[torqueusers] Nodes properties problem ?
Mathieu OUDART
Mathieu.Oudart at cnes.fr
Tue Oct 25 02:37:49 MDT 2005
Hi all,
we use Torque 1.2.0p6 / Maui 3.2.6p14 on our Linux cluster and we have a
problem with node properties.
On our site, node properties are used for partitioning (nodes to queues
mapping).
It seems that any user can override the "resources_default.neednodes =
property" by setting explicitly the execution host : qsub -l
nodes=nodename batch.pbs
Why this behaviour ?
Is there a Torque (or Maui) parameter to prevent this ?
Regards.
####################################################################
Here is a simple testcase of my problem, using a single-node cluster :
Queue configuration :
#
# Create and define queue batch
#
create queue batch
set queue batch queue_type = Execution
set queue batch resources_default.neednodes = property
set queue batch enabled = True
set queue batch started = True
$PBS_SPOOL/server_priv/nodes file :
pc-hot2 np=2 other
[maui at pc-hot2 ~]$ qsub dummy.pbs
7.pc-hot2.cst.cnes.fr
[maui at pc-hot2 ~]$ qstat
Job id Name User Time Use S Queue
---------------- ---------------- ---------------- -------- - -----
7.pc-hot2 dummy.pbs maui 0 Q batch
=> The job is deferred as there is no node available to the job
[maui at pc-hot2 ~]$ qsub -l nodes=pc-hot2 dummy.pbs
11.pc-hot2.cst.cnes.fr
[maui at pc-hot2 ~]$ qstat
Job id Name User Time Use S Queue
---------------- ---------------- ---------------- -------- - -----
7.pc-hot2 dummy.pbs maui 0 Q batch
11.pc-hot2 dummy.pbs maui 0 R batch
=> the job is runnning, even if the node have not the required property !
[maui at pc-hot2 ~]$ checkjob -v 11
checking job 11 (RM job '11.pc-hot2.cst.cnes.fr')
State: Running
Creds: user:maui group:torque class:batch qos:DEFAULT
WallTime: 00:00:00 of 00:00:00
SubmitTime: Tue Oct 25 12:07:12
(Time Queued Total: 00:00:01 Eligible: 00:00:01)
StartTime: Tue Oct 25 12:07:13
Total Tasks: 1
Req[0] TaskCount: 1 Partition: DEFAULT
Network: [NONE] Memory >= 0 Disk >= 0 Swap >= 0
Opsys: [NONE] Arch: [NONE] Features: [property]
Exec: '' ExecSize: 0 ImageSize: 0
Dedicated Resources Per Task: PROCS: 1
Utilized Resources Per Task: [NONE]
Avg Util Resources Per Task: [NONE]
Max Util Resources Per Task: [NONE]
NodeAccess: SHARED
NodeCount: 0
Allocated Nodes:
[pc-hot2:1]
Task Distribution: pc-hot2
IWD: [NONE] Executable: [NONE]
Bypass: 0 StartCount: 1
PartitionMask: [ALL]
Flags: HOSTLIST RESTARTABLE
HostList:
[pc-hot2:1]
Reservation '11' (00:00:00 -> 00:00:01 Duration: 00:00:01)
PE: 1.00 StartPriority: 25005
=> You can notice the "Features: [property]" on the job, but it seems to
be ignored.
--
Mathieu OUDART
More information about the torqueusers
mailing list