Slurm requires that you define the amount of free memory, after OS/Applications are considered, when reporting memory as a resource. If the reported memory is too low, then Slurm will reject this node. To overcome this, by default we dampen the memory by 5% or 1g, whichever is larger.
To change this dampening, there are two options.
1) You can define slurm.dampen_memory=X where X is an integer percentage (5 == 5%)
2) Crate a default_resource definition in the /opt/azurehpc/slurm/autoscale.json file.
Fixes: #30
Related new documentation:
Slurm requires that you define the amount of free memory, after OS/Applications are considered, when reporting memory as a resource. If the reported memory is too low, then Slurm will reject this node. To overcome this, by default we dampen the memory by 5% or 1g, whichever is larger.
To change this dampening, there are two options. 1) You can define
slurm.dampen_memory=X
where X is an integer percentage (5 == 5%) 2) Crate a default_resource definition in the /opt/azurehpc/slurm/autoscale.json file.Default resources are a powerful tool that the underlying library ScaleLib provides. see the ScaleLib documentation