Nextflow

ICA supports running pipelines defined using Nextflow. See this tutorial for an example.

In order to run Nextflow pipelines, the following process-level attributes within the Nextflow definition must be considered.

System Information

Info	Details
Nextflow version	20.10.0 (deprecated ⚠️), 22.04.3 (supported ✅), 24.10.2 (default ⭐)
Executor	Kubernetes

The following table shows when which Nextflow version is

• default (⭐) This version will be proposed when creating a new Nextflow pipeline.

• supported (✅) This version can be selected when you do not want the default Nextflow version.

• deprecated (⚠️) This version can not be selected for new pipelines, but pipelines using this version will still work.

• removed (❌). This version can not be selected when creating new pipelines and pipelines using this version will no longer work.

The switchover always happens in the January release of that year.

Nextflow Version	2025	2026	2027	2028
v20.10.0	⚠️	❌	❌	❌
v22.04.3	✅	⚠️	❌	❌
v24.10.2	⭐	⭐	✅	✅
v25.10.x		✅	⭐	✅
v26.10.x			✅	⭐
v27.10.x				✅

Nextflow Version

You can select the Nextflow version while building a pipeline as follows:

interface
GUI	Select the Nextflow version at Projects > your_project > flow > pipelines > your_pipeline > Details tab.
API	Select the Nextflow version by setting it in the optional field "pipelineLanguageVersionId". When not set, a default Nextflow version will be used for the pipeline.

Compute Node

For each compute type, you can choose between the scheduler.illumina.com/lifecycle: standard (default - AWS on-demand) or scheduler.illumina.com/lifecycle: economy (AWS spot instance) tiers.

Compute Type

To specify a compute type for a Nextflow process, use the pod directive within each process. Set the annotation to scheduler.illumina.com/presetSize and the value to the desired compute type. A list of available compute types can be found here. The default compute type, when this directive is not specified, is standard-small (2 CPUs and 8 GB of memory).

pod annotation: 'scheduler.illumina.com/presetSize', value: 'fpga2-medium'

Often, there is a need to select the compute size for a process dynamically based on user input and other factors. The Kubernetes executor used on ICA does not use the cpu and memorydirectives, so instead, you can dynamically set the pod directive, as mentioned here. e.g.

process foo {
    // Assuming that params.compute_size is set to a valid size such as 'standard-small', 'standard-medium', etc.
    pod annotation: 'scheduler.illumina.com/presetSize', value: "${params.compute_size}"
}

Additionally, it can also be specified in the configuration file. Example configuration file:

// Set the default pod
pod = [
    annotation: 'scheduler.illumina.com/presetSize',
    value     : 'standard-small'
]

withName: 'big_memory_process' {
    pod = [
        annotation: 'scheduler.illumina.com/presetSize',
        value     : 'himem-large'
    ]
}

// Use an FPGA2 instance for dragen processes
withLabel: 'dragen' {
    pod = [
        annotation: 'scheduler.illumina.com/presetSize',
        value     : 'fpga2-medium'
    ]
}

Inputs

Inputs are specified via the XML input form or JSON-based input form. The specified code in the XML will correspond to the field in the params object that is available in the workflow. Refer to the tutorial for an example.

Outputs

Outputs for Nextflow pipelines are uploaded from the out folder in the attached shared filesystem. The publishDir directive can be used to symlink (recommended), copy or move data to the correct folder. Data will be uploaded to the ICA project after the pipeline execution completes.

publishDir 'out', mode: 'symlink'

Nextflow version 20.10.10 (Deprecated)

For Nextflow version 20.10.10 on ICA, using the "copy" method in the publishDir directive for uploading output files that consume large amounts of storage may cause workflow runs to complete with missing files. The underlying issue is that file uploads may silently fail (without any error messages) during the publishDir process due to insufficient disk space, resulting in incomplete output delivery.

Solutions:

1. Use "symlink" instead of "copy" in the publishDir directive. Symlinking creates a link to the original file rather than copying it, which doesn’t consume additional disk space. This can prevent the issue of silent file upload failures due to disk space limitations.

2. Use Nextflow 22.04.0 or later and enable the "failOnError" publishDir option. This option ensures that the workflow will fail and provide an error message if there's an issue with publishing files, rather than completing silently without all expected outputs.

Nextflow Configuration

During execution, the Nextflow pipeline runner determines the environment settings based on values passed via the command-line or via a configuration file (see Nextflow Configuration documentation). When creating a Nextflow pipeline, use the nextflow.config tab in the UI (or API) to specify a nextflow configuration file to be used when launching the pipeline.

Syntax highlighting is determined by the file type, but you can select alternative syntax highlighting with the drop-down selection list.

nextflowconfig-0

If no Docker image is specified, Ubuntu will be used as default.

The following configuration settings will be ignored if provided as they are overridden by the system:

executor.name
executor.queueSize
k8s.namespace
k8s.serviceAccount
k8s.launchDir
k8s.projectDir
k8s.workDir
k8s.storageClaimName
k8s.storageMountPath
trace.enabled
trace.file
trace.fields
timeline.enabled
timeline.file
report.enabled
report.file
dag.enabled
dag.file

Best Practices

Process Time

Setting a timeout to between 2 and 4 times the expected processing time with the time directive for processes or task will ensure that no stuck processes remain indefinitely. Stuck process keep incurring costs for the occupied resources, so if the process can not complete within that timespan, it is safer and more economical to end the process and retry.

Sample Sheet File Ingestion

When you want to use a sample sheet with references to files as Nextflow input, add an extra input to the pipeline. This extra input lets the user select the samplesheet-mentioned files from their project. At run time, those files will get staged in the working directory, and when Nextflow parses the samplesheet and looks for those files without paths, it will find them there. You can not use file paths in a sample sheet without selecting the files in the input form because files are only passed as file/folder ids in the API payload when the analysis is launched.

You can include public data such as http urls because Nextflow is able download those. Nextflow is also able to download publicly accessible S3 urls (s3://...). You can not use Illumina's urn:ilmn:ica:region:... structure.