Directives

Directives are optional settings that affect the execution of the current process. They are declared at the top of a process definition, before any input, output, or script section.

By default, directives are evaluated when the process is defined. However, if the value is a dynamic string or closure, it will be evaluated separately for each task. See Dynamic directives for more information.

Directives can also be specified in a configuration file using the process config scope, which is the recommended way to configure a pipeline for a particular compute environment. See Process configuration for more information.

Some directives are only supported by specific executors. See Executors for more information about each executor.

The following directives are available:

Directive	Description
accelerator	The number of hardware accelerators (e.g. GPUs) required by each task.
afterScript	A Bash snippet to execute after the task script.
arch	The CPU architecture to build the software required by each task.
array	Submit tasks as job arrays, for executors that support it.
beforeScript	A Bash snippet to execute before the task script.
cache	Whether and how task executions are cached.
clusterOptions	Additional submission options for grid executors.
conda	The set of Conda packages required by each task.
container	The container image required by each task.
containerOptions	Additional options for the underlying container runtime.
cpus	The number of CPUs required by each task.
debug	Print the standard output of each task to the pipeline standard output.
disk	The amount of disk storage required by each task.
errorStrategy	How to handle task failures.
executor	The underlying system where tasks are executed.
ext	A generic directive for user-defined properties.
fair	Emit process outputs in the order in which they were received.
hints	Executor-specific hints, specified as key-value pairs.
label	A custom label, used to apply shared process configuration.
machineType	A predefined machine type for cloud executors that support it.
maxErrors	The maximum number of task failures allowed for a process.
maxForks	The maximum number of concurrent task executions for a process.
maxRetries	The maximum number of times a task can be retried.
maxSubmitAwait	How long a task can remain in the submission queue before failing.
memory	The amount of memory required by each task.
module	The set of Environment Modules required by each task.
penv	The parallel environment to use when submitting tasks to SGE.
pod	Pod-specific settings for the Kubernetes executor.
publishDir	Publish matching process output files to a target directory.
queue	The queue to which tasks should be submitted.
resourceLabels	Custom name-value pairs to attach to task executions.
resourceLimits	Environment-specific limits for task resource requests.
scratch	Execute each task in a temporary directory local to the compute node.
secret	The secrets required by each task.
shell	A custom shell command for process scripts.
spack	The set of Spack packages required by each task.
stageInMode	How input files are staged into the task directory.
stageOutMode	How output files are staged out of the scratch directory.
storeDir	Store task outputs in a permanent store directory.
tag	A custom identifier for each task execution.
time	The maximum runtime for each task.

See also

• Dynamic directives: How to define a directive value based on the task inputs.
• Dynamic task resources: How to define task resources that scale with each retry.
• Task directive values: How to access directive values in a process body.