Broken pipe exception on high and rapid map zoom

[Grandro]

qlever-petrimaps/src/qlever-petrimaps/server/Server.cpp

Lines 869 to 882 in 338df82

	int sock = *((int*)png_get_io_ptr(png_ptr));
	size_t writes = 0;
	while (writes != length) {
	int64_t out =
	write(sock, reinterpret_cast<char*>(data) + writes, length - writes);
	if (out < 0) {
	if (errno == EWOULDBLOCK || errno == EAGAIN || errno == EINTR) continue;
	break;
	}
	writes += out;
	}
	}

In line 875 I sometimes get a Broken pipe exception when zooming on high zoom (where object dots/heatmap are bigger). This can also happen when zooming out from high zoom again. I think it only happens when zooming rapidly.

[patrickbr]

I suspect this isn't a broken pipe exception (write() is a C function and therefore doesn't throw any exceptions), but a SIGPIPE signal.

Will try to replicate this and handle SIGPIPEs correctly here.

[patrickbr]

Can you try again using the latest master? SIGPIPE is now handled by SIG_IGN (ignore) explicitly in Server.cpp (it is already set so in util::Server, so I don't quite understand why you still received them). Additionally, all write() calls to the socket have been replaced by send() (write() with additional flags), and the MSG_NOSIGNAL flag is now always set).

Just out of curiosity: which OS are you using? I was unable to replicate this on an Ubuntu and a Debian system, even when adding a sleep(10) before line 874 and closing the socket immediately after the request, and under heavy load testing with hundreds of parallel requests per second. I had to explicitly re-set the default signal handler for SIGPIPE to get the behavior your described, so my guess is that some defaults differ between your system and our testing systems.

[Grandro]

Excuse my late reply. I am using Windows 11 + WSL but I was also not able to replicate the problem on Windows 10 + WSL back then. I just spent a few minutes rapidly zooming in and out, and the issue seems to be resolved. Thank you.