[Moabusers] Re: Default Nodesets being ignored.

[wightman]

We have not been able to reproduce this locally.  When a job is
submitted with the nodesets flag it is properly scheduled.  There may be
something unique about your environment that we have not correctly
duplciated, however.

Can you send the output of the following commands, as well as your
moab.cfg file:

mdiag -n -v
mdiag -t

Thanks,

- Douglas

On Mon, 2006-08-28 at 16:27 -0400, Justin Bronder wrote:
> We are attempting to use defaut nodesets on a per queue basis to get
> certain
> queues to favor certain machines.  The following output should clarify
> the issue.
> This method used to work in patch 4.  We are currently using patch 7.
> 
> 
> jbronder at panopticon:~/src$ qsub -I -l nodes=1:ppn=1 -l walltime=25:00
> -q darwin-admin -A systemTest -l nodeset=oneof:feature:31  
> \qsub: waiting for job 1062.echelon.acrl.clusters.umaine.edu to start
> qsub: job 1062.echelon.acrl.clusters.umaine.edu ready
> jbronder at node128 ~ $  
> 
> So I'm on node128, however:
> 
> root at echelon:/usr/src/moab-4.5.0p7/include# checknode node128
> node node128
> 
> State:   Running  (in current state for 00:00:26)
> Expected State:     Idle   SyncDeadline: Mon Aug 28 16:33:34
> Configured Resources: PROCS: 2  MEM: 1973M  SWAP: 3979M  DISK: 1M
> Utilized   Resources: PROCS: 1
> Dedicated  Resources: PROCS: 2  MEM: 1973M  SWAP: 3979M  DISK: 1M
> Opsys:      linux     Arch:      ppc64 
> Speed:      1.00      CPULoad:   0.000
> Network Load: 0.01 kB/s
> Flags:      rmdetected
> Network:    DEFAULT
> Features:   pRACK1,RACK1,17
> Attributes: [Batch]
> Classes:    [default 2:2][linux-spool 2:2][darwin-spool
> 2:2][linux-batch 2:2][darwin-batch 2:2][linux-admin 2:2][darwin-admin
> 1:2][kearney 2:2]
> RM[base]:   TYPE=PBS
> NodeAccessPolicy: SINGLEJOB
> 
> Total Time: 5:21:17:12  Up: 3:19:29:44 (64.76%)  Active: 00:02:05
> (0.02%)
> 
> Reservations:
>   1062x1  Job:Running  -00:00:56 -> 00:24:04 (00:25:00)
> Jobs:        1062
> ALERT:  node has 2 procs dedicated but load is low (0.000)
> 
> As you can see, node128 does not have the 31 feature.  There are
> plenty of nodes
> (20 to be exact) that do have this feature and can be verified to be
> available using
> checknode.  Is there something else that needs to be done to force the
> nodesets to
> be used?
> 
> Just in case it matters, here is the checkjob output:
> 
> root at echelon:/usr/src/moab-4.5.0p7/include# checkjob 1062
> job 1062
> 
> AName: STDIN
> State: Running 
> Creds:  user:jbronder  group:clusteradmin  account:systemTest
> class:darwin-admin
> WallTime:   00:02:42 of 00:25:00
> SubmitTime: Mon Aug 28 16:23:34
>   (Time Queued  Total: 00:00:00  Eligible: -00:00:01)
> 
> StartTime: Mon Aug 28 16:23:34
> Total Requested Tasks: 1
> 
> Req[0]  TaskCount: 1  Partition: RACK1
> Memory >= 0  Disk >= 0  Swap >= 0
> Opsys:   ---  Arch: ---  Features: ---
> NodeSet=ONEOF:FEATURE:31
> 
> Allocated Nodes:
> [node128:1]
> 
> 
> StartCount:     1
> StartPriority:  2
> Reservation '1062' (-00:02:47 -> 00:22:13  Duration: 00:25:00)
> 
> 
> 
> Thanks,
> Justin.
>