[slurm-users] job is pending but resources are available

Ole Holm Nielsen Ole.H.Nielsen at fysik.dtu.dk
Wed Oct 13 08:30:24 UTC 2021


On 10/13/21 9:59 AM, Adam Xu wrote:
> 
> 在 2021/10/13 9:22, Brian Andrus 写道:
>>
>> Something is very odd when you have the node reporting:
>>
>> RealMemory=1 AllocMem=0 FreeMem=47563 Sockets=2 Boards=1
>>
>> What do you get when you run ‘slurmd -C’ on the node?
>>
> # slurmd -C
> NodeName=apollo CPUs=36 Boards=1 SocketsPerBoard=2 CoresPerSocket=18 
> ThreadsPerCore=1 RealMemory=128306
> UpTime=22-16:14:48

Maybe try to replace "Boards=1 SocketsPerBoard=2" by "Sockets=2".  The 
"Boards" used to give problems in Slurm 20.02, see 
https://wiki.fysik.dtu.dk/niflheim/Slurm_configuration#compute-node-configuration
but should be fixed in 20.11.

Your node apollo has CPUAlloc=28 CPUTot=36 which is OK, since there are 8 
available cores.

Maybe your 8 GPUs are all in use by the 7 jobs?  Can you send your 
gres.conf file defining GPUs?

To view the node apollo jobs, I recommend to download the "pestat" command 
from https://github.com/OleHolmNielsen/Slurm_tools/tree/master/pestat

Just type "pestat -G" to view the GRES resources for each job on all nodes.

