Processes

In Nextflow, a process is the basic processing primitive to execute a user script.

The process definition starts with the keyword process, followed by process name and finally the process body delimited by curly braces. The process body must contain a string which represents the command or, more generally, a script that is executed by it. A basic process looks like the following example:

process sayHello {
    """
    echo 'Hello world!' > file
    """
}

A process may contain any of the following definition blocks: directives, inputs, outputs, when clause, and the process script. The syntax is defined as follows:

process < name > {

  [ directives ]

  input:
    < process inputs >

  output:
    < process outputs >

  when:
    < condition >

  [script|shell|exec]:
    < user script to be executed >

}

Script

The script block defines, as a string expression, the script that is executed by the process.

A process may contain only one script, and if the script guard is not explicitly declared, the script must be the final statement in the process block.

The script string is executed as a Bash script in the host environment. It can be any command or script that you would normally execute on the command line or in a Bash script. Naturally, the script may only use commands that are available in the host environment.

The script block can be a simple string or a multi-line string. The latter approach makes it easier to write scripts with multiple commands spanning multiple lines. For example:

process doMoreThings {
  """
  blastp -db $db -query query.fa -outfmt 6 > blast_result
  cat blast_result | head -n 10 | cut -f 2 > top_hits
  blastdbcmd -db $db -entry_batch top_hits > sequences
  """
}

As explained in the script tutorial section, strings can be defined using single-quotes or double-quotes, and multi-line strings are defined by three single-quote or three double-quote characters.

There is a subtle but important difference between them. Like in Bash, strings delimited by a " character support variable substitutions, while strings delimited by ' do not.

In the above code fragment, the $db variable is replaced by the actual value defined elsewhere in the pipeline script.

Warning

Since Nextflow uses the same Bash syntax for variable substitutions in strings, you must manage them carefully depending on whether you want to evaluate a Nextflow variable or a Bash variable.

When you need to access a system environment variable in your script, you have two options.

If you don’t need to access any Nextflow variables, you can define your script block with single-quotes:

process printPath {
  '''
  echo The path is: $PATH
  '''
}

Otherwise, you can define your script with double-quotes and escape the system environment variables by prefixing them with a back-slash \ character, as shown in the following example:

process doOtherThings {
  """
  blastp -db \$DB -query query.fa -outfmt 6 > blast_result
  cat blast_result | head -n $MAX | cut -f 2 > top_hits
  blastdbcmd -db \$DB -entry_batch top_hits > sequences
  """
}

In this example, $MAX is a Nextflow variable that must be defined elsewhere in the pipeline script. Nextflow replaces it with the actual value before executing the script. Meanwhile, $DB is a Bash variable that must exist in the execution environment, and Bash will replace it with the actual value during execution.

Tip

Alternatively, you can use the Shell block definition, which allows a script to contain both Bash and Nextflow variables without having to escape the first.

Scripts à la carte

The process script is interpreted by Nextflow as a Bash script by default, but you are not limited to Bash.

You can use your favourite scripting language (Perl, Python, R, etc), or even mix them in the same pipeline.

A pipeline may be composed of processes that execute very different tasks. With Nextflow, you can choose the scripting language that best fits the task performed by a given process. For example, for some processes R might be more useful than Perl, whereas for others you may need to use Python because it provides better access to a library or an API, etc.

To use a language other than Bash, simply start your process script with the corresponding shebang. For example:

process perlTask {
    """
    #!/usr/bin/perl

    print 'Hi there!' . '\n';
    """
}

process pythonTask {
    """
    #!/usr/bin/python

    x = 'Hello'
    y = 'world!'
    print "%s - %s" % (x,y)
    """
}

workflow {
    perlTask()
    pythonTask()
}

Tip

Since the actual location of the interpreter binary file can differ across platforms, it is wise to use the env command followed by the interpreter name, e.g. #!/usr/bin/env perl, instead of the absolute path, in order to make your script more portable.

Conditional scripts

So far, our script block has always been a simple string expression, but in reality, the script block is just Groovy code that returns a string. This means that you can write arbitrary Groovy code to determine the script to execute, as long as the final statement is a string (remember that the return keyword is optional in Groovy).

For example, you can use flow control statements (if, switch, etc) to execute a different script based on the process inputs. The only difference here is that you must explicitly declare the script guard, whereas before it was not required. Here is an example:

mode = 'tcoffee'

process align {
    input:
    path sequences

    script:
    if( mode == 'tcoffee' )
        """
        t_coffee -in $sequences > out_file
        """

    else if( mode == 'mafft' )
        """
        mafft --anysymbol --parttree --quiet $sequences > out_file
        """

    else if( mode == 'clustalo' )
        """
        clustalo -i $sequences -o out_file
        """

    else
        error "Invalid alignment mode: ${mode}"
}

In the above example, the process will execute one of the script fragments depending on the value of the mode parameter. By default it will execute the tcoffee command, but changing the mode variable will cause a different branch to be executed.

Template

Process scripts can be externalised to template files, which can be reused across different processes and tested independently from the overall pipeline execution.

A template is simply a shell script file that Nextflow is able to execute by using the template function as shown below:

process templateExample {
    input:
    val STR

    script:
    template 'my_script.sh'
}

workflow {
    Channel.of('this', 'that') | templateExample
}

By default, Nextflow looks for the my_script.sh template file in the templates directory located alongside the Nextflow script and/or the module script in which the process is defined. Any other location can be specified by using an absolute template path.

The template script may contain any code that can be executed by the underlying environment. For example:

#!/bin/bash
echo "process started at `date`"
echo $STR
echo "process completed"

Tip

The dollar character ($) is interpreted as a Nextflow variable when the script is run as a Nextflow template, whereas it is evaluated as a Bash variable when run as a Bash script. This can be very useful for testing your script independently from Nextflow execution. You only need to provide a Bash environment variable for each of the Nextflow variables that are referenced in your script. For example, it would be possible to execute the above script with the following command in the terminal: STR='foo' bash templates/my_script.sh

Tip

As a best practice, the template script should not contain any \$ escaped variables, because these variables will not be evaluated properly when the script is executed directly.

Shell

The shell block is a string expression that defines the script that is executed by the process. It is an alternative to the Script definition with one important difference: it uses the exclamation mark ! character, instead of the usual dollar $ character, to denote Nextflow variables.

This way, it is possible to use both Nextflow and Bash variables in the same script without having to escape the latter, which makes process scripts easier to read and maintain. For example:

process myTask {
    input:
    val str

    shell:
    '''
    echo "User $USER says !{str}"
    '''
}

workflow {
    Channel.of('Hello', 'Hola', 'Bonjour') | myTask
}

In the above example, $USER is treated as a Bash variable, while !{str} is treated as a Nextflow variable.

Note

  • Shell script definitions require the use of single-quote ' delimited strings. When using double-quote " delimited strings, dollar variables are interpreted as Nextflow variables as usual. See String interpolation.

  • Variables prefixed with ! must always be enclosed in curly brackets, i.e. !{str} is a valid variable whereas !str is ignored.

  • Shell scripts support the use of the Template mechanism. The same rules are applied to the variables defined in the script template.

Native execution

Nextflow processes can also execute native Groovy code as the task itself, using the exec block. Whereas the script block defines a script to be executed, the exec block defines Groovy code to be executed directly.

For example:

process simpleSum {
    input:
    val x

    exec:
    println "Hello Mr. $x"
}

workflow {
    Channel.of('a', 'b', 'c') | simpleSum
}

will display:

Hello Mr. b
Hello Mr. a
Hello Mr. c

Stub

New in version 20.11.0-edge.

You can define a command stub, which replaces the actual process command when the -stub-run or -stub command-line option is enabled:

process INDEX {
  input:
    path transcriptome

  output:
    path 'index'

  script:
    """
    salmon index --threads $task.cpus -t $transcriptome -i index
    """

  stub:
    """
    mkdir index
    touch index/seq.bin
    touch index/info.json
    touch index/refseq.bin
    """
}

The stub block can be defined before or after the script block. When the pipeline is executed with the -stub-run option and a process’s stub is not defined, the script block is executed.

This feature makes it easier to quickly prototype the workflow logic without using the real commands. The developer can use it to provide a dummy script that mimics the execution of the real one in a quicker manner. In other words, it is a way to perform a dry-run.

