dbqp on Natty Narwhal is leaving lots of drizzled servers hanging about

[Patrick Crews]

It seems that during the course of executing the test suite(s), a number of lt-drizzled process are left hanging about, but only on Natty Narwhal.

dbqp is able to continually use the same ports (9306, etc) so these leftover processes aren't blocking / interfering (they also used same ports).

Need to determine the cause and get this sorted. No trace of this behavior on Lucid test box.

[Stewart Smith (stewart)]

we should maybe make sure that servers don't detach, so killing dbqp kills them.

this could be done by instead of having --start-and-exit to having (like mtr2) just --start - where ctrl-c on that process then kills all servers.

[Patrick Crews (patrick-crews)]

Well, we track this with the server_manager code (ie all active servers) and the 'finally' block *should* always go through and shutdown everything (ie this *always* happens...normal shutdown, ctrl-c, etc).

This appears to be an issue when we switch servers (due to option changes).

In such a case, we should shutdown the server, update the options, then restart it with new options.

Something appears to be going horribly awry during the shutdown / restart bits. It appears that we're left with zombie lt-drizzled's (like one for every option switch) that can't be connected to.

Also, this insanity appears to be only on Natty : /

However, the --start option isn't a bad idea and can be done.
./dbqp --mode=cleanup is an extra step, but it looks for kicked-off pid files and kills anything (useful after a --start-and-exit play session). Additionally, we automatically do this on a fresh dbqp run

[Patrick Crews (patrick-crews)]

Added code so that we monitor the server pid's on shutdown and we kill them after a timeout (ala test-run.pl)

Not a proper fix, but it makes us not leave a mess when we run tests. Will open a separate bug to properly track down the cause of this.