dask_jobqueue.SGECluster

class dask_jobqueue.SGECluster(n_workers=0, job_cls: Optional[dask_jobqueue.core.Job] = None, loop=None, security=None, shared_temp_directory=None, silence_logs='error', name=None, asynchronous=False, dashboard_address=None, host=None, scheduler_options=None, scheduler_cls=<class 'distributed.scheduler.Scheduler'>, interface=None, protocol=None, config_name=None, **job_kwargs)

Launch Dask on an SGE cluster

Note

If you want a specific amount of RAM, both memory and resource_spec must be specified. The exact syntax of resource_spec is defined by your GridEngine system administrator. The amount of memory requested should match the resource_spec, so that Dask’s memory management system can perform accurately.

Parameters
queuestr

Destination queue for each worker job. Passed to #$ -q option.

projectstr

Accounting string associated with each worker job. Passed to #$ -A option.

coresint

Total number of cores per job

memory: str

Total amount of memory per job

processesint

Cut the job up into this many processes. Good for GIL workloads or for nodes with many cores. By default, process ~= sqrt(cores) so that the number of processes and the number of threads per process is roughly the same.

interfacestr

Network interface like ‘eth0’ or ‘ib0’. This will be used both for the Dask scheduler and the Dask workers interface. If you need a different interface for the Dask scheduler you can pass it through the scheduler_options argument: interface=your_worker_interface, scheduler_options={'interface': your_scheduler_interface}.

nannybool

Whether or not to start a nanny process

local_directorystr

Dask worker local directory for file spilling.

death_timeoutfloat

Seconds to wait for a scheduler before closing workers

extralist

Additional arguments to pass to dask-worker

env_extralist

Other commands to add to script before launching worker.

header_skiplist

Lines to skip in the header. Header lines matching this text will be removed

log_directorystr

Directory to use for job scheduler logs.

shebangstr

Path to desired interpreter for your batch submission script.

pythonstr

Python executable used to launch Dask workers. Defaults to the Python that is submitting these jobs

config_namestr

Section to use from jobqueue.yaml configuration file.

namestr

Name of Dask worker. This is typically set by the Cluster

n_workersint

Number of workers to start by default. Defaults to 0. See the scale method

silence_logsstr

Log level like “debug”, “info”, or “error” to emit here if the scheduler is started locally

asynchronousbool

Whether or not to run this cluster object with the async/await syntax

securitySecurity or Bool

A dask.distributed security object if you’re using TLS/SSL. If True, temporary self-signed credentials will be created automatically.

scheduler_optionsdict

Used to pass additional arguments to Dask Scheduler. For example use scheduler_options={'dashboard_address': ':12435'} to specify which port the web dashboard should use or scheduler_options={'host': 'your-host'} to specify the host the Dask scheduler should run on. See distributed.Scheduler for more details.

scheduler_clstype

Changes the class of the used Dask Scheduler. Defaults to Dask’s distributed.Scheduler.

shared_temp_directorystr

Shared directory between scheduler and worker (used for example by temporary security certificates) defaults to current working directory if not set.

resource_specstr

Request resources and specify job placement. Passed to #$ -l option.

walltimestr

Walltime for each worker job.

job_extralist

List of other SGE options, for example -w e. Each option will be prepended with the #$ prefix.

Examples

>>> from dask_jobqueue import SGECluster
>>> cluster = SGECluster(
...     queue='regular',
...     project="myproj",
...     cores=24,
...     memory="500 GB"
... )
>>> cluster.scale(jobs=10)  # ask for 10 jobs

>>> from dask.distributed import Client
>>> client = Client(cluster)

This also works with adaptive clusters. This automatically launches and kill workers based on load.

>>> cluster.adapt(maximum_jobs=20)
__init__(n_workers=0, job_cls: Optional[dask_jobqueue.core.Job] = None, loop=None, security=None, shared_temp_directory=None, silence_logs='error', name=None, asynchronous=False, dashboard_address=None, host=None, scheduler_options=None, scheduler_cls=<class 'distributed.scheduler.Scheduler'>, interface=None, protocol=None, config_name=None, **job_kwargs)

Methods

__init__([n_workers, job_cls, loop, ...])

adapt(*args[, minimum_jobs, maximum_jobs])

Scale Dask cluster automatically based on scheduler activity.

close([timeout])

from_name(name)

Create an instance of this class to represent an existing cluster by name.

get_logs([cluster, scheduler, workers])

Return logs for the cluster, scheduler and workers

job_script()

logs(*args, **kwargs)

new_worker_spec()

Return name and spec for the next worker

scale([n, jobs, memory, cores])

Scale cluster to specified configurations.

scale_down(workers)

scale_up([n, memory, cores])

Scale cluster to n workers

sync(func, *args[, asynchronous, ...])

Attributes

asynchronous

dashboard_link

job_header

job_name

name

observed

plan

requested

scheduler_address

previous

dask_jobqueue.PBSCluster

next

dask_jobqueue.SLURMCluster