@Joe is this upstream patch by any chance in the latest HHVM builds you have been testing?

Nope of course, but I guess we can add it to our next build. How serious is this?

@bd808 @Anomie where did you see this error? which version of hhvm?

Already discussed on IRC, but for posterity: Error occurs on both mw1017 (3.6.1) and mw1050 (3.3.1) using either the test script in the upstream bug report or the following variant that uses SessionHandlerInterface:

<?php

echo PHP_VERSION . "\n";

// These implement a do-nothing session handler
class S implements SessionHandlerInterface {
	function open( $path, $name ){ return true; }
	function close(){ return true; }
	function read( $id ){ return ''; }
	function write( $id, $data ){ return true; }
	function destroy( $id ){ return true; }
	function gc( $t ){ return true; }
}

session_set_save_handler(new S(), true);

// Test
echo "start\n";

session_start();
session_destroy();
session_start();

echo "ok\n";

@Joe do we have a "next HHVM build" tracking bug that we can attach this too so it doesn't get lost?

Change 226286 had a related patch set uploaded (by Giuseppe Lavagetto):
Perserve session handler on session_destroy

https://gerrit.wikimedia.org/r/226286

Change 226286 merged by Giuseppe Lavagetto:
Perserve session handler on session_destroy

https://gerrit.wikimedia.org/r/226286

Testing on beta, it appears that (once deployed) the new package will fix this bug.

@hashar: Please upgrade the CI slaves with the new package from T106483, when it's ready.

Canary appservers updated.

In T97675#1471267, @Anomie wrote:

Testing on beta, it appears that (once deployed) the new package will fix this bug.

@hashar: Please upgrade the CI slaves with the new package from T106483, when it's ready.

Forked as T106699: Upgrade HHVM related packages on Trusty Jenkins slaves.

All appservers upgraded. I'm resolving the ticket now, you can still reopen it later