On unknown circumstances, when trying to close a TCPConnecti

Library versions: <div class="snippet-clipboard-content notranslate position-relat

Trying to get StreamSocketSlot but have none. about eventcore HOT 1 OPEN

vibe-d commented on June 23, 2024

Trying to get StreamSocketSlot but have none.

from eventcore.

Comments (1)

TETYYS commented on June 23, 2024

Library versions:

"botan": "1.12.9",
"botan-math": "1.0.3",
"diet-ng": "1.4.5",
"eventcore": "0.8.35",
"libasync": "0.8.3",
"libevent": "2.0.2+2.0.16",
"memutils": "0.4.10",
"mir-linux-kernel": "1.0.0",
"openssl": "1.1.6+1.0.1g",
"stdx-allocator": "2.77.2",
"taggedalgebraic": "0.10.11",
"vibe-core": "1.4.0",
"vibe-d": "0.8.4"

I think this issue happens because each worker thread have their own event loops and each event loop has their own FD pool. When connect and close are combined with runWorkerTaskH, sometimes close is called on different event loop the connection was created on.

This code snippet reproduces this issue (runTask instead of runWorkerTaskH gives same result):

import vibe.vibe;

void fx()
{
	TCPConnection conn;
	setTimer(2.seconds, {
		conn.close();
	});
	conn = connectTCP("google.com", 80);
}

void main()
{
	foreach (x; 0 .. 100) {
		runWorkerTaskH(&fx);
	}

	runApplication();
}

When this runs fine:

import vibe.vibe;

void fx()
{
	TCPConnection conn;
	runTask((TCPConnection c), {
		sleep(2.seconds);
		c.close();
	}, conn);
	conn = connectTCP("google.com", 80);
}

void main()
{
	foreach (x; 0 .. 100) {
		runWorkerTaskH(&fx);
	}

	runApplication();
}

Now this points that this is reference count issue.

from eventcore.

Recommend Projects