Inputs

The input block allows you to define the input channels of a process, similar to function arguments. A process may have at most one input block, and it must contain at least one input.

The input block follows the syntax shown below:

input:
  <input qualifier> <input name>

An input definition consists of a qualifier and a name. The input qualifier defines the type of data to be received. This information is used by Nextflow to apply the semantic rules associated with each qualifier, and handle it properly depending on the target execution platform (grid, cloud, etc).

When a process is invoked in a workflow block, it must be provided a channel for each channel in the process input block, similar to calling a function with specific arguments. The examples provided in the following sections demonstrate how a process is invoked with input channels.

The following input qualifiers are available:

  • val: Access the input value by name in the process script.

  • file: (DEPRECATED) Handle the input value as a file, staging it properly in the execution context.

  • path: Handle the input value as a path, staging the file properly in the execution context.

  • env: Use the input value to set an environment variable in the process script.

  • stdin: Forward the input value to the process stdin special file.

  • tuple: Handle a group of input values having any of the above qualifiers.

  • each: Execute the process for each element in the input collection.

Input type val

The val qualifier accepts any data type. It can be accessed in the process script by using the specified input name, as shown in the following example:

process basicExample {
  input:
  val x

  "echo process job $x"
}

workflow {
  def num = Channel.of(1,2,3)
  basicExample(num)
}

In the above example, the process is executed three times: once for each value emitted by the num channel. The resulting output is similar to the one shown below:

process job 3
process job 1
process job 2

Note

While channels do emit items in the order that they are received, processes do not necessarily process items in the order that they are received. In the above example, the value 3 was processed before the others.

Note

When the process declares exactly one input, the pipe | operator can be used to provide inputs to the process, instead of passing it as a parameter. Both methods have identical semantics:

process basicExample {
  input:
  val x

  "echo process job $x"
}

workflow {
  Channel.of(1,2,3) | basicExample
}

Input type file

Note

The file qualifier was the standard way to handle input files prior to Nextflow 19.10.0. In later versions of Nextflow, the path qualifier should be preferred over file.

The file qualifier is identical to path, with one important difference. When a file input receives a value that is not a file, it automatically converts the value to a string and saves it to a temporary file. This behavior is useful in some cases, but tends to be confusing in general. The path qualifier instead interprets string values as the path location of the input file and automatically converts to a file object.

Input type path

The path qualifier allows you to provide input files to the process execution context. Nextflow will stage the files into the process execution directory, and they can be accessed in the script by using the specified input name. For example:

process blastThemAll {
  input:
  path query_file

  "blastp -query ${query_file} -db nr"
}

workflow {
  def proteins = Channel.fromPath( '/some/path/*.fa' )
  blastThemAll(proteins)
}

In the above example, all the files ending with the suffix .fa are sent over the channel proteins. These files are received by the process, which executes a BLAST query on each of them.

It’s worth noting that in the above example, the name of the file in the file-system is not used. You can access the file without even knowing its name, because you can reference it in the process script by the input name.

There may be cases where your task needs to use a file whose name is fixed, i.e. it does not have to change along with the actual provided file. In this case, you can specify a fixed name with the name attribute in the input file parameter definition, as shown in the following example:

input:
path query_file, name: 'query.fa'

or, using a shorter syntax:

input:
path 'query.fa'

The previous example can be re-written as shown below:

process blastThemAll {
  input:
  path 'query.fa'

  "blastp -query query.fa -db nr"
}

workflow {
  def proteins = Channel.fromPath( '/some/path/*.fa' )
  blastThemAll(proteins)
}

In this example, each file received by the process is staged with the name query.fa in a different execution context (i.e. the folder where a task is executed).

Tip

This feature allows you to execute the process command multiple times without worrying about the file names changing. In other words, Nextflow helps you write pipeline tasks that are self-contained and decoupled from the execution environment. As a best practice, you should avoid referencing files in your process script other than those defined in your input block.

Channel factories like Channel.fromPath produce file objects, but a path input can also accept a string literal path. The string value should be an absolute path, i.e. it must be prefixed with a / character or a supported URI protocol (file://, http://, s3://, etc), and it cannot contain special characters (\n, etc).

process foo {
  input:
  path x

  """
  your_command --in $x
  """
}

workflow {
  foo('/some/data/file.txt')
}

Available options:

arity

New in version 23.09.0-edge.

Specify the number of expected files. Can be a number or a range:

input:
    path('one.txt', arity: '1')         // exactly one file is expected
    path('pair_*.txt', arity: '2')      // exactly two files are expected
    path('many_*.txt', arity: '1..*')   // one or more files are expected

When a task is created, Nextflow will check whether the received files for each path input match the declared arity, and fail if they do not.

stageAs

Specify how the file should be named in the task work directory:

process foo {
  input:
  path x, stageAs: 'data.txt'

  """
  your_command --in data.txt
  """
}

workflow {
  foo('/some/data/file.txt')
}

Can be a name or a pattern as described in the Multiple input files section.

Note

Process path inputs have nearly the same interface as described in Files and I/O, with one difference which is relevant when files are staged into a subdirectory. Given the following input:

path x, stageAs: 'my-dir/*'

In this case, x.name returns the file name with the parent directory (e.g. my-dir/file.txt), whereas normally it would return the file name (e.g. file.txt). You can use x.fileName.name to get the file name.

Multiple input files

A path input can also accept a collection of files instead of a single value. In this case, the input variable will be a Groovy list, and you can use it as such.

When the input has a fixed file name and a collection of files is received by the process, the file name will be appended with a numerical suffix representing its ordinal position in the list. For example:

process blastThemAll {
    input:
    path 'seq'

    "echo seq*"
}

workflow {
    def fasta = Channel.fromPath( "/some/path/*.fa" ).buffer(size: 3)
    blastThemAll(fasta)
}

will output:

seq1 seq2 seq3
seq1 seq2 seq3
...

The target input file name may contain the * and ? wildcards, which can be used to control the name of staged files. The following table shows how the wildcards are replaced depending on the cardinality of the received input collection.

Cardinality

Name pattern

Staged file names

any

*

named as the source file

1

file*.ext

file.ext

1

file?.ext

file1.ext

1

file??.ext

file01.ext

many

file*.ext

file1.ext, file2.ext, file3.ext, ..

many

file?.ext

file1.ext, file2.ext, file3.ext, ..

many

file??.ext

file01.ext, file02.ext, file03.ext, ..

many

dir/*

named as the source file, created in dir subdirectory

many

dir??/*

named as the source file, created in a progressively indexed subdirectory e.g. dir01/, dir02/, etc.

many

dir*/*

(as above)

The following example shows how a wildcard can be used in the input file definition:

process blastThemAll {
    input:
    path 'seq?.fa'

    "cat seq1.fa seq2.fa seq3.fa"
}

workflow {
    def fasta = Channel.fromPath( "/some/path/*.fa" ).buffer(size: 3)
    blastThemAll(fasta)
}

Note

Rewriting input file names according to a named pattern is an extra feature and not at all required. The normal file input syntax introduced in the Input type path section is valid for collections of multiple files as well. To handle multiple input files while preserving the original file names, use a variable identifier or the * wildcard.

Dynamic input file names

When the input file name is specified by using the name option or a string literal, you can also use other input values as variables in the file name string. For example:

process simpleCount {
  input:
  val x
  path "${x}.fa"

  """
  cat ${x}.fa | grep '>'
  """
}

In the above example, the input file name is determined by the current value of the x input value.

This approach allows input files to be staged in the task directory with a name that is coherent with the current execution context.

Tip

In most cases, you won’t need to use dynamic file names, because each task is executed in its own directory, and input files are automatically staged into this directory by Nextflow. This behavior guarantees that input files with the same name won’t overwrite each other.

An example of when you may have to deal with that is when you have many input files in a task, and some of these files may have the same filename. In this case, a solution would be to use the stageAs option.

Input type env

The env qualifier allows you to define an environment variable in the process execution context based on the input value. For example:

process printEnv {
    input:
    env HELLO

    '''
    echo $HELLO world!
    '''
}

workflow {
    Channel.of('hello', 'hola', 'bonjour', 'ciao') | printEnv
}
hello world!
ciao world!
bonjour world!
hola world!

