HTTPServerConnection::onServerStopped does not halt poll

[micheleselea]

When you call httpServer->stoppAll(true) we end up in
HTTPServerConnection::onServerStopped -> with abortConnection true
if the http connection is in keepalive waiting for socket().poll(_keepAliveTimeout, Socket::SELECT_READ)
on windows you are waiting for epoll_wait that call GetQueuedCompletionStatusEx
the .close() done in onServerStopped is not enough to exit without waiting for keepalive timeout occur

[micheleselea]

I tried a couple of changes but the only way I found rigth now is to change
bool HTTPServerSession::hasMoreRequests()
for doing something like

#if defined(_WIN32 )&&defined(POCO_HAVE_FD_EPOLL)
		Poco::Timestamp tsKeepAlive;
		do {
			if (!connected()) return false;
			if ((buffered() > 0) || socket().poll(Poco::Timespan(0, 250000), Socket::SELECT_READ)) return true;
		} while (!tsKeepAlive.isElapsed(_keepAliveTimeout.totalMicroseconds()));
		return false;
#else
		return (buffered() > 0) || socket().poll(_keepAliveTimeout, Socket::SELECT_READ);
#endif

[micheleselea]

An alternative way to handle that is to do this
piscisaureus/wepoll#20

[micheleselea]

I think that this function
PostQueuedCompletionStatus(_epollfd, 0, 0, NULL);
can work in PollSet too instead using a tcp listening connection

[lovelycheep]

i had the same problem with version 1.10.1.

Here is the crash stack:

I tried to simulate the stack, but the position was never right.

Does anyone know how to analyze and solve this problem

[micheleselea]

@lovelycheep I think the best solution is to do
piscisaureus/wepoll#20
so cheanging PollSet wakeUp like this
so you can avoid creating _eventfd

	void wakeUp()
	{		
#ifdef WEPOLL_H_
		if (_port > 0) {
			StreamSocket ss(SocketAddress("127.0.0.1", _port));
		}
		else if (_epollfd > 0) {
			PostQueuedCompletionStatus(_epollfd, 0, 0, NULL);
		}
#else
		uint64_t val = 1;
		// This is guaranteed to write into a valid fd,
		// or 0 (meaning PollSet is being destroyed).
		// Errors are ignored.
		write(_eventfd, &val, sizeof(val));
#endif
	}

[aleks-f]

This is likely addressed by this fix

But, generally speaking, code that is reproducing the issue is the best way to report and get it fixed - create a test case that fails send a pull.

[aleks-f]

I think that this function PostQueuedCompletionStatus(_epollfd, 0, 0, NULL); can work in PollSet too instead using a tcp listening connection

it doesn't

testPollSetWakeUp:
Net\testsuite\binA64\TestSuited.exe (process 15108) exited with code -1073741819.

[micheleselea]

I managed to work the PostQueuedCompletionStatus working, so no problem, I already fix my code, if you think is not the correct way to handle it, never mind

[vm2mv]

@aleks-f, are you planning to fix this problem?
I'm forced to use POCO without POCO_HAVE_FD_EPOLL now, but I don't like this option.

[aleks-f]

@vm2mv the solution proposed here was segfaulting when I tried it.

Generally speaking, "here's a fix from my heavily tweaked poco code" is not an efficient way to get the issues fixed. If you have a problem and want it fixed quick, send a pull request to devel branch, let's see if CI goes green with it and then we'll be in a much better place to fix it.

/cc @matejk