CI
Docs
Package
License

ncar-jobqueue

ncar-jobqueue
- Installation
- Configuration
- Usage
  - Casper
  - Cheyenne
  - Hobart
  - Izumi
  - Non-NCAR machines

ncar-jobqueue provides utilities for configuring dask-jobqueue with appropriate default settings for NCAR's clusters.

The following compute servers are supported:

Cheyenne (cheyenne.ucar.edu)
Casper (DAV) (casper.ucar.edu)
Hobart (hobart.cgd.ucar.edu)
Izumi (izumi.unified.ucar.edu)

Installation

NCAR-jobqueue can be installed from PyPI with pip:

python -m pip install ncar-jobqueue

NCAR-jobqueue is also available from conda-forge for conda installations:

conda install -c conda-forge ncar-jobqueue

Configuration

ncar-jobqueue provides a custom configuration file with appropriate default settings for different clusters. This configuration file resides in ~/.config/dask/ncar-jobqueue.yaml:

ncar-jobqueue.yaml

cheyenne:
  pbs:
    #project: XXXXXXXX
    name: dask-worker-cheyenne
    cores: 18 # Total number of cores per job
    memory: '109GB' # Total amount of memory per job
    processes: 18 # Number of Python processes per job
    interface: ib0 # Network interface to use like eth0 or ib0
    queue: regular
    walltime: '01:00:00'
    resource-spec: select=1:ncpus=36:mem=109GB
    log-directory: '/glade/scratch/${USER}/dask/cheyenne/logs'
    local-directory: '/glade/scratch/${USER}/dask/cheyenne/local-dir'
    job-extra: []
    env-extra: []
    death-timeout: 60

casper-dav:
  pbs:
    #project: XXXXXXXX
    name: dask-worker-casper-dav
    cores: 2 # Total number of cores per job
    memory: '25GB' # Total amount of memory per job
    processes: 1 # Number of Python processes per job
    interface: ib0
    walltime: '01:00:00'
    resource-spec: select=1:ncpus=1:mem=25GB
    queue: casper
    log-directory: '/glade/scratch/${USER}/dask/casper-dav/logs'
    local-directory: '/glade/scratch/${USER}/dask/casper-dav/local-dir'
    job-extra: []
    env-extra: []
    death-timeout: 60

hobart:
  pbs:
    name: dask-worker-hobart
    cores: 10 # Total number of cores per job
    memory: '96GB' # Total amount of memory per job
    processes: 10 # Number of Python processes per job
    # interface: null              # ib0 doesn't seem to be working on Hobart
    queue: medium
    walltime: '08:00:00'
    resource-spec: nodes=1:ppn=48
    log-directory: '/scratch/cluster/${USER}/dask/hobart/logs'
    local-directory: '/scratch/cluster/${USER}/dask/hobart/local-dir'
    job-extra: ['-r n']
    env-extra: []
    death-timeout: 60

izumi:
  pbs:
    name: dask-worker-izumi
    cores: 10 # Total number of cores per job
    memory: '96GB' # Total amount of memory per job
    processes: 10 # Number of Python processes per job
    # interface: null              # ib0 doesn't seem to be working on Hobart
    queue: medium
    walltime: '08:00:00'
    resource-spec: nodes=1:ppn=48
    log-directory: '/scratch/cluster/${USER}/dask/izumi/logs'
    local-directory: '/scratch/cluster/${USER}/dask/izumi/local-dir'
    job-extra: ['-r n']
    env-extra: []
    death-timeout: 60

Note:

To configure a default project account that is used by dask-jobqueue when submitting batch jobs, uncomment the project key/line in ~/.config/dask/ncar-jobqueue.yaml and set it to an appropriate value.

Usage

Note:

⚠️ Online documentation for dask-jobqueue is available here. ⚠️

Casper

>>> from ncar_jobqueue import NCARCluster
>>> from dask.distributed import Client
>>> cluster = NCARCluster(project='XXXXXXXX')
>>> cluster
PBSCluster(0f23b4bf, 'tcp://xx.xxx.x.x:xxxx', workers=0, threads=0, memory=0 B)
>>> cluster.scale(jobs=2)
>>> client = Client(cluster)

Cheyenne

>>> from ncar_jobqueue import NCARCluster
>>> from dask.distributed import Client
>>> cluster = NCARCluster(project='XXXXXXXX')
>>> cluster
PBSCluster(0f23b4bf, 'tcp://xx.xxx.x.x:xxxx', workers=0, threads=0, memory=0 B)
>>> cluster.scale(jobs=2)
>>> client = Client(cluster)

Hobart

>>> from ncar_jobqueue import NCARCluster
>>> from dask.distributed import Client
>>> cluster = NCARCluster()
>>> cluster
PBSCluster(0f23b4bf, 'tcp://xx.xxx.x.x:xxxx', workers=0, threads=0, memory=0 B)
>>> cluster.scale(jobs=2)
>>> client = Client(cluster)

Izumi

>>> from ncar_jobqueue import NCARCluster
>>> from dask.distributed import Client
>>> cluster = NCARCluster()
>>> cluster
PBSCluster(0f23b4bf, 'tcp://xx.xxx.x.x:xxxx', workers=0, threads=0, memory=0 B)
>>> cluster.scale(jobs=2)
>>> client = Client(cluster)

Non-NCAR machines

On non-NCAR machines, ncar-jobqueue will warn the user, and it will use distributed.LocalCluster:

>>> from ncar_jobqueue import NCARCluster
.../ncar_jobqueue/cluster.py:17: UserWarning: Unable to determine which NCAR cluster you are running on... Returning a `distributed.LocalCluster` class.
warn(message)
>>> from dask.distributed import Client
>>> cluster = NCARCluster()
>>> cluster
LocalCluster(3a7dd0f6, 'tcp://127.0.0.1:64184', workers=4, threads=8, memory=17.18 GB)

andersy005 / ncar-jobqueue