Input type stdin

The stdin qualifier allows you to forward the input value to the standard input of the process script. For example:

process printAll {
  input:
  stdin str

  """
  cat -
  """
}

workflow {
  Channel.of('hello', 'hola', 'bonjour', 'ciao')
    | map { it + '\n' }
    | printAll
}

will output:

hola
bonjour
ciao
hello

Input type tuple

The tuple qualifier allows you to group multiple values into a single input definition. It can be useful when a channel emits tuples of values that need to be handled separately. Each element in the tuple is associated with a corresponding element in the tuple definition. For example:

process tupleExample {
    input:
    tuple val(x), path('input.txt')

    """
    echo "Processing $x"
    cat input.txt > copy
    """
}

workflow {
  Channel.of( [1, 'alpha.txt'], [2, 'beta.txt'], [3, 'delta.txt'] ) | tupleExample
}

In the above example, the tuple input consists of the value x and the file input.txt.

A tuple definition may contain any of the following qualifiers, as previously described: val, env, path and stdin. Files specified with the path qualifier are treated exactly the same as standalone path inputs.

Input repeaters (each)

The each qualifier allows you to repeat the execution of a process for each item in a collection, each time a new value is received. For example:

process alignSequences {
  input:
  path seq
  each mode

  """
  t_coffee -in $seq -mode $mode > result
  """
}

workflow {
  sequences = Channel.fromPath('*.fa')
  methods = ['regular', 'espresso', 'psicoffee']

  alignSequences(sequences, methods)
}

In the above example, each time a file of sequences is emitted from the sequences channel, the process executes three tasks, each running a T-coffee alignment with a different value for the mode parameter. This behavior is useful when you need to repeat the same task over a given set of parameters.

Input repeaters can be applied to files as well. For example:

process alignSequences {
  input:
  path seq
  each mode
  each path(lib)

  """
  t_coffee -in $seq -mode $mode -lib $lib > result
  """
}

workflow {
  sequences = Channel.fromPath('*.fa')
  methods = ['regular', 'espresso']
  libraries = [ file('PQ001.lib'), file('PQ002.lib'), file('PQ003.lib') ]

  alignSequences(sequences, methods, libraries)
}

In the above example, each sequence input file emitted by the sequences channel triggers six alignment tasks, three with the regular method against each library file, and three with the espresso method.

Note

When multiple repeaters are defined, the process is executed for each combination of them.

Note

Input repeaters currently do not support tuples. However, you can emulate an input repeater on a channel of tuples by using the combine or cross operator with other input channels to produce all of the desired input combinations.

Multiple input channels

A key feature of processes is the ability to handle inputs from multiple channels.

When two or more channels are declared as process inputs, the process waits until there is a complete input configuration, i.e. until it receives a value from each input channel. When this condition is satisfied, the process consumes a value from each channel and launches a new task, repeating this logic until one or more channels are empty.

As a result, channel values are consumed sequentially and any empty channel will cause the process to wait, even if the other channels have values.

For example:

process foo {
  input:
  val x
  val y

  script:
  """
  echo $x and $y
  """
}

workflow {
  x = Channel.of(1, 2)
  y = Channel.of('a', 'b', 'c')
  foo(x, y)
}

The process foo is executed two times because the x channel emits only two values, therefore the c element is discarded. It outputs:

1 and a
2 and b

A different semantic is applied when using a value channel. This kind of channel is created by the Channel.value factory method or implicitly when a process is invoked with an argument that is not a channel. By definition, a value channel is bound to a single value and it can be read an unlimited number of times without consuming its content. Therefore, when mixing a value channel with one or more (queue) channels, it does not affect the process termination because the underlying value is applied repeatedly.

To better understand this behavior, compare the previous example with the following one:

process bar {
  input:
  val x
  val y

  script:
  """
  echo $x and $y
  """
}

workflow {
  x = Channel.value(1)
  y = Channel.of('a', 'b', 'c')
  foo(x, y)
}

The above example executes the bar process three times because x is a value channel, therefore its value can be read as many times as needed. The process termination is determined by the contents of y. It outputs:

1 and a
1 and b
1 and c

Note

In general, multiple input channels should be used to process combinations of different inputs, using the each qualifier or value channels. Having multiple queue channels as inputs is equivalent to using the merge operator, which is not recommended as it may lead to non-deterministic process inputs.

See also: Channel types.

Outputs

The output block allows you to define the output channels of a process, similar to function outputs. A process may have at most one output block, and it must contain at least one output.

The output block follows the syntax shown below:

output:
  <output qualifier> <output name> [, <option>: <option value>]

An output definition consists of a qualifier and a name. Some optional attributes can also be specified.

When a process is invoked, each process output is returned as a channel. The examples provided in the following sections demonstrate how to access the output channels of a process.

The following output qualifiers are available:

  • val: Emit the variable with the specified name.

  • file: (DEPRECATED) Emit a file produced by the process with the specified name.

  • path: Emit a file produced by the process with the specified name.

  • env: Emit the variable defined in the process environment with the specified name.

  • stdout: Emit the stdout of the executed process.

  • tuple: Emit multiple values.

  • eval: Emit the result of a script or command evaluated in the task execution context.

Output type val

The val qualifier allows you to output any Nextflow variable defined in the process. A common use case is to output a variable that was defined in the input block, as shown in the following example:

process foo {
  input:
  each x

  output:
  val x

  """
  echo $x > file
  """
}

workflow {
  methods = ['prot', 'dna', 'rna']

  receiver = foo(methods)
  receiver.view { "Received: $it" }
}

The output value can be a value literal, an input variable, any other Nextflow variable in the process scope, or a value expression. For example:

process foo {
  input:
  path infile

  output:
  val x
  val 'BB11'
  val "${infile.baseName}.out"

  script:
  x = infile.name
  """
  cat $x > file
  """
}

workflow {
  ch_dummy = Channel.fromPath('*').first()
  (ch_var, ch_str, ch_exp) = foo(ch_dummy)

  ch_var.view { "ch_var: $it" }
  ch_str.view { "ch_str: $it" }
  ch_exp.view { "ch_exp: $it" }
}

Output type file

Note

The file qualifier was the standard way to handle input files prior to Nextflow 19.10.0. In later versions of Nextflow, the path qualifier should be preferred over file.

The file qualifier is similar to path, but with some differences. The file qualifier interprets : as a path separator, therefore file 'foo:bar' captures two files named foo and bar, whereas path 'foo:bar' captures a single file named foo:bar. Additionally, file does not support all of the extra options provided by path.

Output type path

The path qualifier allows you to output one or more files produced by the process. For example:

process randomNum {
  output:
  path 'result.txt'

  '''
  echo $RANDOM > result.txt
  '''
}

workflow {
  numbers = randomNum()
  numbers.view { "Received: ${it.text}" }
}

In the above example, the randomNum process creates a file named result.txt which contains a random number. Since a path output with the same name is declared, that file is emitted by the corresponding output channel. A downstream process with a compatible input channel will be able to receive it.

Available options:

arity

New in version 23.09.0-edge.

Specify the number of expected files. Can be a number or a range:

output:
    path('one.txt', arity: '1')         // exactly one file is expected
    path('pair_*.txt', arity: '2')      // exactly two files are expected
    path('many_*.txt', arity: '1..*')   // one or more files are expected

When a task completes, Nextflow will check whether the produced files for each path output match the declared arity, and fail if they do not. If the arity is 1, a sole file object will be emitted. Otherwise, a list will always be emitted, even if only one file is produced.

followLinks

When true, target files are returned in place of any matching symlink (default: true)

glob

When true, the specified name is interpreted as a glob pattern (default: true)

hidden

When true, hidden files are included in the matching output files (default: false)

includeInputs

When true and the output path is a glob pattern, any input files matching the pattern are also included in the output (default: false)

maxDepth

Maximum number of directory levels to visit (default: no limit).

type

Type of paths returned, either file, dir or any (default: any, or file if the specified file name pattern contains a double star (**))

Multiple output files

When an output file name contains a * or ? wildcard character, it is interpreted as a glob path matcher. This allows you to capture multiple files into a list and emit the list as a single value. For example:

process splitLetters {
    output:
    path 'chunk_*'

    '''
    printf 'Hola' | split -b 1 - chunk_
    '''
}

workflow {
    splitLetters
        | flatten
        | view { "File: ${it.name} => ${it.text}" }
}

It prints:

File: chunk_aa => H
File: chunk_ab => o
File: chunk_ac => l
File: chunk_ad => a

By default, all the files matching the specified glob pattern are emitted as a single list. However, as the above example demonstrates, the flatten operator can be used to transform the list of files into a channel that emits each file individually.

