Plugins

This page describes how to create, test, and publish third-party plugins.

Plugin structure

The best way to get started with your own plugin is to refer to the nf-hello repository. This repository provides a minimal plugin implementation with several examples of different extension points and instructions for building, testing, and publishing.

Plugins can be written in Java or Groovy.

The minimal dependencies are as follows:

dependencies {
    compileOnly project(':nextflow')
    compileOnly 'org.slf4j:slf4j-api:1.7.10'
    compileOnly 'org.pf4j:pf4j:3.4.1'

    testImplementation project(':nextflow')
    testImplementation "org.codehaus.groovy:groovy:4.0.23"
    testImplementation "org.codehaus.groovy:groovy-nio:4.0.23"
}

The plugin subproject directory name must begin with the prefix nf- and must include a file named src/resources/META-INF/MANIFEST.MF which contains the plugin metadata. The manifest content looks like the following:

Manifest-Version: 1.0
Plugin-Class: the.plugin.ClassName
Plugin-Id: the-plugin-id
Plugin-Provider: Your Name or Organization
Plugin-Version: 0.0.0

Extension points

Nextflow’s plugin system exposes a variety of extension points for plugins. This section describes how to use these extension points when writing a plugin, as well as how they are used in a pipeline.

Note

If you would like to implement something in a plugin that isn’t covered by any of the following sections, feel free to create an issue on GitHub and describe your use case. In general, any class in the Nextflow codebase that implements ExtensionPoint can be extended by a plugin, and existing plugins are a great source of examples when writing new plugins.

Note

Plugin extension points must be added to extensions.idx in the plugin repository to make them discoverable. See the nf-hello plugin for an example.

Commands

Plugins can define custom CLI commands that can be executed with the nextflow plugin command.

To implement a plugin-specific command, implement the PluginExecAware interface in your plugin entrypoint (the class that extends BasePlugin). Alternatively, you can implement the PluginAbstractExec trait, which provides an abstract implementation with some boilerplate code. This trait requires you to implement two methods, getCommands() and exec():

import nextflow.cli.PluginAbstractExec
import nextflow.plugin.BasePlugin

class MyPlugin extends BasePlugin implements PluginAbstractExec {
    @Override
    List<String> getCommands() {
        [ 'hello' ]
    }

    @Override
    int exec(String cmd, List<String> args) {
        if( cmd == 'hello' ) {
            println "Hello! You gave me these arguments: ${args.join(' ')}"
            return 0
        }
        else {
            System.err.println "Invalid command: ${cmd}"
            return 1
        }
    }
}

You can then execute this command using the nextflow plugin command:

nextflow plugin my-plugin:hello --foo --bar

See the plugin CLI command for usage information.

Configuration

Plugins can access the resolved Nextflow configuration through the session object using session.config.navigate(). Several extension points provide the session object for this reason. This method allows you to query any configuration option in a safe manner – if the option isn’t defined, it will return null. A common practice is to define any configuration for your plugin in a custom config scope.

For example, you can query a config option in a trace observer hook:

import nextflow.Session
import nextflow.trace.TraceObserver

class MyObserver implements TraceObserver {

    @Override
    void onFlowCreate(Session session) {
        final message = session.config.navigate('myplugin.create.message')
        println message
    }
}

You can then set this option in your config file:

// dot syntax
myplugin.create.message = "I'm alive!"

// closure syntax
myplugin {
    create {
        message = "I'm alive!"
    }
}

Executors

Plugins can define custom executors that can then be used with the executor process directive.

To implement an executor, create a class in your plugin that extends the Executor class and implements the ExtensionPoint interface. Add the @ServiceName annotation to your class with the name of your executor:

import nextflow.executor.Executor
import nextflow.util.ServiceName
import org.pf4j.ExtensionPoint

@ServiceName('my-executor')
class MyExecutor extends Executor implements ExtensionPoint {

    // ...

}

You can then use this executor in your pipeline:

process foo {
    executor 'my-executor'
}

Tip

Refer to the source code of Nextflow’s built-in executors to see how to implement the various components of an executor. You might be able to implement most of your executor by simply reusing existing code.

Functions

New in version 22.09.0-edge.

Plugins can define custom functions, which can then be included in Nextflow pipelines.

To implement a custom function, create a class in your plugin that extends the PluginExtensionPoint class, and implement your function with the Function annotation:

import nextflow.Session
import nextflow.plugin.extension.Function
import nextflow.plugin.extension.PluginExtensionPoint

class MyExtension extends PluginExtensionPoint {

    @Override
    void init(Session session) {}

    @Function
    String reverseString(String origin) {
        origin.reverse()
    }

}

You can then use this function in your pipeline:

include { reverseString } from 'plugin/my-plugin'

channel.of( reverseString('hi') )

You can also use an alias:

include { reverseString as anotherReverseMethod } from 'plugin/my-plugin'

Operators

New in version 22.04.0.

