| Summary: | Temporary deadlocks in multi-threaded TAO application | ||
|---|---|---|---|
| Product: | TAO | Reporter: | milan.cvetkovic |
| Component: | ORB | Assignee: | DOC Center Support List (internal) <tao-support> |
| Status: | NEW --- | ||
| Severity: | normal | ||
| Priority: | P3 | ||
| Version: | 2.2.1 | ||
| Hardware: | All | ||
| OS: | Linux | ||
| URL: | https://groups.google.com/d/topic/comp.soft-sys.ace/LubMZ2EG7bo/discussion | ||
| Attachments: |
client.cpp
client.cpp Bug_4128_Regression.mpc MyInterfaceImpl.h MyInterfaceImpl.cpp Test.idl server.cpp README |
||
Created attachment 1473 [details]
client.cpp
Created attachment 1474 [details]
Bug_4128_Regression.mpc
Created attachment 1475 [details]
MyInterfaceImpl.h
Created attachment 1476 [details]
MyInterfaceImpl.cpp
Created attachment 1477 [details]
Test.idl
Created attachment 1478 [details]
server.cpp
Created attachment 1479 [details]
README
Couple of notes: - If there is no relative timeout policy in client orb, the RPC does not complete until sleep in the other thread is complete. - if sleep() is preceded by: orb_->orb_core()->lf_strategy().set_upcall_thread ( orb_->orb_core()->leader_follower()); RPC returns correctly with CORBA::TRANSIENT. - If number of threads is decreased to 1, and only RPC timer started, without lf_strategy code above, RPC also returns correctly with CORBA::TRANSIENT |
Created attachment 1472 [details] client.cpp TAO VERSION: 1.2.1 ACE VERSION: 6.2.1 HOST MACHINE and OPERATING SYSTEM: Linux debian 6.0 on amd64 COMPILER NAME AND VERSION: g++-4.4.5 $ACE_ROOT/ace/config.h FILE: config-linux.h THE $ACE_ROOT/include/makeinclude/platform_macros.GNU FILE: platform-linux.GNU CONTENTS OF $ACE_ROOT/bin/MakeProjectCreator/config/default.features: unmidified AREA/CLASS/EXAMPLE AFFECTED: LF_Strategy DOES THE PROBLEM AFFECT: EXECUTION: Yes SYNOPSIS: In 2 threaded application, we experience intermittent CORBA::TIMEOUT exceptions from a thread invoking RPC call, while other thread is busy. DESCRIPTION: I have 2 threads, both of them running orb->run() in TP_Reactor. Main task of the application is executing every 5 minutes, and it run by a timer. Application runs client RPC every 1 minute to another CORBA server, which normally returns immediately. Before invoking RPC, this thread invokes set_upcall_thread(). What I am seeing is that this RPC throws CORBA::TIMEOUT. It does not appear that remote CORBA server is busy at this time. Here is code layout: int Obj1::handle_timeout (...) { // very long op, perhaps 2minutes (1) ACE_OS::sleep (20); } int Obj2::handle_timeout (...) { try { remote->ping(); } catch (CORBA::TIMEOUT&) { // the RPC comes back with TIMEOUT } } It appears that the thread which executes RPC is not capable of handling the response. If I put this line after comment (1) orb->orb_core()->lf_strategy().set_upcall_thread ( orb->orb_core()->leader_follower()); I get the expected behavior - if server is up, I get response, if not, I get immediate CORBA::TRANSIENT Attached is the sample code exposing the problem