Some caveats on glob pattern behavior:

  • Input files are not included (unless includeInputs is true)

  • Directories are included, unless the ** pattern is used to recurse through directories

Warning

Although the input files matching a glob output declaration are not included in the resulting output channel, these files may still be transferred from the task scratch directory to the original task work directory. Therefore, to avoid unnecessary file copies, avoid using loose wildcards when defining output files, e.g. path '*'. Instead, use a prefix or a suffix to restrict the set of matching files to only the expected ones, e.g. path 'prefix_*.sorted.bam'.

Read more about glob syntax at the following link What is a glob?

Dynamic output file names

When an output file name needs to be expressed dynamically, it is possible to define it using a dynamic string which references variables in the input block or in the script global context. For example:

process align {
  input:
  val species
  path seq

  output:
  path "${species}.aln"

  """
  t_coffee -in $seq > ${species}.aln
  """
}

In the above example, each process execution produces an alignment file whose name depends on the actual value of the species input.

Tip

The management of output files in Nextflow is often misunderstood.

With other tools it is generally necessary to organize the output files into some kind of directory structure or to guarantee a unique file name scheme, so that result files don’t overwrite each other and so they can be referenced unequivocally by downstream tasks.

With Nextflow, in most cases, you don’t need to manage the naming of output files, because each task is executed in its own unique directory, so files produced by different tasks can’t overwrite each other. Also, metadata can be associated with outputs by using the tuple output qualifier, instead of including them in the output file name.

One example in which you’d need to manage the naming of output files is when you use the publishDir directive to have output files also in a specific path of your choice. If two tasks have the same filename for their output and you want them to be in the same path specified by publishDir, the last task to finish will overwrite the output of the task that finished before. You can dynamically change that by adding the saveAs option to your publishDir directive.

To sum up, the use of output files with static names over dynamic ones is preferable whenever possible, because it will result in simpler and more portable code.

Output type env

The env qualifier allows you to output a variable defined in the process execution environment:

process myTask {
    output:
    env FOO

    script:
    '''
    FOO=$(seq 5)
    '''
}

workflow {
    myTask | view
}

Changed in version 23.12.0-edge: Prior to this version, if the environment variable contained multiple lines of output, the output would be compressed to a single line by converting newlines to spaces.

Output type stdout

The stdout qualifier allows you to output the stdout of the executed process:

process sayHello {
    output:
    stdout

    """
    echo Hello world!
    """
}

workflow {
    sayHello | view { "I say... $it" }
}

Output type eval

New in version 24.02.0-edge.

The eval qualifier allows you to capture the standard output of an arbitrary command evaluated the task shell interpreter context:

process sayHello {
    output:
    eval('bash --version')

    """
    echo Hello world!
    """
}

workflow {
    sayHello | view
}

Only one-line Bash commands are supported. You can use a semi-colon ; to specify multiple Bash commands on a single line, and many interpreters can execute arbitrary code on the command line, e.g. python -c 'print("Hello world!")'.

If the command fails, the task will also fail. In Bash, you can append || true to a command to suppress any command failure.

Output type tuple

The tuple qualifier allows you to output multiple values in a single channel. It is useful when you need to associate outputs with metadata, for example:

process blast {
  input:
    val species
    path query

  output:
    tuple val(species), path('result')

  script:
    """
    blast -db nr -query $query > result
    """
}

workflow {
  ch_species = Channel.from('human', 'cow', 'horse')
  ch_query = Channel.fromPath('*.fa')

  blast(ch_species, ch_query)
}

In the above example, a blast task is executed for each pair of species and query that are received. Each task produces a new tuple containing the value for species and the file result.

A tuple definition may contain any of the following qualifiers, as previously described: val, path, env and stdout. Files specified with the path qualifier are treated exactly the same as standalone path inputs.

Note

While parentheses for input and output qualifiers are generally optional, they are required when specifying elements in an input/output tuple.

Here’s an example with a single path output (parentheses optional):

process foo {
    output:
    path 'result.txt', hidden: true

    '''
    echo 'another new line' >> result.txt
    '''
}

And here’s an example with a tuple output (parentheses required):

process foo {
    output:
    tuple path('last_result.txt'), path('result.txt', hidden: true)

    '''
    echo 'another new line' >> result.txt
    echo 'another new line' > last_result.txt
    '''
}

Additional options

The following options are available for all process outputs:

emit: <name>

Defines the name of the output channel, which can be used to access the channel by name from the process output:

process FOO {
    output:
    path 'hello.txt', emit: hello
    path 'bye.txt', emit: bye

    """
    echo "hello" > hello.txt
    echo "bye" > bye.txt
    """
}

workflow {
    FOO()
    FOO.out.hello.view()
}

See Invoking processes for more details.

optional: true | false

Normally, if a specified output is not produced by the task, the task will fail. Setting optional: true will cause the task to not fail, and instead emit nothing to the given output channel.

output:
path("output.txt"), optional: true

In this example, the process is normally expected to produce an output.txt file, but in the cases where the file is missing, the task will not fail. The output channel will only contain values for those tasks that produced output.txt.

Note

While this option can be used with any process output, it cannot be applied to individual elements of a tuple output. The entire tuple must be optional or not optional.

topic: <name>

New in version 23.11.0-edge.

Experimental: may change in a future release.

Defines the channel topic to which the output will be sent.

When

The when block allows you to define a condition that must be satisfied in order to execute the process. The condition can be any expression that returns a boolean value.

It can be useful to enable/disable the process execution depending on the state of various inputs and parameters. For example:

process find {
  input:
  path proteins
  val dbtype

  when:
  proteins.name =~ /^BB11.*/ && dbtype == 'nr'

  script:
  """
  blastp -query $proteins -db nr
  """
}

Tip

As a best practice, it is better to define such control flow logic in the workflow block, i.e. with an if statement or with channel operators, to make the process more portable.

Directives

Directives are optional settings that affect the execution of the current process.

They must be entered at the top of the process body, before any other declaration blocks (input, output, etc), and have the following syntax:

// directive with simple value
name value

// directive with list value
name arg1, arg2, arg3

// directive with map value
name key1: val1, key2: val2

// directive with value and options
name arg, opt1: val1, opt2: val2

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, which allows task-specific variables like task and val inputs to be used.

Some directives are generally available to all processes, while others depend on the executor currently defined.

accelerator

New in version 19.09.0-edge.

The accelerator directive allows you to request hardware accelerators (e.g. GPUs) for the task execution. For example:

process foo {
    accelerator 4, type: 'nvidia-tesla-k80'

    script:
    """
    your_gpu_enabled --command --line
    """
}

The above examples will request 4 GPUs of type nvidia-tesla-k80.

Note

This directive is only used by certain executors. Refer to the Executors page to see which executors support this directive.

Note

The accelerator type option depends on the target execution platform. Refer to the platform-specific documentation for details on the available accelerators:

The accelerator type option is not supported for AWS Batch. You can control the accelerator type indirectly through the allowed instance types in your Compute Environment. See the AWS Batch FAQs for more information.

afterScript

The afterScript directive allows you to execute a custom (Bash) snippet immediately after the main process has run. This may be useful to clean up your staging area.

Note

When combined with the container directive, the afterScript will be executed outside the specified container. In other words, the afterScript is always executed in the host environment.

arch

The arch directive allows you to define the CPU architecture to build the software in use by the process’ task. For example:

process cpu_task {
    spack 'blast-plus@2.13.0'
    arch 'linux/x86_64', target: 'cascadelake'

    """
    blastp -query input_sequence -num_threads ${task.cpus}
    """
}

The example above declares that the CPU generic architecture is linux/x86_64 (X86 64 bit), and more specifically that the microarchitecture is cascadelake (a specific generation of Intel CPUs).

This directive is currently used by the following Nextflow functionalities:

  • by the spack directive, to build microarchitecture-optimised applications;

  • by the Wave containers service, to build containers for one of the generic families of CPU architectures (see below);

  • by the spack strategy within Wave containers, to optimise the container builds for specific CPU microarchitectures.