Best regards,
Ole



>> *From: *Adam Xu <mailto:adam_xu at adagene.com.cn>
>> *Sent: *Tuesday, October 12, 2021 6:07 PM
>> *To: *slurm-users at lists.schedmd.com
>> *Subject: *Re: [slurm-users] job is pending but resources are available
>>
>> 在 2021/10/12 21:21, Adam Xu 写道:
>>
>>     Hi All,
>>
>>     OS: Rocky Linux 8.4
>>
>>     slurm version: 20.11.7
>>
>>     the partition's name is apollo. the node's name is apollo too. the
>>     node has 36 cpu cores and 8GPUs in it.
>>
>>     partition info
>>
>>     $ scontrol show partition apollo
>>     PartitionName=apollo
>>        AllowGroups=ALL AllowAccounts=ALL AllowQos=ALL
>>        AllocNodes=ALL Default=NO QoS=N/A
>>        DefaultTime=NONE DisableRootJobs=NO ExclusiveUser=NO GraceTime=0
>>     Hidden=NO
>>        MaxNodes=UNLIMITED MaxTime=UNLIMITED MinNodes=0 LLN=NO
>>     MaxCPUsPerNode=UNLIMITED
>>        Nodes=apollo
>>        PriorityJobFactor=1 PriorityTier=1 RootOnly=NO ReqResv=NO
>>     OverSubscribe=YES:36
>>        OverTimeLimit=NONE PreemptMode=OFF
>>        State=UP TotalCPUs=36 TotalNodes=1 SelectTypeParameters=NONE
>>        JobDefaults=(null)
>>        DefMemPerNode=UNLIMITED MaxMemPerNode=UNLIMITED
>>
>>     node info
>>
>>     $ scontrol show node apollo
>>     NodeName=apollo Arch=x86_64 CoresPerSocket=18
>>        CPUAlloc=28 CPUTot=36 CPULoad=7.02
>>        AvailableFeatures=(null)
>>        ActiveFeatures=(null)
>>        Gres=gpu:v100:8,mps:v100:800
>>        NodeAddr=apollo NodeHostName=apollo Version=20.11.7
>>        OS=Linux 4.18.0-305.19.1.el8_4.x86_64 #1 SMP Wed Sep 15 19:12:32
>>     UTC 2021
>>        RealMemory=1 AllocMem=0 FreeMem=47563 Sockets=2 Boards=1
>>        State=MIXED ThreadsPerCore=1 TmpDisk=0 Weight=1 Owner=N/A
>>     MCS_label=N/A
>>        Partitions=apollo
>>        BootTime=2021-09-20T23:43:49 SlurmdStartTime=2021-10-12T16:55:44
>>        CfgTRES=cpu=36,mem=1M,billing=36
>>        AllocTRES=cpu=28
>>        CapWatts=n/a
>>        CurrentWatts=0 AveWatts=0
>>        ExtSensorsJoules=n/s ExtSensorsWatts=0 ExtSensorsTemp=n/s
>>        Comment=(null)
>>
>>     Now I have 7 jobs running but when I submit 8th job, the status of
>>     the job is pending beacuse Resources.
>>
>>     $ squeue
>>                  JOBID PARTITION     NAME     USER ST       TIME NODES
>>     NODELIST(REASON)
>>                    879    apollo    do.sh zhining_ PD       0:00 1
>>     (Resources)
>>                    489    apollo    do.sh zhining_  R 13-12:50:45 1 apollo
>>                    490    apollo    do.sh zhining_  R 13-12:41:00 1 apollo
>>                    592    apollo runme-gp junwen_f  R 4-12:42:31 1 apollo
>>                    751    apollo runme-gp junwen_f  R 1-12:48:20 1 apollo
>>                    752    apollo runme-gp junwen_f  R 1-12:48:10 1 apollo
>>                    871    apollo runme-gp junwen_f  R    7:13:45 1 apollo
>>                    872    apollo runme-gp junwen_f  R    7:12:42 1 apollo
>>
>>     $ scontrol show job 879
>>     JobId=879 JobName=do.sh
>>        UserId=zhining_wan(1001) GroupId=zhining_wan(1001) MCS_label=N/A
>>        Priority=4294900882 Nice=0 Account=(null) QOS=(null)
>>        JobState=PENDING Reason=Resources Dependency=(null)
>>        Requeue=1 Restarts=0 BatchFlag=1 Reboot=0 ExitCode=0:0
>>        RunTime=00:00:00 TimeLimit=UNLIMITED TimeMin=N/A
>>        SubmitTime=2021-10-12T16:29:29 EligibleTime=2021-10-12T16:29:29
>>        AccrueTime=2021-10-12T16:29:29
>>        StartTime=2021-10-12T21:17:41 EndTime=Unknown Deadline=N/A
>>        SuspendTime=None SecsPreSuspend=0 LastSchedEval=2021-10-12T21:17:39
>>        Partition=apollo AllocNode:Sid=sms:1281191
>>        ReqNodeList=(null) ExcNodeList=(null)
>>        NodeList=(null) SchedNodeList=apollo
>>        NumNodes=1-1 NumCPUs=4 NumTasks=4 CPUs/Task=1 ReqB:S:C:T=0:0:*:*
>>        TRES=cpu=4,node=1,billing=4
>>        Socks/Node=* NtasksPerN:B:S:C=0:0:*:* CoreSpec=*
>>        MinCPUsNode=1 MinMemoryNode=0 MinTmpDiskNode=0
>>        Features=(null) DelayBoot=00:00:00
>>        OverSubscribe=YES Contiguous=0 Licenses=(null) Network=(null)
>>     Command=/home/zhining_wan/job/2021/20210603_ctla4_double_bilayer/final_pdb_minimize/amber/nolipid/test/do.sh
>>
>>     WorkDir=/home/zhining_wan/job/2021/20210603_ctla4_double_bilayer/final_pdb_minimize/amber/nolipid/test
>>
>>     StdErr=/home/zhining_wan/job/2021/20210603_ctla4_double_bilayer/final_pdb_minimize/amber/nolipid/test/slurm-879.out
>>
>>        StdIn=/dev/null
>>     StdOut=/home/zhining_wan/job/2021/20210603_ctla4_double_bilayer/final_pdb_minimize/amber/nolipid/test/slurm-879.out
>>
>>        Power=
>>        TresPerNode=gpu:1
>>        NtasksPerTRES:0
>>
>>     After running 7 jobs, the node has 8 cpu cores and 1 gpu left, so I
>>     can be sure that the remaining resources are sufficient. but why the
>>     job is pending with reason "Resources"?
>>
>> Some information to add:
>>
>> I have killed some jobs with kill instead of scancle, Could this be the 
>> cause of this result?




More information about the slurm-users mailing list