| Summary: | Reactive Flushing Strategy race condition | ||
|---|---|---|---|
| Product: | TAO | Reporter: | Carlos O'Ryan <coryan> |
| Component: | ORB | Assignee: | Carlos O'Ryan <coryan> |
| Status: | RESOLVED FIXED | ||
| Severity: | blocker | ||
| Priority: | P3 | ||
| Version: | 1.1.14 | ||
| Hardware: | All | ||
| OS: | All | ||
| Bug Depends on: | |||
| Bug Blocks: | 132, 296 | ||
This one is mine. Until this bug is fixed 132 should be open. This bug would probably take 1 week, maybe 2. I think I can start work on it by early next week (i.e. May, 14th) I thought I had accepted this bug already. The bug was fixed last week. At least the one test that used to hang due to this problem does not anymore. There are some other bugs that we are trying to track down, I suspect they will require some minor fixes, should they not we will re-enter the bugs as needed. |
The reactive flushing strategy flushes a transport using: TAO_ORB_Core *orb_core = transport->orb_core (); while (!transport->queue_is_empty ()) { int result = orb_core->run (0, 1, ACE_TRY_ENV); ACE_TRY_CHECK; if (result == -1) return -1; } unfortunately this loop has a race condition. The queue could be cleaned up by another thread, before we can enter the event loop in orb_core->run(). Currently the problem only shows up when using AMI requests, for example, $TAO_ROOT/tests/LongUpcalls. A simple workaround is to use the Block_Flushing_Strategy. The only solution that I can think of right now involves adding more complexity to the Leader/Followers code, when the ORB flushes a transport it would need to follow the same rules that "client" threads follow when they join the group. But such a change is too complicated for me to take on at this point.