| Summary: | Problem in LoadManager | ||
|---|---|---|---|
| Product: | TAO | Reporter: | Johnny Willemsen <jwillemsen> |
| Component: | Load Balancer | Assignee: | DOC Center Support List (internal) <tao-support> |
| Status: | ASSIGNED --- | ||
| Severity: | normal | ||
| Priority: | P3 | ||
| Version: | 1.5.1 | ||
| Hardware: | x86 | ||
| OS: | Windows NT | ||
I recall someone from OCI, Chad perhaps, suggesting using the WFMO reactor on Windows to work around the problem. problems still happens, added an ACE_ERROR_RETURN instead of -1 so that on the test output it is visible why things don't run |
I am looking why some tests are failing and I see that the LoadBalancer test orbsvcs/tests/LoadBalancing/GenericFactory/Application_Controlled is failing in all tests. I have started a debugger and I see a problem in orbsvcs/LoadBalancer/LoadManager.cpp. There I see the following code: // Activate/register the signal handler that (attempts) to // ensure graceful shutdown of the LoadManager so that remote // resources created by the LoadManager can be cleaned up. TAO_LB_Signal_Handler signal_handler (orb.in (), root_poa.in ()); if (signal_handler.activate () != 0) return -1; // @@ There is a subtle race condition here. If the signal // handler thread shuts down the ORB before it is run, the // below call to ORB::run() will throw a CORBA::BAD_INV_ORDER // exception. orb->run (ACE_ENV_SINGLE_ARG_PARAMETER); ACE_TRY_CHECK; The signal handler is tried to be activated using the ACE_TP_Reactor and this returns a notsup for registering the handler. This then results in the fact that the LoadManager is not started and the test fails. When I just comment out the check then things start and I get another error related to PortableGroup. I have changed something in that library during my refactoring so I will debug this further, but can you have a look at this? How do you want to handle this, shall I make a bugzilla entry?