Bug 1869

Summary: AMI doesn't survive failed invocations
Product: TAO Reporter: Ulrich Becker <bec>
Component: AMIAssignee: DOC Center Support List (internal) <tao-support>
Status: NEW ---    
Severity: major    
Priority: P3    
Version: 1.4.1   
Hardware: x86   
OS: Linux   
URL: http://www4.cs.fau.de/~ubecker/AMI_bug.tar
Attachments: Test case that shows failure of AMI while SMI works

Description Ulrich Becker 2004-07-07 07:39:30 CDT 
The problem occurs with AMI invocations to a server with a persistent IOR. After
one invocation failed because the server went down, all following invocations
after a server restart fail silently. But when I use SMI instead, all
invocations after a server restart return successfully.

Reproducy by: I've written a simple example to reproduce the problem -
sourcecode available under the URL above. Just adapt the ACE_ROOT in the
Makefile, then:
execute ./runtest_ami.sh to build and run the AMI version
execute ./runtest_smi.sh to build and run the SMI version

The testscript starts the server and the client. After 5 seconds, it kills the
server. The client gets an exception and prints "failed" with both versions.
After another 5 seconds, the server is restarted. With the SMI version, all
following invocations succeed again. With the AMI version, all following
invocations fail silently (neither exception nor successfull return).

Btw, I discovered similary behaviour with oneway invocations in the past.

Workaround: None, except to use many threads and SMI to achieve "non-blocking"
behaviour, but i wanted to get around this by using AMI...
Comment 1 Ulrich Becker 2004-07-08 09:11:09 CDT 
Created attachment 275 [details]
Test case that shows failure of AMI while SMI works
Comment 2 Nanbor Wang 2004-07-11 06:46:32 CDT 
Assigning the bug to me. Should be able to take a look. If this misses this 
beta, we will surely get it in for the next
Comment 3 Torsten Kuepper 2005-01-26 08:14:17 CST 
corbaloc url is used by the client. 
Obviously AMI can't yet deal with location forward reply.
Comment 4 Ulrich Becker 2005-02-01 04:36:32 CST 
I verified that this problem is related to corbaloc URLs. I modified the
testcase to use a stringified IOR, and this works.
Comment 5 Johnny Willemsen 2006-01-13 07:25:17 CST 
Fri Jan 13 13:24:12 UTC 2006  Johnny Willemsen  <jwillemsen@remedy.nl>

        * tests/Bug_1869_Regression/*:
          Added this regression to cvs, no fix yet and regression could
          be improved more to give more output to the user
Comment 6 Nanbor Wang 2006-08-08 01:38:12 CDT 
Changed component to AMI
Comment 7 Johnny Willemsen 2007-09-20 03:09:00 CDT 
to pool