Plugins can define custom channel factories and operators, which can then be included into Nextflow pipelines.

To implement a custom factory or operator, create a class in your plugin that extends the PluginExtensionPoint class, and implement your function with the Factory or Operator annotation:

import groovyx.gpars.dataflow.DataflowReadChannel
import groovyx.gpars.dataflow.DataflowWriteChannel
import nextflow.Session
import nextflow.plugin.extension.Factory
import nextflow.plugin.extension.Operator
import nextflow.plugin.extension.PluginExtensionPoint

class MyExtension extends PluginExtensionPoint {

    @Override
    void init(Session session) {}

    @Factory
    DataflowWriteChannel fromQuery(Map opts, String query) {
        // ...
    }

    @Operator
    DataflowWriteChannel sqlInsert(DataflowReadChannel source, Map opts) {
        // ...
    }

}

You can then use them in your pipeline:

include { sqlInsert; fromQuery as fromTable } from 'plugin/nf-sqldb'

def sql = 'select * from FOO'
channel
    .fromTable(sql, db: 'test', emitColumns: true)
    .sqlInsert(into: 'BAR', columns: 'id', db: 'test')

The above snippet is based on the nf-sqldb plugin. The fromQuery factory is included under the alias fromTable.

Process directives

Plugins that implement a custom executor will likely need to access process directives that affect the task execution. When an executor receives a task, the process directives can be accessed through that task’s configuration. As a best practice, custom executors should try to support all process directives that have executor-specific behavior and are relevant to your executor.

Nextflow does not provide the ability to define custom process directives in a plugin. Instead, you can use the ext directive to provide custom process settings to your executor. Try to use specific names that are not likely to conflict with other plugins or existing pipelines.

For example, a custom executor can use existing process directives as well as a custom setting through the ext directive:

class MyExecutor extends Executor {

    @Override
    TaskHandler createTaskHandler(TaskRun task) {
        final cpus = task.config.cpus
        final memory = task.config.memory
        final myOption = task.config.ext.myOption

        println "This task is configured with cpus=${cpus}, memory=${memory}, myOption=${myOption}"

        // ...
    }

    // ...

}

Trace observers

A trace observer in Nextflow is an entity that can listen and react to workflow events, such as when a workflow starts, a task completes, a file is published, etc. Several components in Nextflow, such as the execution report and DAG visualization, are implemented as trace observers.

Plugins can define custom trace observers that react to workflow events with custom behavior. To implement a trace observer, create a class that implements the TraceObserver trait and another class that implements the TraceObserverFactory interface. Implement any of the hooks defined in TraceObserver, and implement the create() method in your observer factory:

// MyObserverFactory.groovy
import nextflow.Session
import nextflow.trace.TraceObserver
import nextflow.trace.TraceObserverFactory

class MyObserverFactory implements TraceObserverFactory {

    @Override
    Collection<TraceObserver> create(Session session) {
        final enabled = session.config.navigate('myplugin.enabled')
        return enabled ? [ new MyObserver() ] : []
    }
}

// MyObserver.groovy
import java.nio.file.Path

import nextflow.processor.TaskHandler
import nextflow.trace.TraceObserver
import nextflow.trace.TraceRecord

class MyObserver implements TraceObserver {
    
    @Override
    void onFlowBegin() {
        println "Okay, let's begin!"
    }

    @Override
    void onProcessComplete(TaskHandler handler, TraceRecord trace) {
        println "I completed a task! It's name is '${handler.task.name}'"
    }

    @Override
    void onProcessCached(TaskHandler handler, TraceRecord trace) {
        println "I found a task in the cache! It's name is '${handler.task.name}'"
    }

    @Override
    void onFilePublish(Path destination, Path source) {
        println "I published a file! It's located at ${path.toUriString()}"
    }
    
    @Override
    void onFlowError(TaskHandler handler, TraceRecord trace) {
        println "Uh oh, something went wrong..."
    }

    @Override
    void onFlowComplete() {
        println 'All done!'
    }
}

You can then use your trace observer by simply enabling the plugin in your pipeline. In the above example, the observer must also be enabled with a config option:

myplugin.enabled = true

Refer to the TraceObserver source code for descriptions of the available workflow events.

Publishing plugins

Nextflow resolves plugins through a plugin registry, which stores metadata for each plugin version, including the publishing date, checksum, and download URL for the plugin binary. The default registry is located on GitHub at nextflow-io/plugins, specifically this file.

