[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] Suspended --> Retiring transition





Jimmy Choi wrote:

As an experiment, I was trying to simulate the transition from Suspended
to Retiring, but couldn't get it to work.

I have specified the following in condor_config:

START = True # make sure that the job starts
WANT_SUSPEND = True
SUSPEND = True # get into the suspended state immediately
CONTINUE = False PREEMPT = True # should trigger the Suspended->Retiring state?
MaxJobRetirementTime = 3600


Basically, the job stays in the suspended state and never moves to the
Retiring state. What's wrong?



When I try the setup you describe above, here's what I see:

10/1 15:24:25 vm1: State change: SUSPEND is TRUE
10/1 15:24:25 vm1: Changing activity: Busy -> Suspended
10/1 15:24:31 vm1: State change: PREEMPT is TRUE
10/1 15:24:31 vm1: Changing activity: Suspended -> Retiring
10/1 15:24:31 vm1: State change: SUSPEND is TRUE
10/1 15:24:31 vm1: Changing activity: Retiring -> Suspended

and it then stays in the suspended state indefinitely. This is behaving as expected (but I am open to criticism). Here's why it behaves this way:

1. The suspension policy applies during retirement, so the fact that a claim is being retired does not interfere with your ability to make Condor play nicely when there is other activity on the machine.

2. Time spent suspended does _not_ count towards a job's maximum retirement time, because otherwise it is impossible for a user to know what the job will really get when specifying a minimum required retirement time.

If you replace the following two lines in your setup, you can see the transition suspended -> retiring.

SUSPEND = $(ActivationTimer) < 30
CONTINUE = $(ActivationTimer) > 30

--Dan Bradley