Allowed values for the arch directive are as follows, grouped by equivalent family (choices available for the sake of compatibility):

  • X86 64 bit: linux/x86_64, x86_64, linux/amd64, amd64

  • ARM 64 bit: linux/aarch64, aarch64, linux/arm64, arm64, linux/arm64/v8

  • ARM 64 bit, older generation: linux/arm64/v7

Examples of values for the architecture target option are cascadelake, icelake, zen2 and zen3. See the Spack documentation for the full and up-to-date list of meaningful targets.

array

New in version 24.04.0.

Warning

Experimental: may change in a future release.

The array directive allows you to submit tasks as job arrays for executors that support it.

A job array is a collection of jobs with the same resource requirements and the same script (parameterized by an index). Job arrays incur significantly less scheduling overhead compared to individual jobs, and as a result they are preferred by HPC schedulers where possible.

The directive should be specified with a given array size, along with an executor that supports job arrays. For example:

process cpu_task {
    executor 'slurm'
    array 100

    '''
    your_command --here
    '''
}

Nextflow currently supports job arrays for the following executors:

A process using job arrays will collect tasks and submit each batch as a job array when it is ready. Any “leftover” tasks will be submitted as a partial job array.

Once a job array is submitted, each “child” task is executed as an independent job. Any tasks that fail (and can be retried) will be retried without interfering with the tasks that succeeded. Retried tasks are submitted individually rather than through a job array, in order to allow for the use of dynamic resources.

The following directives must be uniform across all tasks in a process that uses job arrays, because these directives are specified once for the entire job array:

For cloud-based executors like AWS Batch, or when using Fusion with any executor, the following additional directives must be uniform:

beforeScript

The beforeScript directive allows you to execute a custom (Bash) snippet before the main process script is run. This may be useful to initialise the underlying cluster environment or for other custom initialisation.

For example:

process foo {
  beforeScript 'source /cluster/bin/setup'

  """
  echo bar
  """
}

When the process is containerized (using the container directive), the beforeScript will be executed in the container only if the executor is container-native (e.g. cloud batch executors, Kubernetes). Otherwise, the beforeScript will be executed outside the container.

cache

The cache directive allows you to store the process results to a local cache. When the cache is enabled and the pipeline is launched with the resume option, any task executions that are already cached will be re-used. See the Caching and resuming page for more information about how the cache works.

The cache is enabled by default, but you can disable it for a specific process by setting the cache directive to false. For example:

process noCacheThis {
  cache false

  script:
  <your command string here>
}

The following options are available:

false

Disable caching.

true (default)

Enable caching. Input file metadata (name, size, last updated timestamp) are included in the cache keys.

'deep'

Enable caching. Input file content is included in the cache keys.

'lenient'

Enable caching. Minimal input file metadata (name and size only) are included in the cache keys.

This strategy provides a workaround for incorrect caching invalidation observed on shared file systems due to inconsistent file timestamps.

clusterOptions

The clusterOptions directive allows the usage of any native configuration option accepted by your cluster submit command. You can use it to request non-standard resources or use settings that are specific to your cluster and not supported out of the box by Nextflow.

The cluster options can be a string:

process foo {
  clusterOptions '-x 1 -y 2'
  // ...
}

Changed in version 24.04.0: Prior to this version, grid executors that require each option to be on a separate line in the job script would attempt to split multiple options using a variety of different conventions. Multiple options can now be specified more clearly using a string list as shown below.

The cluster options can also be a string list:

process foo {
  clusterOptions '-x 1', '-y 2', '--flag'
  // ...
}

Grid executors that require one option per line will write each option to a separate line, while grid executors that allow multiple options per line will write all options to a single line, the same as with a string. This form is useful to control how the options are split across lines when it is required by the scheduler.

Note

This directive is only used by grid executors. Refer to the Executors page to see which executors support this directive.

Warning

While you can use the clusterOptions directive to specify options that are supported as process directives (queue, memory, time, etc), you should not use both at the same time, as it will cause undefined behavior. Most HPC schedulers will either fail or simply ignore one or the other.

conda

The conda directive allows for the definition of the process dependencies using the Conda package manager.

Nextflow automatically sets up an environment for the given package names listed by in the conda directive. For example:

process foo {
  conda 'bwa=0.7.15'

  '''
  your_command --here
  '''
}

Multiple packages can be specified separating them with a blank space e.g. bwa=0.7.15 fastqc=0.11.5. The name of the channel from where a specific package needs to be downloaded can be specified using the usual Conda notation i.e. prefixing the package with the channel name as shown here bioconda::bwa=0.7.15.

The conda directive also allows the specification of a Conda environment file path or the path of an existing environment directory. See the Conda environments page for further details.

container

The container directive allows you to execute the process script in a Docker container.

It requires the Docker daemon to be running in machine where the pipeline is executed, i.e. the local machine when using the local executor or the cluster nodes when the pipeline is deployed through a grid executor.

For example:

process runThisInDocker {
  container 'dockerbox:tag'

  """
  <your holy script here>
  """
}

Simply replace in the above script dockerbox:tag with the name of the Docker image you want to use.

Tip

Containers are a very useful way to execute your scripts in a reproducible self-contained environment or to run your pipeline in the cloud.

Note

This directive is ignored for processes that are executed natively.

containerOptions

The containerOptions directive allows you to specify any container execution option supported by the underlying container engine (ie. Docker, Singularity, etc). This can be useful to provide container settings only for a specific process e.g. mount a custom path:

process runThisWithDocker {
    container 'busybox:latest'
    containerOptions '--volume /data/db:/db'

    output:
    path 'output.txt'

    '''
    your_command --data /db > output.txt
    '''
}

Warning

This feature is not supported by the Kubernetes and Google Life Sciences executors.

cpus

The cpus directive allows you to define the number of (logical) CPU required by the process’ task. For example:

process big_job {
  cpus 8
  executor 'sge'

  """
  blastp -query input_sequence -num_threads ${task.cpus}
  """
}

This directive is required for tasks that execute multi-process or multi-threaded commands/tools and it is meant to reserve enough CPUs when a pipeline task is executed through a cluster resource manager.

See also: penv, memory, time, queue, maxForks

debug

By default the stdout produced by the commands executed in all processes is ignored. By setting the debug directive to true, you can forward the process stdout to the current top running process stdout file, showing it in the shell terminal.

For example:

process sayHello {
  debug true

  script:
  "echo Hello"
}
Hello

Without specifying debug true, you won’t see the Hello string printed out when executing the above example.

disk

The disk directive allows you to define how much local disk storage the process is allowed to use. For example:

process big_job {
    disk '2 GB'
    executor 'cirrus'

    """
    your task script here
    """
}

The following memory unit suffix can be used when specifying the disk value:

Unit

Description

B

Bytes

KB

Kilobytes

MB

Megabytes

GB

Gigabytes

TB

Terabytes

See MemoryUnit for more information.

Note

This directive is only used by certain executors. Refer to the Executors page to see which executors support this directive.

See also: cpus, memory time, queue and Dynamic computing resources.

echo

Deprecated since version 22.04.0: Use debug instead

errorStrategy

The errorStrategy directive allows you to define how an error condition is managed by the process. By default when an error status is returned by the executed script, the process stops immediately. This in turn forces the entire pipeline to terminate.

The following error strategies are available:

terminate (default)

When a task fails, terminate the pipeline immediately. Pending and running jobs are killed.

finish

When a task fails, wait for pending and running tasks to finish and then terminate the pipeline.

ignore

Ignore all task failures and complete the pipeline execution successfully.

New in version 24.05.0-edge: When the workflow.failOnIgnore config option is set to true, the pipeline will return a non-zero exit code if one or more failed tasks were ignored.

retry

When a task fails, retry it.

When setting the errorStrategy directive to ignore the process doesn’t stop on an error condition, it just reports a message notifying you of the error event.

For example:

process ignoreAnyError {
  errorStrategy 'ignore'

  script:
  <your command string here>
}

Note

By definition, a command script fails when it ends with a non-zero exit status.

The retry error strategy allows you to re-submit for execution a process returning an error condition. For example:

process retryIfFail {
  errorStrategy 'retry'

  script:
  <your command string here>
}

The number of times a failing process is re-executed is defined by the maxRetries and maxErrors directives.

Tip

More complex strategies depending on the task exit status or other parametric values can be defined using a dynamic errorStrategy. See the Dynamic directives section for details.

See also: maxErrors, maxRetries and Dynamic computing resources.

executor

The executor defines the underlying system where processes are executed. By default a process uses the executor defined globally in the nextflow.config file.

