Keyword Arguments

To make submission easier, this module defines a number of keyword arguments in the file that can be used for all submission and Job() functions. These include things like ‘cores’ and ‘nodes’ and ‘mem’.

The following is a complete list of arguments that can be used in this version

Common: Used in every mode

Option Description Type Default
depends A job or list of jobs to depend on list None
clean_files Auto clean script files when fetching outputs bool None
clean_outputs Auto clean output files when fetching outputs bool None
cores Number of cores to use for the job int 1
modules Modules to load with the module load command list None
syspaths Paths to add to _sys.path for submitted functions list None
scriptpath Folder to write cluster script files to, must be accessible to the compute nodes. str .
outpath Folder to write cluster output files to, must be accessible to the compute nodes. str .
runpath The working directory for the job str .
suffix A suffix to append to job files (e.g. job.suffix.qsub) str cluster
outfile File to write STDOUT to str None
errfile File to write STDERR to str None

Func: Used for function calls

Option Description Type Default
imports Imports to be used in function calls (e.g. sys, os) list None

Cluster: Options that work in both slurm and torque

Option Description Type Default
nodes Number of nodes to request int 1
features A comma-separated list of node features to require list None
qos A quality of service to require str None
time Walltime in HH:MM:SS str 12:00:00
mem Memory to use in MB (e.g. 4000) [‘int’, ‘str’] 4000
partition The partition/queue to run in (e.g. local/batch) str None
account Account to be charged str None
export Comma separated list of environmental variables to export str None

Slurm: Used for slurm only

Option Description Type Default
begin Start after this much time str None


Synonym Option
depend depends
dependency depends
dependencies depends
stdout outfile
stderr errfile
queue partition
memory mem
cpus cores
threads cores
walltime time
delete_files clean_files
delete_outputs clean_outputs
filedir scriptpath
filepath scriptpath
dir runpath
path runpath
paths syspaths
syspath syspaths
scriptdir scriptpath
cleanfiles clean_files
delfiles clean_files
cleanouts clean_outputs
delouts clean_outputs
deloutputs clean_outputs
cleanoutputs clean_outputs

Note: Type is enforced, any provided argument must match that python type (automatic conversion is attempted), the default is just a recommendation and is not currently used. These arguments are passed like regular arguments to the submission and Job() functions, eg:

Job(nodes=1, cores=4, mem='20MB')

This will be interpretted correctly on any system. If torque or slurm are not available, any cluster arguments will be ignored. The module will attempt to honor the cores request, but if it exceeds the maximum number of cores on the local machine, then the request will be trimmed accordingly (i.e. a 50 core request will become 8 cores on an 8 core machine).

Adding your own keywords

There are many more options available for torque and slurm, to add your own, edit the file, and look for CLUSTER_OPTS (or TORQUE/SLURM if your keyword option is only availble on one system). Add your option using the same format as is present in that file. The format is:

('name', {'slurm': '--option-str={}', 'torque': '--torque-option={}',
          'help': 'This is an option!', 'type': str, 'default': None})

You can also add list options, but they must include ‘sjoin’ and ‘tjoin’ keys to define how to merge the list for slurm and torque, or you must write custom option handling code in fyrd.options.options_to_string(). For an excellent example of both approaches included in a single option, see the ‘features’ keyword above.