Documentation vs reality: workflow.container

[keiranmraine]

Bug report

Expected behavior and actual behavior

The metadata documentation indicates that workflow.container returns a map of all containers:

Docker image used to run workflow tasks. When more than one image is used it returns a map object containing [process name, image name] pair entries.

A map is not returned when multiple containers defined, only the top level value when defined (process.container).

If top level process.container has not been defined we get [:].

nextflow.config

params {
    // output folder options
    outdir          = 'results'
    tracedir        = "${params.outdir}/pipeline_info"
}

docker.enabled = true

process {
    container     = 'ubuntu:18.04' // gives this value as result, comment to get `[:]`

    withName: bash_step {
        executor = 'local'
        container = null
    }

    withName: job_a {
        container = 'quay.io/wtsicgp/expansion_hunter:5.0.0'
    }

    withName: job_b {
        container = 'quay.io/wtsicgp/pcap-core:5.7.0'
    }
}

main.nf

#!/usr/bin/env nextflow
nextflow.enable.dsl=2

process bash_step {
    input:
        val(example)
    output:
        path '*.out', emit: wfl
    script:
        """
        echo ${workflow.container} > ${example}.out
        """
}

process job_a {
    input:
        path(wfl)
    output:
        path '*.out.gz', emit: gzip
    script:
        """
        gzip -c ${wfl} > ${wfl}.gz
        """
}

process job_b {
    input:
        path(wfl)
        path(gzip)
    
    output:
        path 'compress.comp'

    publishDir "${params.tracedir}", mode: "copy"

    shell:
        '''
        echo native: $(wc -c !{wfl}) > compress.comp
        echo compressed: $(wc -c !{gzip}) >> compress.comp
        echo "## content follows ##" >> compress.comp
        cat !{wfl} >> compress.comp
        echo "## EOC ##" >> compress.comp
        '''
}

workflow {
    main:
        bash_step('bob')
        job_a(bash_step.out.wfl)
        job_b(bash_step.out.wfl, job_a.out.gzip)
}

Run with:

nextflow run -stub-run main.nf
# cat results/pipeline_info/compress.comp

Gives:

native: 13 bob.out
compressed: 41 bob.out.gz
## content follows ##
ubuntu:18.04
## EOC ##

Would expect:

[bash_step : null, job_a : 'quay.io/wtsicgp/expansion_hunter:5.0.0', job_b: 'quay.io/wtsicgp/pcap-core:5.7.0']

(possibly no entry for bash_step)

Steps to reproduce the problem

As above

Program output

As above

Environment

• Nextflow version: 22.10.2.5832
• Java version: openjdk version "17.0.5" 2022-10-18
• Operating system: Linux (GitHub CodeSpaces)
• Bash version: GNU bash, version 5.1.4(1)-release (x86_64-pc-linux-gnu)

Additional context

Appears related to #1758, if this is not a suitable way to introspect this value please update documentation.

[Z-Zen]

Bumping this

One could do like this:

#!/usr/bin/env nextflow
nextflow.enable.dsl=2

// creating an empty dictionary
mapContainers = [:]

process bash_step {
    input:
        val(example)
    output:
        path '*.out', emit: wfl
    script:
    if (!mapContainers.containsKey(task.container)) {
      mapContainers[task.container] = task.process
     }
        """
        echo ${workflow.container} > ${example}.out
        """
}

process job_a {
    input:
        path(wfl)
    output:
        path '*.out.gz', emit: gzip
    script:
    if (!mapContainers.containsKey(task.container)) {
      mapContainers[task.container] = task.process
     }
        """
        gzip -c ${wfl} > ${wfl}.gz
        """
}

process job_b {
    input:
        path(wfl)
        path(gzip)
    
    output:
        path 'compress.comp'

    publishDir "${params.tracedir}", mode: "copy"

    shell:
    if (!mapContainers.containsKey(task.container)) {
      mapContainers[task.container] = task.process
     }
        '''
        echo native: $(wc -c !{wfl}) > compress.comp
        echo compressed: $(wc -c !{gzip}) >> compress.comp
        echo "## content follows ##" >> compress.comp
        cat !{wfl} >> compress.comp
        echo "## EOC ##" >> compress.comp
        '''
}

workflow {
    main:
        bash_step('bob')
        job_a(bash_step.out.wfl)
        job_b(bash_step.out.wfl, job_a.out.gzip)
}


// print the mapContainers here
workflow.onComplete {
  def status = "NA"
  if(workflow.success) {

    status = "SUCCESS"

    println("""
    Workflow container    : ${mapContainers}
    """
    )
  }
}

the problem is that this approach won't work when using with modules....

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

[bentsherman]

The relevant code is here:

nextflow/modules/nextflow/src/main/groovy/nextflow/Session.groovy

Lines 1270 to 1274 in a66b0e6

	config.process.each { String name, value ->
	if( name.startsWith('$') && value instanceof Map && value.container ) {
	result[name] = resolveClosure(value.container)
	}
	}

However when I inspect the process config at this point, I get the following:

process.container = ubuntu:18.04
process.withName:bash_step = [executor:local, container:null]
process.withName:job_a = [container:quay.io/wtsicgp/expansion_hunter:5.0.0]
process.withName:job_b = [container:quay.io/wtsicgp/pcap-core:5.7.0]

@pditommaso can you explain the logic here? I don't know what the $ means, but it doesn't seem to be valid anymore.

[pditommaso]

I have to say I have no memory about that. Tho there's even a test for that

nextflow/modules/nextflow/src/test/groovy/nextflow/SessionTest.groovy

Lines 513 to 559 in 3051cd1

	def 'should fetch containers definition' () {
	String text
	when:
	text = '''
	process.container = 'beta'
	'''
	then:
	new Session(cfg(text)).fetchContainers() == 'beta'
	when:
	text = '''
	process {
	$proc1 { container = 'alpha' }
	$proc2 { container ='beta' }
	}
	'''
	then:
	new Session(cfg(text)).fetchContainers() == ['$proc1': 'alpha', '$proc2': 'beta']
	when:
	text = '''
	process {
	$proc1 { container = 'alpha' }
	$proc2 { container ='beta' }
	}
	process.container = 'gamma'
	'''
	then:
	new Session(cfg(text)).fetchContainers() == ['$proc1': 'alpha', '$proc2': 'beta', default: 'gamma']
	when:
	text = '''
	process.container = { "ngi/rnaseq:${workflow.getRevision() ?: 'latest'}" }
	'''
	def meta = Mock(WorkflowMetadata); meta.getRevision() >> '1.2'
	def session = new Session(cfg(text))
	session.binding.setVariable('workflow',meta)
	then:
	session.fetchContainers() == 'ngi/rnaseq:1.2'
	}

[bentsherman]

The original container code was added long ago at a time when process names had to have a $ prefix in the config (docs).

I will draft a fix.