The executor directive allows you to configure what executor has to be used by the process, overriding the default configuration.

The following executors are available:

Name

Executor

awsbatch

AWS Batch service

azurebatch

Azure Batch service

condor

HTCondor job scheduler

google-lifesciences

Google Genomics Pipelines service

k8s

Kubernetes cluster

local

The computer where Nextflow is launched

lsf

Platform LSF job scheduler

moab

Moab job scheduler

nqsii

NQSII job scheduler

oge

Alias for the sge executor

pbs

PBS/Torque job scheduler

pbspro

PBS Pro job scheduler

sge

Sun Grid Engine / Open Grid Engine

slurm

SLURM workload manager

uge

Alias for the sge executor

The following example shows how to set the process’s executor:

process doSomething {
  executor 'sge'

  script:
  <your script here>
}

Note

Each executor supports additional directives and executor configuration options. Refer to the Executors page to see what each executor supports.

ext

The ext is a special directive used as namespace for user custom process directives. This can be useful for advanced configuration options. For example:

process mapping {
  container "biocontainers/star:${task.ext.version}"

  input:
  path genome
  tuple val(sampleId), path(reads)

  """
  STAR --genomeDir $genome --readFilesIn $reads ${task.ext.args ?: ''}
  """
}

In the above example, the process container version is controlled by ext.version, and the script supports additional command line arguments through ext.args.

The ext directive can be set in the process definition:

process mapping {
  ext version: '2.5.3', args: '--foo --bar'
}

Or in the Nextflow configuration:

process.ext.version = '2.5.3'
process.ext.args = '--foo --bar'

fair

New in version 22.12.0-edge.

The fair directive, when enabled, guarantees that process outputs will be emitted in the order in which they were received. For example:

process foo {
    fair true

    input:
    val x
    output:
    tuple val(task.index), val(x)

    script:
    """
    sleep \$((RANDOM % 3))
    """
}

workflow {
    channel.of('A','B','C','D') | foo | view
}

The above example produces:

[1, A]
[2, B]
[3, C]
[4, D]

label

The label directive allows the annotation of processes with mnemonic identifier of your choice. For example:

process bigTask {
  label 'big_mem'

  '''
  <task script>
  '''
}

The same label can be applied to more than a process and multiple labels can be applied to the same process using the label directive more than one time.

Note

A label must consist of alphanumeric characters or _, must start with an alphabetic character and must end with an alphanumeric character.

Labels are useful to organise workflow processes in separate groups which can be referenced in the configuration file to select and configure subset of processes having similar computing requirements. See the Process selectors documentation for details.

See also: resourceLabels

machineType

New in version 19.07.0.

The machineType can be used to specify a predefined Google Compute Platform machine type when running using the Google Life Sciences executor.

This directive is optional and if specified overrides the cpus and memory directives:

process foo {
  machineType 'n1-highmem-8'

  """
  <your script here>
  """
}

See also: cpus and memory.

maxSubmitAwait (experimental)

The maxSubmitAwait directives allows you to specify how long a task can remain in submission queue without being executed. Elapsed this time the task execution will fail.

When used along with retry error strategy, it can be useful to re-schedule the task to a difference queue or resource requirement. For example:

process foo {
  errorStrategy 'retry'
  maxSubmitAwait '10 mins'
  maxRetries 3
  queue "${task.submitAttempt==1 : 'spot-compute' : 'on-demand-compute'}"
  script:
  '''
  your_job --here
  '''
}

In the above example the task is submitted to the spot-compute on the first attempt (task.submitAttempt==1). If the task execution does not start in the 10 minutes, a failure is reported and a new submission is attempted using the queue named on-demand-compute.

maxErrors

The maxErrors directive allows you to specify the maximum number of times a process can fail when using the retry error strategy. By default this directive is disabled, you can set it as shown in the example below:

process retryIfFail {
  errorStrategy 'retry'
  maxErrors 5

  """
  echo 'do this as that .. '
  """
}

Note

This setting considers the total errors accumulated for a given process, across all instances. If you want to control the number of times a process instance (aka task) can fail, use maxRetries.

See also: errorStrategy and maxRetries.

maxForks

The maxForks directive allows you to define the maximum number of process instances that can be executed in parallel. By default this value is equals to the number of CPU cores available minus 1.

If you want to execute a process in a sequential manner, set this directive to one. For example:

process doNotParallelizeIt {
  maxForks 1

  '''
  <your script here>
  '''
}

maxRetries

The maxRetries directive allows you to define the maximum number of times a process instance can be re-submitted in case of failure. This value is applied only when using the retry error strategy. By default only one retry is allowed, you can increase this value as shown below:

process retryIfFail {
    errorStrategy 'retry'
    maxRetries 3

    """
    echo 'do this as that .. '
    """
}

Note

There is a subtle but important difference between maxRetries and the maxErrors directive. The latter defines the total number of errors that are allowed during the process execution (the same process can launch different execution instances), while the maxRetries defines the maximum number of times the same process execution can be retried in case of an error.

See also: errorStrategy and maxErrors.

memory

The memory directive allows you to define how much memory the process is allowed to use. For example:

process big_job {
    memory '2 GB'
    executor 'sge'

    """
    your task script here
    """
}

The following memory unit suffix can be used when specifying the memory value:

Unit

Description

B

Bytes

KB

Kilobytes

MB

Megabytes

GB

Gigabytes

TB

Terabytes

See MemoryUnit for more information.

See also: cpus, time, queue and Dynamic computing resources.

module

Environment Modules is a package manager that allows you to dynamically configure your execution environment and easily switch between multiple versions of the same software tool.

If it is available in your system you can use it with Nextflow in order to configure the processes execution environment in your pipeline.

In a process definition you can use the module directive to load a specific module version to be used in the process execution environment. For example:

process basicExample {
  module 'ncbi-blast/2.2.27'

  """
  blastp -query <etc..>
  """
}

You can repeat the module directive for each module you need to load. Alternatively multiple modules can be specified in a single module directive by separating all the module names by using a : (colon) character as shown below:

 process manyModules {

   module 'ncbi-blast/2.2.27:t_coffee/10.0:clustalw/2.1'

   """
   blastp -query <etc..>
   """
}

penv

The penv directive allows you to define the parallel environment to be used when submitting a parallel task to the SGE resource manager. For example:

process big_job {
  cpus 4
  penv 'smp'
  executor 'sge'

  """
  blastp -query input_sequence -num_threads ${task.cpus}
  """
}

This configuration depends on the parallel environment provided by your grid engine installation. Refer to your cluster documentation or contact your admin to learn more about this.

See also: cpus, memory, time

pod

The pod directive allows the definition of pod specific settings, such as environment variables, secrets, and config maps, when using the Kubernetes executor.

For example:

process your_task {
  pod env: 'FOO', value: 'bar'

  '''
  echo $FOO
  '''
}

The above snippet defines an environment variable named FOO whose value is bar.

When defined in the Nextflow configuration file, a pod setting can be defined as a map:

process {
  pod = [env: 'FOO', value: 'bar']
}

Or as a list of maps:

process {
  pod = [
    [env: 'FOO', value: 'bar'],
    [secret: 'my-secret/key1', mountPath: '/etc/file.txt']
  ]
}

The following options are available:

affinity: <config>

New in version 22.01.0-edge.

Specifies the pod affinity with the given configuration.

annotation: '<name>', value: '<value>'

Can be specified multiple times

Defines a pod annotation with the given name and value.

automountServiceAccountToken: true | false

New in version 22.01.0-edge.

Specifies whether to automount service account token into the pod (default: true).

config: '<configMap>/<key>', mountPath: '</absolute/path>'

Can be specified multiple times

Mounts a ConfigMap with name and optional key to the given path. If the key is omitted, the path is interpreted as a directory and all entries in the ConfigMap are exposed in that path.

csi: '<name>', mountPath: '</absolute/path>'

New in version 22.11.0-edge.

Can be specified multiple times

Mounts a CSI ephemeral volume by name to the given path.

emptyDir: <config>, mountPath: '</absolute/path>'

New in version 22.11.0-edge.

Can be specified multiple times

Mounts an emptyDir with the given configuration to the given path.

env: '<name>', config: '<configMap>/<key>'

Can be specified multiple times

Defines an environment variable whose value is defined by the given ConfigMap and key.

env: '<name>', fieldPath: '<fieldPath>'

New in version 21.09.1-edge.

