[slurm-users] Defining an empty partition

[Steve Brasier]

Thank you Tina, I hadn't realised that would show as "n/a" not "down" in
that case (which IMO would have been confusing). For anyone else hitting
this I think the minimum you can do is something like:

PartitionName=compute Default=YES <options> State=UP Nodes=nosuch
NodeName=nosuch

The documented approach would have been easier in my case given
the constraints of the template logic generating this but at least it's a
workaround.

thanks
Steve

http://stackhpc.com/
Please note I work Tuesday to Friday.

On 18/12/2020 12:45:26, Tina Friedrich wrote:

> Yeah, I had that problem as well (trying to set up a partition that
> didn't have any nodes - they're not here yet). I figured that one can have
> partitions with nodes that don't exist,
> though. As in, not even in DNS. I currently have this: [arc-slurm ~]$ sinfo
> PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
> short        up   12:00:00      1  down* arc-c023
> short        up   12:00:00      1  alloc arc-c001
> short        up   12:00:00     43   idle arc-c[002-022,024-045]
> medium       up 2-00:00:00      0    n/a
> long*        up   infinite      0    n/a with medium & long partition
> containing nodes 'arc-c[046-297]': PartitionName=medium
>    AllowGroups=ALL AllowAccounts=ALL AllowQos=ALL
>    AllocNodes=ALL Default=NO QoS=N/A
>    DefaultTime=12:00:00 DisableRootJobs=NO ExclusiveUser=NO GraceTime=0
> Hidden=NO
>    MaxNodes=UNLIMITED MaxTime=2-00:00:00 MinNodes=0 LLN=NO
> MaxCPUsPerNode=UNLIMITED
>    Nodes=arc-c[046-297]... which don't exist as of today: [arc-slurm ~]$
> host arc-c046
> Host arc-c046 not found: 3(NXDOMAIN) which - as you can see - simply ends
> up with SLURM showing the partition
> with no nodes. So you could just put a dummy nodename in the slurm.conf
> file? Tina



On Fri, 18 Dec 2020 at 11:13, Steve Brasier <steveb at stackhpc.com> wrote:

> Having tried just not even defining any partitions you hit this this
> <https://github.com/SchedMD/slurm/blob/master/src/common/node_conf.c#L383>check
> which seems to ensure you can't create a cluster with no nodes. Is it
> possible to create a control node without any compute nodes, e.g. as part
> of a staged deployment?
>
> http://stackhpc.com/
> Please note I work Tuesday to Friday.
>
>
> On Fri, 18 Dec 2020 at 10:56, Steve Brasier <steveb at stackhpc.com> wrote:
>
>> Hi all,
>>
>> According to the relevant manpage
>> <https://slurm.schedmd.com/archive/slurm-20.02.5/slurm.conf.html> it's
>> possible to define an empty partition using "Nodes= ".
>>
>> However this doesn't seem to work (slurm 20.2.05):
>>
>> [centos at testohpc-login-0 ~]$ grep -n Partition /etc/slurm/slurm.conf
>> 72:PriorityWeightPartition=1000
>> 105:PartitionName=compute Default=YES MaxTime=86400 State=UP Nodes=
>>
>> (note there is a space after that final "=" but I've tried both with and
>> without)
>>
>> [centos at testohpc-login-0 ~]$ sinfo
>> sinfo: error: Parse error in file /etc/slurm/slurm.conf line 105: "
>> Nodes= "
>> sinfo: fatal: Unable to process configuration file
>>
>> Is this a bug, or am I doing it wrong?
>>
>> thanks for any suggestions
>>
>> Steve
>>
>> http://stackhpc.com/
>> Please note I work Tuesday to Friday.
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.schedmd.com/pipermail/slurm-users/attachments/20201218/0dc07e2a/attachment.htm>