To publish a plugin, you must create the plugin release, publish it to GitHub, and make a pull request against the main plugin registry with the metadata for your plugin release.

A plugin release is a ZIP archive containing the compiled plugin classes, the required dependencies, and a JSON file with the plugin metadata. See the nf-hello example to see how to create a plugin release.

Here is an example meta file for a plugin release:

{
    "version": "0.2.0",
    "url": "https://github.com/nextflow-io/nf-amazon/releases/download/0.2.0/nf-amazon-0.2.0.zip",
    "date": "2020-10-12T10:05:44.28+02:00",
    "sha512sum": "9e9e33695c1a7c051271..."
}

Testing plugins

Plugins must be declared in the nextflow.config file using the plugins scope, for example:

plugins {
    id 'nf-amazon@0.2.0'
}

If a plugin is not locally available, Nextflow checks the repository index for the download URL, downloads and extracts the plugin archive, and installs the plugin into the directory specified by NXF_PLUGINS_DIR (default: ${NXF_HOME}/plugins).

Since each Nextflow run can have a different set of plugins (and versions), each run keeps a local plugins directory called .nextflow/plr/<session-id> which links the exact set of plugins required for the given run.

Additionally, the “default plugins” (defined in the Nextflow resources file modules/nextflow/src/main/resources/META-INF/plugins-info.txt) are always made available for use. To disable the use of default plugins, set the environment variable NXF_PLUGINS_DEFAULT=false.

When running in development mode, the plugin system uses the DevPluginClasspath to load plugin classes from each plugin project build path, e.g. plugins/nf-amazon/build/classes and plugins/nf-amazon/build/target/libs (for deps libraries).

Environment variables

The following environment variables are available when developing and testing plugins:

NXF_PLUGINS_MODE

The plugin execution mode, either prod for production or dev for development (see below for details).

NXF_PLUGINS_DIR

The path where the plugin archives are loaded and stored (default: $NXF_HOME/plugins in production, ./plugins in development).

NXF_PLUGINS_DEFAULT

Whether to use the default plugins when no plugins are specified in the Nextflow configuration (default: true).

NXF_PLUGINS_DEV

Comma-separated list of development plugin root directories.

NXF_PLUGINS_TEST_REPOSITORY

New in version 23.04.0.

Comma-separated list of URIs for additional plugin registries or meta files, which will be used in addition to the default registry.

The URI should refer to a plugins repository JSON file or a specific plugin JSON meta file. In the latter case it should match the pattern https://host.name/some/path/<plugin id>-X.Y.Z-meta.json.

For example:

# custom plugin repository at https://github.com/my-org/plugins
export NXF_PLUGINS_TEST_REPOSITORY="https://raw.githubusercontent.com/my-org/plugins/main/plugins.json"

# custom plugin release
export NXF_PLUGINS_TEST_REPOSITORY="https://github.com/nextflow-io/nf-hello/releases/download/0.3.0/nf-hello-0.3.0-meta.json"

nextflow run <pipeline> -plugins nf-hello

This variable is useful for testing a plugin release before publishing it to the main registry.

Gradle Tasks

The following build tasks are defined in the build.gradle of each core plugin as well as the nf-hello example plugin.

makeZip

Create the plugin archive and JSON meta file in the subproject build/libs directory.

$ ls -l1 plugins/nf-tower/build/libs/
nf-tower-0.1.0.jar
nf-tower-0.1.0.json
nf-tower-0.1.0.zip

copyPluginLibs

Copy plugin dependencies JAR files into the subproject build/target/libs directory. This is needed only when launching the plugin in development mode.

copyPluginZip

Copy the plugin archive to the root project build directory, i.e. build/plugins.

uploadPlugin

Upload the plugin archive and meta file to the corresponding GitHub repository. Options:

release

The plugin version, e.g. 1.0.1.

repo

The GitHub repository name, e.g. nf-amazon.

owner

The GitHub owning organization, e.g. nextflow-io.

skipExisting

Do not upload a file if it already exists, i.e. checksum is the same (default: true).

dryRun

Execute the tasks without uploading file (default: false).

overwrite

Prevent to overwrite a remote file already existing (default: false).

userName

The user name used to authenticate GitHub API requests.

authToken

The personal token used to authenticate GitHub API requests.

upload

Upload the plugin archive and meta file.

publishIndex

Upload the plugins index to the repository hosted at nextflow-io/plugins, which makes them publicly accessible at this URL.

Additional Resources

• PF4J

• Understanding Gradle: The Build Lifecycle