caused increased contention and DTC
timeouts, leading to session exhaustion and the error you report.
p.s. - Recommend to your DBA that he always keep a session
active against Oracle with sysadmin privileges. If all sessions
become locked in an "Inactive" state then there is no way to
recover since you can't log in. I've had to resort to a hardware
reset more than once during extreme load testing, and neither
Solaris nor Oracle were all too happy about it.
Michael D. Long
Microsoft MVP - Windows SDK
"Srikanth Nellore" <srikanth...@barwil.com> wrote in message
news:022701c27208$004916d0$9de62ecf@tkmsftngxs01...
m.shi...@gmail.com
we are also facing same issue right now. Did you guys manage to fix this issue in the past ?