launchpad-dev team mailing list archive

[Maris Fogels <maris.fogels@xxxxxxxxxxxxx>]

On 08/13/2010 10:39 AM, Jonathan Lange wrote:
> On Fri, Aug 13, 2010 at 3:37 PM, Guilherme Salgado
> <salgado@xxxxxxxxxxxxx>  wrote:
> > On Fri, 2010-08-13 at 14:42 +0100, Jonathan Lange wrote:
> > > On Fri, Aug 13, 2010 at 3:53 AM, Tim Penhey<tim.penhey@xxxxxxxxxxxxx>  wrote:
> > > > Hi All,
> > > >
> > > > I've had a number of situations where an ec2 instance hasn't shut down after
> > > > it has sent out the email.  The instance was still running almost two hours
> > > > after sending the email.
> > >
> > > According to Bazaar, the code that does this hasn't changed recently.
> > > The most likely candidate is a 2010-07-25 change (r11224) that times
> > > out the server using "shutdown +TIMEOUT" rather than a call to "at".
> >
> > Well, the timeout is set to 8h and the test suite takes around 3h IIRC,
> > so the shutdown wouldn't have kicked in when Tim noticed the instance
> > was still running.
> >
> > This is not the code responsible for terminating the instances, though
> > -- it was added to prevent these rogue instances from being kept running
> > for too long.  Maybe we can change the timeout to 6h or maybe even 5h?
>
> Yeah, I didn't think it was a very likely candidate, just the most
> likely of a poor bunch.

So it turns out you can't call 'shutdown -P now' if another call to shutdown is 
already running.  ec2 test calls 'shutdown -P now' after the test suite 
finishes, but the call fails because the failsafe shutdown is still running. 
The failsafe shutdown wins, and the server stays up for the full 8 hours.

A patch is forthcoming.

--
Māris Fogels -- https://launchpad.net/~mars
Launchpad.net -- cross-project collaboration and hosting