slurm parsing

[Fred Smith]

I am using slurm latest version 14.03.8 on CentOS 6.2. It's compiled following the instructions in the installation guide for rpmbuild and rpm install. The plugin rpm is installed, nevertheless slurm does not start when using the topology sample from the Topology Guide (the node names are also configured in the slurm.conf NodeName):

SwitchName=s0 Nodes=node-0-[4-5]
SwitchName=s1 Nodes=node-0-[6-7]
SwitchName=s2 Switches=s[0-1]

slurm is configured with TopologyPlugin=topology/tree .
However the start fails with parsing errors:

scontrol: error: Parsing error at unrecognized key: SwitchName
scontrol: error: Parse error in file /etc/slurm/topology.conf line 1:
"SwitchName=s0 Nodes=hpc-0-[4-5] "
scontrol: error: Parsing error at unrecognized key: SwitchName
scontrol: error: Parse error in file /etc/slurm/topology.conf line 2:
"SwitchName=s1 Nodes=hpc-0-[6-7] "
scontrol: error: Parsing error at unrecognized key: SwitchName
scontrol: error: Parse error in file /etc/slurm/topology.conf line 3:
"SwitchName=s2 Switches=s[0-1]"
scontrol: error: "Include" failed in file /etc/slurm/slurm.conf line 85
scontrol: fatal: Unable to process configuration file

the plugin is installed: topology_tree.so in the default location and does indeed include
strings : SwitchName LinkSpeed Nodes Switches. The slurm start script does include the default LIBDIR setting.





Any hint what is missing to get the topology to work?

[j***@public.gmane.org]

This is just a guess, but is your topology information below in a file
topology.conf (NOT part of slurm.conf)?

Post by Fred Smith
I am using slurm latest version 14.03.8 on CentOS 6.2. It's compiled
following the instructions in the installation guide for rpmbuild
and rpm install. The plugin rpm is installed, nevertheless slurm
does not start when using the topology sample from the Topology
SwitchName=s0 Nodes=node-0-[4-5]
SwitchName=s1 Nodes=node-0-[6-7]
SwitchName=s2 Switches=s[0-1]
slurm is configured with TopologyPlugin=topology/tree .
scontrol: error: Parsing error at unrecognized key: SwitchName
"SwitchName=s0 Nodes=hpc-0-[4-5] "
scontrol: error: Parsing error at unrecognized key: SwitchName
"SwitchName=s1 Nodes=hpc-0-[6-7] "
scontrol: error: Parsing error at unrecognized key: SwitchName
"SwitchName=s2 Switches=s[0-1]"
scontrol: error: "Include" failed in file /etc/slurm/slurm.conf line 85
scontrol: fatal: Unable to process configuration file
the plugin is installed: topology_tree.so in the default location and does indeed include
strings : SwitchName LinkSpeed Nodes Switches. The slurm start
script does include the default LIBDIR setting.
Any hint what is missing to get the topology to work?

[Fred Smith]

yes, the file is included in slurm.conf.

I also tried the lines in the slurm.conf with the same failure.

scontrol: error: Parsing error at unrecognized key: SwitchName
scontrol: error: Parse error in file /etc/slurm/slurm.conf line 82:

[j***@public.gmane.org]

That's your problem. topology.conf is a completely separate file.

Post by Fred Smith
yes, the file is included in slurm.conf.
I also tried the lines in the slurm.conf with the same failure.
scontrol: error: Parsing error at unrecognized key: SwitchName

[Lloyd Brown]

I'm not an expert, but I think that topology.conf is handled separately
by the corresponding plugin, and shouldn't actually be included in
slurm.conf. If you are including it in slurm.conf, I could see that
being a syntax problem.

But something may have changed since I looked at it last.

Lloyd Brown
Systems Administrator
Fulton Supercomputing Lab
Brigham Young University
http://marylou.byu.edu

Post by Fred Smith
yes, the file is included in slurm.conf.
I also tried the lines in the slurm.conf with the same failure.
scontrol: error: Parsing error at unrecognized key: SwitchName