Re: [Condor-users] Reboot the submit machine and not restart jobs?

["Horvatth Szabolcs" <szabolcs@xxxxxxxxxxxxx>]

Sometimes the restarted DAGMan job does not restart properly,
the DAG structure gets freezed at the time of reboot and no new jobs are submitted.
I could not reproduce the problem for sure but it does happen quite often.

Cheers,
Szabolcs




*********** REPLY SEPARATOR  ***********

On 1/12/2006 at 8:42 AM Matt Hope wrote:

>On 1/11/06, Finch, Ralph <rfinch@xxxxxxxxxxxx> wrote:
>> condor -version
>> $CondorVersion: 6.7.13 Nov  7 2005 $
>> $CondorPlatform: INTEL-WINNT50 $
>
>given this
>
>> Occasionally after submitting jobs from my machine, I need to reboot my
>> machine (it's windows, after all).  However, this means all my jobs must
>> restart (since this is windows, it is only the vanilla universe and they
>> are not checkpointed).  Since the jobs take a few hours to complete, I
>> was wondering if it's possible for the jobs running in the pool on other
>> machines to not restart, but simply reconnect with new condor_shadows
>> when my submit machine comes back after reboot.
>
>The answer is yes, but only if the startd's are also 6.7 series and
>are configured to enable leasing.
>
>http://www.cs.wisc.edu/condor/manual/v6.7/2_15Special_Environment.html#SECTION003154000000000000000
>
>A proviso on this for windows is that if you shutdown normally then
>the lease will not happen - it will trigger an eviction. To make it
>work on shutdown you need to hard kill your condor subsystem (pskill
>is going to be your friend here) as if your machine reset without
>warning.!
>
>I dislike this behaviour intensely (especially given windows likely
>use of more vanilla non checkpointing jobs) - since if you have some
>jobs which *can* checkpoint and others which can't then you're SOL.
>
>When my farm goes to 6.8 I can see this being the most common "why
>can't I do this?" request I get.
>
>Matt
>
>_______________________________________________
>Condor-users mailing list
>Condor-users@xxxxxxxxxxx
>https://lists.cs.wisc.edu/mailman/listinfo/condor-users