Can be specified multiple times

Defines an environment variable whose value is defined by the given field path value.

For example, the following pod option:

pod = [env: 'MY_NODE_NAME', fieldPath: 'spec.nodeName']

Maps to the following pod spec:

env:
  - name: MY_NODE_NAME
    valueFrom:
      fieldRef:
        fieldPath: spec.nodeName
env: '<name>', secret: '<secret>/<key>'

Can be specified multiple times

Defines an environment variable whose value is defined by the given Secret and key.

env: '<name>', value: '<value>'

Can be specified multiple times

Defines an environment variable with the given name and value.

hostPath: '/host/absolute/path', mountPath: '</pod/absolute/path>'

New in version 23.10.0.

Can be specified multiple times

Allows creating hostPath volume and access it with the specified mountPath in the pod.

imagePullPolicy: 'IfNotPresent' | 'Always' | 'Never'

Specifies the image pull policy used by the pod to pull the container image.

imagePullSecret: '<name>'

Specifies the image pull secret used to access a private container image registry.

label: '<name>', value: '<value>'

Can be specified multiple times

Defines a pod label with the given name and value.

nodeSelector: <config>

Specifies the node selector with the given configuration.

The configuration can be a map or a string:

// map
pod = [nodeSelector: [disktype: 'ssd', cpu: 'intel']]

// string
pod = [nodeSelector: 'disktype=ssd,cpu=intel']
priorityClassName: '<name>'

New in version 22.01.0-edge.

Specifies the priority class name for pods.

privileged: true | false

New in version 22.05.0-edge.

Specifies whether the pod should run as a privileged container (default: false).

runAsUser: '<uid>'

Specifies the user ID with which to run the container. Shortcut for the securityContext option.

schedulerName: '<name>'

Specifies which scheduler is used to schedule the container.

secret: '<secret>/<key>', mountPath: '</absolute/path>'

Can be specified multiple times

Mounts a Secret with name and optional key to the given path. If the key is omitted, the path is interpreted as a directory and all entries in the Secret are exposed in that path.

securityContext: <config>

Specifies the pod security context with the given configuration.

toleration: <config>

New in version 22.04.0.

Can be specified multiple times

Specifies the pod toleration with the given configuration.

The configuration should be a map corresponding to a single toleration rule. For example, the following pod options:

pod = [
    [toleration: [key: 'key1', operator: 'Equal', value: 'value1', effect: 'NoSchedule']],
    [toleration: [key: 'key1', operator: 'Exists', effect: 'NoSchedule']],
]

Maps to the following pod spec:

tolerations:
  - key: "key1"
    operator: "Equal"
    value: "value1"
    effect: "NoSchedule"
  - key: "key1"
    operator: "Exists"
    effect: "NoSchedule"
ttlSecondsAfterFinished

New in version 24.02.0-edge.

Specifies the TTL mechanism for finished jobs in seconds. Applies to both successful and failed jobs.

volumeClaim: '<name>', mountPath: '</absolute/path>' [, subPath: '<path>', readOnly: true | false]

Can be specified multiple times

Mounts a Persistent volume claim with the given name to the given path.

The subPath option can be used to mount a sub-directory of the volume instead of its root.

The readOnly option can be used to mount the volume as read-only (default: false)

publishDir

The publishDir directive allows you to publish the process output files to a specified folder. For example:

process foo {
    publishDir '/data/chunks'

    output:
    path 'chunk_*'

    '''
    printf 'Hola' | split -b 1 - chunk_
    '''
}

The above example splits the string Hola into file chunks of a single byte. When complete the chunk_* output files are published into the /data/chunks folder.

Note

Only files that match the declaration in the output block are published, not all the outputs of the process.

Tip

The publishDir directive can be specified more than once in order to publish output files to different target directories based on different rules.

By default files are published to the target folder creating a symbolic link for each process output that links the file produced into the process working directory. This behavior can be modified using the mode option, for example:

process foo {
    publishDir '/data/chunks', mode: 'copy', overwrite: false

    output:
    path 'chunk_*'

    '''
    printf 'Hola' | split -b 1 - chunk_
    '''
}

Warning

Files are copied into the specified directory in an asynchronous manner, so they may not be immediately available in the publish directory at the end of the process execution. For this reason, downstream processes should not try to access output files through the publish directory, but through channels.

Available options:

contentType

New in version 22.10.0.

Experimental: currently only supported for S3.

Allow specifying the media content type of the published file a.k.a. MIME type. If set to true, the content type is inferred from the file extension (default: false).

enabled

Enable or disable the publish rule depending on the boolean value specified (default: true).

failOnError

Changed in version 24.03.0-edge: The default value was change from false to true

When true abort the execution if some file can’t be published to the specified target directory or bucket for any cause (default: true)

mode

The file publishing method. Can be one of the following values:

  • 'copy': Copies the output files into the publish directory.

  • 'copyNoFollow': Copies the output files into the publish directory without following symlinks ie. copies the links themselves.

  • 'link': Creates a hard link in the publish directory for each output file.

  • 'move': Moves the output files into the publish directory. Note: this is only supposed to be used for a terminal process i.e. a process whose output is not consumed by any other downstream process.

  • 'rellink': Creates a relative symbolic link in the publish directory for each output file.

  • 'symlink': Creates an absolute symbolic link in the publish directory for each output file (default).

overwrite

When true any existing file in the specified folder will be overridden (default: true during normal pipeline execution and false when pipeline execution is resumed).

path

Specifies the directory where files need to be published. Note: the syntax publishDir '/some/dir' is a shortcut for publishDir path: '/some/dir'.

pattern

Specifies a glob file pattern that selects which files to publish from the overall set of output files.

saveAs

A closure which, given the name of the file being published, returns the actual file name or a full path where the file is required to be stored. This can be used to rename or change the destination directory of the published files dynamically by using a custom strategy. Return the value null from the closure to not publish a file. This is useful when the process has multiple output files, but you want to publish only some of them.

storageClass

New in version 22.12.0-edge.

Experimental: currently only supported for S3.

Allow specifying the storage class to be used for the published file.

tags

New in version 21.12.0-edge.

Experimental: currently only supported for S3.

Allow the association of arbitrary tags with the published file e.g. tags: [FOO: 'Hello world'].

queue

The queue directive allows you to set the queue where jobs are scheduled when using a grid based executor in your pipeline. For example:

process grid_job {
    queue 'long'
    executor 'sge'

    """
    your task script here
    """
}

Multiple queues can be specified by separating their names with a comma for example:

process grid_job {
    queue 'short,long,cn-el6'
    executor 'sge'

    """
    your task script here
    """
}

Note

This directive is only used by certain executors. Refer to the Executors page to see which executors support this directive.

resourceLabels

New in version 22.09.1-edge.

The resourceLabels directive allows you to specify custom name-value pairs that Nextflow applies to the computing resource used to carry out the process execution. Resource labels can be specified using the syntax shown below:

process my_task {
    resourceLabels region: 'some-region', user: 'some-username'

    '''
    <task script>
    '''
}

The limits and the syntax of the corresponding cloud provider should be taken into consideration when using resource labels.

Resource labels are currently supported by the following executors:

New in version 23.09.0-edge: Resource labels are supported for Azure Batch when using automatic pool creation.

Resource labels in Azure are added to pools, rather than jobs, in order to facilitate cost analysis. A new pool will be created for each new set of resource labels, therefore it is recommended to also set azure.batch.deletePoolsOnCompletion = true when using process-specific resource labels.

See also: label

resourceLimits

New in version 24.04.0.

The resourceLimits directive allows you to specify environment-specific limits for task resource requests. Resource limits can be specified in a process as follows:

process my_task {
  resourceLimits cpus: 24, memory: 768.GB, time: 72.h

  script:
  '''
  your_command --here
  '''
}

Or in the Nextflow configuration:

process {
    resourceLimits = [ cpus: 24, memory: 768.GB, time: 72.h ]
}

Resource limits can be defined for the following directives:

Resource limits are a useful way to specify environment-specific limits alongside tasks with dynamic resources. Normally, if a task requests more resources than can be provisioned (e.g. a task requests 32 cores but the largest node in the cluster has 24), the task will either fail or cause the pipeline to hang forever as it will never be scheduled. If the resourceLimits directive is defined with these limits, the task resources will be automatically reduced to comply with these limits before the job is submitted.

scratch

The scratch directive allows you to execute the process in a temporary folder that is local to the execution node.

