Assertion failed: Invalid argument (src\select.cpp:111) resp. hang in zmq_ctx_destroy

[sigiesec]

Issue description

I get an assertion failure during context shutdown.

Environment

• libzmq version (commit hash if unreleased): unreleased, 7783d02
• OS: Windows 7 64-Bit

Minimal test code / Steps to reproduce the issue

1. Change count in test_get_peer_state to a larger number, e.g. 100000
2. Run test_router_mandatory

What's the actual result? (include assertion message & call stack if applicable)

Assertion failed: Invalid argument (src\select.cpp:111) with the following stacktrace:

 	KernelBase.dll!RaiseException�()	Unknown
>	libzmq-v140-mt-gd-4_2_3.dll!zmq::zmq_abort(const char * errmsg_) Line 89	C++
 	libzmq-v140-mt-gd-4_2_3.dll!zmq::select_t::rm_fd(unsigned __int64 handle_) Line 111	C++
 	libzmq-v140-mt-gd-4_2_3.dll!zmq::io_object_t::rm_fd(unsigned __int64 handle_) Line 72	C++
 	libzmq-v140-mt-gd-4_2_3.dll!zmq::stream_engine_t::unplug() Line 277	C++
 	libzmq-v140-mt-gd-4_2_3.dll!zmq::stream_engine_t::error(zmq::stream_engine_t::error_reason_t reason) Line 998	C++
 	libzmq-v140-mt-gd-4_2_3.dll!zmq::stream_engine_t::in_event() Line 327	C++
 	libzmq-v140-mt-gd-4_2_3.dll!zmq::select_t::loop() Line 315	C++
 	libzmq-v140-mt-gd-4_2_3.dll!zmq::select_t::worker_routine(void * arg_) Line 390	C++
 	libzmq-v140-mt-gd-4_2_3.dll!thread_routine(void * arg_) Line 47	C++
 	ucrtbased.dll!invoke_thread_procedure(unsigned int(*)(void *) procedure, void * const context) Line 92	C++
 	ucrtbased.dll!thread_start<unsigned int (__cdecl*)(void * __ptr64)>(void * const parameter) Line 115	C++
 	kernel32.dll!BaseThreadInitThunk�()	Unknown
 	ntdll.dll!RtlUserThreadStart�()	Unknown

reason is connection_error here

The main thread is here:

>	libzmq-v140-mt-gd-4_2_3.dll!zmq::signaler_t::wait(int timeout_) Line 268	C++
 	libzmq-v140-mt-gd-4_2_3.dll!zmq::mailbox_t::recv(zmq::command_t * cmd_, int timeout_) Line 81	C++
 	libzmq-v140-mt-gd-4_2_3.dll!zmq::ctx_t::terminate() Line 179	C++
 	libzmq-v140-mt-gd-4_2_3.dll!zmq_ctx_term(void * ctx_) Line 175	C++
 	test_router_mandatory.exe!test_get_peer_state() Line 144	C++
 	test_router_mandatory.exe!main() Line 220	C++

What's the expected result?

Successful completion of zmq_ctx_destroy

[bjovke]

@sigiesec What is "test_get_peer_state", a file, function, option? I can't seem to find it.

[sigiesec]

Sorry, test_get_peer_state is a function in test_router_mandatory.cpp

[bjovke]

@sigiesec I'm sorry, I can't seem to find it in test_router_mandatory.cpp. In which branch/commit is it?

[sigiesec]

@bjovke It's in the branch commit mentioned under "Environment" above: 7783d02

[bjovke]

@sigiesec Found it, thank you.

[bjovke]

@sigiesec I'm getting only infinite blocking wait on rc = zmq_ctx_term (ctx);, no assertion.
I'm using VS 2015, Win7, x64 build. Which Visual Studio did you use?

[sigiesec]

@bjovke Thanks a lot for having a look into this :) I am using the same compiler etc.
I am building with ZMQ_USE_SELECT, maybe you are using ZMQ_USE_POLL?

[bjovke]

@sigiesec I'm also using ZMQ_USE_SELECT, that is, option usepoll=No.
But I've pasted the whole test code of test_router_mandatory.cpp to existing project I already had setup for linking with libzmq. Maybe the difference is in compile/link options for tests build?
How did you build this test, by using CMake or in some other way?

[bjovke]

@sigiesec And also, I've ran only static runtime + static libzmq + debug build. Is this happening in some other build type?

[sigiesec]

I use the CMake Debug build, which uses the dynamic runtime and dynamic libzmq AFAIK.
If you copied only test_router_mandatory.cpp, is the rest of libzmq the same as in the branch?

[bjovke]

@sigiesec Yes, I've pulled exactly this branch. But I will double check. I didn't use CMake at all, for libzmq I've used VS solution from build directory. I've noticed before that some things might be different than CMake build.

[bjovke]

@sigiesec I've built VS solution with CMake: 64 bit debug, libzmq as DLL, dynamic runtime. I've ran test_router_mandatory.cpp with const size_t count = 1000000; and still the same result, rc = zmq_ctx_term (ctx); hangs forever.

Here are my VS project files:

libzmq.vcxproj.txt
test_router_mandatory.vcxproj.txt

[sigiesec]

well, hanging forever is not ok as well. do you have stacktraces where the threads are hanging?
i can imagine it is not deterministic, but i had assertion failures every time i tried

[sigiesec]

i can't look at the project files right now, will check them later