Cannot release, possible deadlock when exit ...

[chenscottus]

Stuck in ~thread_pool() => threads_[i].join();

145 pthread_join_common.c: No such file or directory. #0 __pthread_clockjoin_ex (threadid=281473323018496, thread_return=0x0, clockid=0, abstime=, block=) at pthread_join_common.c:145 #1 0x0000aaaaaadded30 in std::thread::join() () #2 0x0000aaaaaabba198 in std::jthread::join() (this=0xffff9804fe90) at /usr/include/c++/11/thread:160 #3 std::jthread::~jthread() (this=0xffff9804fe90, __in_chrg=) at /usr/include/c++/11/thread:130 #4 dp::thread_pool<std::function<void ()>, std::jthread>::~thread_pool() (this=0xffff9e7c6c58, __in_chrg=) at /home/dev/projects/vcam/project/Tracker/../common/thread_pool/thread_pool.h:

Thanks!

-Scott

[chenscottus]

If I change the code to, it will like work.

		int threads_size = 0;

		threads_size = threads_.size();
		for (std::size_t i = 0; i < threads_size; ++i) {
			threads_[i].request_stop();
		}

		threads_size = tasks_.size();
		for (std::size_t i = 0; i < threads_size; ++i) {
			tasks_[i].signal.release();
		}

		threads_size = threads_.size();
		for (std::size_t i = 0; i < threads_size; ++i) {
			//threads_[i].request_stop();
			//tasks_[i].signal.release();
			if (threads_[i].joinable() == true)
			{
				threads_[i].join();
				//threads_[i].~jthread();
			}
		}

[DeveloperPaul123]

Thanks for the report.

Can you post what your task looks like that you're running on the thread pool?

[chenscottus]

dp::thread_pool<>* dp_pool_vars; ... dp::thread_pool dp_pool(32); dp_pool_vars = &dp_pool; ... dp_pool_vars->enqueue(write_frames_original_task, &wf_original_args_main); ... ... dp_pool.~thread_pool(); ....

[DeveloperPaul123]

Few things:

• Why are keeping a pointer to the thread pool?
• When the thread pool goes out of scope the deconstructor is automatically called. There's not need to call the thread pool dtor directly.
• Can you post the actually code that's doing the work? The lock may be due to that code.

[chenscottus]

1. I need to use the thread pool in different classes and function. I use only one thread pool. And I need to declare it first and later, initialize it. That is why I need a pointer
2. correct: When the thread pool goes out of scope the deconstructor is automatically called. There's not need to call the thread pool dtor directly.
3. my project is too big, the zip of the source code is more than 1MB.
4. I just use enqueue function to run: dp_pool_vars->enqueue(write_frames_original_task, &wf_original_args_main);
5. I have been runing a few days, there is no problem.
6. The only problem is when the app finish, it need to release the resources.

[DeveloperPaul123]

Okay no problem, I see.

One possible issue is that you're using enqueue() but you're not using the returned future. Consider using enqueue_detach().

[DeveloperPaul123]

Any updates on this issue? Are you still facing problems?

[chenscottus]

Thanks, no problem now!

On Fri, Apr 26, 2024 at 8:12 AM Paul T @.***> wrote:

Any updates on this issue? Are you still facing problems?

— Reply to this email directly, view it on GitHub https://github.com/DeveloperPaul123/thread-pool/issues/52#issuecomment-2079586392, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABHZOKZQLK2D34EZHZDEJHLY7JVFDAVCNFSM6AAAAAA37D7B56VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANZZGU4DMMZZGI . You are receiving this because you authored the thread.Message ID: @.***>