This is useful when your pipeline is launched by using a grid executor, because it allows you to decrease the NFS overhead by running the pipeline processes in a temporary directory in the local disk of the actual execution node. Only the files declared as output in the process definition will be copied in the pipeline working area.

In its basic form simply specify true at the directive value, as shown below:

process simpleTask {
  scratch true

  output:
  path 'data_out'

  '''
  <task script>
  '''
}

By doing this, it tries to execute the script in the directory defined by the variable $TMPDIR in the execution node. If this variable does not exist, it will create a new temporary directory by using the Linux command mktemp.

Note

Cloud-based executors use scratch = true by default, since the work directory resides in object storage.

The following values are supported:

false

Do not use a scratch directory.

true

Create a scratch directory in the directory defined by the $TMPDIR environment variable, or $(mktemp /tmp) if $TMPDIR is not set.

'$YOUR_VAR'

Create a scratch directory in the directory defined by the given environment variable, or $(mktemp /tmp) if that variable is not set. The value must use single quotes, otherwise the environment variable will be evaluated in the pipeline script context.

'/my/tmp/path'

Create a scratch directory in the specified directory.

'ram-disk'

Create a scratch directory in the RAM disk /dev/shm/.

shell

The shell directive allows you to define a custom shell command for process scripts. By default, script blocks are executed with /bin/bash -ue.

process doMoreThings {
    shell '/bin/bash', '-euo', 'pipefail'

    '''
    your_command_here
    '''
}

The same directive could be specified in your Nextflow configuration as follows:

process.shell = ['/bin/bash', '-euo', 'pipefail']

spack

The spack directive allows for the definition of the process dependencies using the Spack package manager.

Nextflow automatically sets up an environment for the given package names listed by in the spack directive. For example:

process foo {
    spack 'bwa@0.7.15'

    '''
    your_command --here
    '''
}

Multiple packages can be specified separating them with a blank space, e.g. bwa@0.7.15 fastqc@0.11.5.

The spack directive also allows the specification of a Spack environment file path or the path of an existing environment directory. See the Spack environments page for further details.

stageInMode

The stageInMode directive defines how input files are staged into the process work directory. The following values are allowed:

'copy'

Input files are staged in the process work directory by creating a copy.

'link'

Input files are staged in the process work directory by creating a hard link for each of them.

'rellink'

Input files are staged in the process work directory by creating a symbolic link with a relative path for each of them.

'symlink'

Input files are staged in the process work directory by creating a symbolic link with an absolute path for each of them (default).

stageOutMode

The stageOutMode directive defines how output files are staged out from the scratch directory to the process work directory. The following values are allowed:

'copy'

Output files are copied from the scratch directory to the work directory.

'fcp'

New in version 23.02.0-edge.

Output files are copied from the scratch directory to the work directory by using the fcp utility (note: it must be available in your cluster computing nodes).

'move'

Output files are moved from the scratch directory to the work directory.

'rclone'

New in version 23.01.0-edge.

Output files are copied from the scratch directory to the work directory by using the rclone utility (note: it must be available in your cluster computing nodes).

'rsync'

Output files are copied from the scratch directory to the work directory by using the rsync utility.

See also: scratch.

storeDir

The storeDir directive allows you to define a directory that is used as a permanent cache for your process results.

In more detail, it affects the process execution in two main ways:

  1. The process is executed only if the files declared in the output block do not exist in the directory specified by the storeDir directive. When the files exist the process execution is skipped and these files are used as the actual process result.

  2. Whenever a process is successfully completed the files listed in the output block are moved into the directory specified by the storeDir directive.

The following example shows how to use the storeDir directive to create a directory containing a BLAST database for each species specified by an input parameter:

process formatBlastDatabases {
  storeDir '/db/genomes'

  input:
  path species

  output:
  path "${dbName}.*"

  script:
  dbName = species.baseName
  """
  makeblastdb -dbtype nucl -in ${species} -out ${dbName}
  """
}

Warning

The storeDir directive is meant for long-term process caching and should not be used to publish output files or organize outputs into a semantic directory structure. In those cases, use the publishDir directive instead.

Note

The use of AWS S3 paths is supported, however it requires the installation of the AWS CLI (i.e. aws) in the target compute node.

tag

The tag directive allows you to associate each process execution with a custom label, so that it will be easier to identify them in the log file or in the trace execution report. For example:

process foo {
  tag "$code"

  input:
  val code

  """
  echo $code
  """
}

workflow {
  Channel.of('alpha', 'gamma', 'omega') | foo
}

The above snippet will print a log similar to the following one, where process names contain the tag value:

[6e/28919b] Submitted process > foo (alpha)
[d2/1c6175] Submitted process > foo (gamma)
[1c/3ef220] Submitted process > foo (omega)

See also Trace execution report

time

The time directive allows you to define how long a process is allowed to run. For example:

process big_job {
    time '1h'

    """
    your task script here
    """
}

The following time unit suffixes can be used when specifying the duration value:

Unit

Description

ms, milli, millis

Milliseconds

s, sec, second, seconds

Seconds

m, min, minute, minutes

Minutes

h, hour, hours

Hours

d, day, days

Days

Multiple units can be used in a single declaration, for example: '1day 6hours 3minutes 30seconds'

See Duration for more information.

Note

This directive is only used by certain executors. Refer to the Executors page to see which executors support this directive.

See also: cpus, memory, queue and Dynamic computing resources.

Dynamic directives

A directive can be assigned dynamically, during the process execution, so that its actual value can be evaluated based on the process inputs.

In order to be defined in a dynamic manner, the directive’s value needs to be expressed using a closure, as in the following example:

process foo {
  executor 'sge'
  queue { entries > 100 ? 'long' : 'short' }

  input:
  tuple val(entries), path('data.txt')

  script:
  """
  < your job here >
  """
}

In the above example, the queue directive is evaluated dynamically, depending on the input value entries. When it is larger than 100, jobs will be submitted to the long queue, otherwise the short queue will be used.

All directives can be assigned a dynamic value except the following:

Tip

Assigning a string value with one or more variables is always resolved in a dynamic manner, and therefore is equivalent to the above syntax. For example, the above directive can also be written as:

queue "${ entries > 100 ? 'long' : 'short' }"

Note, however, that the latter syntax can be used both for a directive’s main argument (as in the above example) and for a directive’s optional named attributes, whereas the closure syntax is only resolved dynamically for a directive’s main argument.

Tip

You can retrieve the current value of a dynamic directive in the process script by using the implicit variable task, which holds the directive values defined in the current task. For example:

process foo {
  queue { entries > 100 ? 'long' : 'short' }

  input:
  tuple val(entries), path('data.txt')

  script:
  """
  echo Current queue: ${task.queue}
  """
}

Dynamic computing resources

It’s a very common scenario that different instances of the same process may have very different needs in terms of computing resources. In such situations requesting, for example, an amount of memory too low will cause some tasks to fail. Instead, using a higher limit that fits all the tasks in your execution could significantly decrease the execution priority of your jobs.

The Dynamic directives evaluation feature can be used to modify the amount of computing resources requested in case of a process failure and try to re-execute it using a higher limit. For example:

process foo {
    memory { 2.GB * task.attempt }
    time { 1.hour * task.attempt }

    errorStrategy { task.exitStatus in 137..140 ? 'retry' : 'terminate' }
    maxRetries 3

    script:
    <your job here>
}

In the above example the memory and execution time limits are defined dynamically. The first time the process is executed the task.attempt is set to 1, thus it will request a two GB of memory and one hour of maximum execution time.

If the task execution fail reporting an exit status in the range between 137 and 140, the task is re-submitted (otherwise terminates immediately). This time the value of task.attempt is 2, thus increasing the amount of the memory to four GB and the time to 2 hours, and so on.

The directive maxRetries set the maximum number of time the same task can be re-executed.

Dynamic Retry with backoff

There are cases in which the required execution resources may be temporary unavailable e.g. network congestion. In these cases immediately re-executing the task will likely result in the identical error. A retry with an exponential backoff delay can better recover these error conditions:

process foo {
  errorStrategy { sleep(Math.pow(2, task.attempt) * 200 as long); return 'retry' }
  maxRetries 5

  script:
  '''
  your_command --here
  '''
}