Build, version, and publish workflows

In this tutorial we look at the flow of creating and managing workflows: from a local, project-based workflow to a versioned, global workflow that can be published to authorized users and listed in the DNAnexus tools library.

Motivation

Workflows created in the DNAnexus platform with an ID workflow-xxxx are data objects stored in a project. Since they are stored in a project it is easy to share the workflow with other users by adding them to the project. Such "local" workflows are also great for fast iterations for development and testing. It is convenient that we can always delete the workflow and create a new one in its place.

However, for users and organizations collaborating on multiple private or public projects, these local workflows may be less suitable for long-term maintenance and collaboration. To address use cases in this context, DNAnexus provides an additional suite of release management features that can be used. For example, an organization administrator or workflow developer may want to restrict execution of a particular workflow to specific users and organizations, or they may want to allow execution of this workflow across cloud regions and providers. These features are applied to objects we call "global workflows".

A global workflow is an executable that can be versioned and published to other users. It is implemented as a wrapper around an existing, project-based workflow.

Examples of when it is useful to make a local workflow global

  • version management: it is easy to maintain the source code of the workflow across versions
  • provenance: global workflows maintain an explicit history of changes for the workflow name and its associated ID. The user can always revert to a previous version and versions are immutable
  • sharing: a global workflow can be maintained by a set of developers across projects and organizations and this workflow can be shared to a separate set of users who cannot modify the workflow but can run it
  • multi-region support: it is convenient to maintain one executable that can be run across multiple regions and cloud providers
  • immutability guarantees: it is not possible to change/override an existing version of a given global worflow
  • ability to list the workflow in the DNAnexus tools library: the authorized users and developers can then discover global workflows shared with them

Step 1: Build a local workflow

The easiest way to build a workflow is by using the web interface, as detailed in the Building Workflows in the UI guide. We can also use the DNAnexus SDK, dx-toolkit. For details on how to create a workflow using dx-toolkit, please check the Intro to Building Workflows tutorial.

In both tutorials we built a similar workflow named "BWA MEM + GATK Exome Workflow" in a project. The creator and all the users that have at least VIEW permissions to this project, can run it in that project.

We will now create a global workflow based on the "BWA MEM + GATK Exome Workflow". Please, make sure your installed dx-toolkit version is at least 0.253 (you can upgrade it using dx upgrade).

Step 2: Make the workflow global

We can use the existing, project-based workflow to create a new object on the DNAnexus platform - a global workflow. Functionally, the global workflow will be identical to the original workflow.

The basic procedure is the following:

  1. use the dx get command to download a JSON representation (dxworkflow.json) of a chosen workflow on the platform
  2. set a unique name (in the global namespace) and version in the downloaded JSON file
  3. use the dx build --globalworkflow command to build the global workflow on the platform

We'll go over each step in more detail below.

Download the original workflow

First, let's download the JSON source code of the workflow, as follows:

dx get "BWA MEM + GATK Exome Workflow"

This will create a directory "BWA MEM + GATK Exome Workflow" locally and a dxworkflow.json file in it.

Alternatively, you can skip this step since in the next section we provide the example dxworkflow.json that you can use.

Build a global workflow on the platform

We will use the downloaded dxworkflow.json to create a new global workflow. The only additional fields that are required for the global workflow are: name and version. The name has to be unique in the global namespace shared by apps and global workflows, that is no other app or global workflow (whether it is available to us or not) should have the same name. We can leave the "stages" and any other fields unchanged.

TIP: We recommend following the Semantic Versioning (external link) conventions for numbering the versions of your global workflow.
{
  "name": "bwa_mem_gatk_exome_demo",
  "version": "0.0.1",
  "stages": [
    ...
  ]
}

Our final dxworkflow.json can be found here.

We can add fields such as title and summary to make our workflow more user-friendly. It is also a good practice to include documentation which can be placed in the Readme.md file in the same directory.

We can now generate our first global workflow as follows:

dx build --globalworkflow "BWA MEM + GATK Exome Workflow"

BWA MEM + GATK Exome Workflow is the name of the directory storing the dxworkflow.json, it can be changed freely.

Describe and run the workflow

The dx build command should return us the unique ID ('globalworkflow-xxxx') of the created version. We can use it to refer to the global workflow from now on, for example:

dx describe globalworkflow-xxxx
TIP: All the valid identifiers of the global workflow are listed here.

Running a global workflow is the same as running any other workflow. We can run our workflow for example:

dx run -ialign_reads.reads_fastqgz=myreads.fastq.gz \
       -ialign_reads.genomeindex_targz=file-xxxx \
       bwa_mem_gatk_exome_demo

Step 3: Add authorized users

We can now specify a list of users with whom we want to share our workflow by using dx add users. We will prepare the list now but the users will be able to find and access the global workflow only after the workflow is published. The list of users can be updated by developers before or after it is published, and it will apply to all to the versions of the workflows, past and future.

For example, to share our workflow with a user and an organization we'll run:

dx add users bwa_mem_gatk_exome_demo user-bob org-partnerorg

Then, we can view who is on the access list for the workflow:

dx list users bwa_mem_gatk_exome_demo

and remove a user by running:

dx remove users bwa_mem_gatk_exome_demo org-partnerorg

The authorized user permissions do not propagate to apps automatically, so if the workflow contain any apps the users need access to them to in order to run the workflow. To add users to an app, the same dx add users command can be used.

Step 4: Release (publish) a version to users

Once we have tested our workflow version we can release it to the authorized users by executing dx publish:

dx publish bwa_mem_gatk_exome_demo/0.0.1

Publishing a workflow version has the following effects:

  • the authorized users can discover the version via dx find globalworkflow
  • the authorized users can describe and run the version
  • the authorized users can download the workflow's dxworkflow.json source code with dx get (as well as the dependencies, e.g. applets), though they cannot build new versions with the same global workflow name

Developers can add and remove users and other developers any time. Adding users to the workflow will give the users access to all the published versions but it will have no effect on the unpublished versions. Unpublished versions are only accessible to the people listed as developers of the workflow.

The "default" alias

Executing dx publish command will make the published version a default one by adding a "default" alias to this version (and removing it from whichever workflow version was previously marked as "default"). The alias indicates that this version will be invoked whenever the workflow name is used without the version. For example, users can run the workflow:

dx run bwa_mem_gatk_exome_demo   # equivalent to "dx run bwa_mem_gatk_exome_demo/0.0.1"

The first created version of a global workflow gets this flag automatically. Please, check dx publish --help for more options for this command.

Next steps

Add developers

We can enable multiple users to update current or build new versions of our global workflow, for example:

dx add developers bwa_mem_gatk_exome_demo user-eve

Either specific users or whole orgs can be developers of a global workflow. We can list who is on the developer list as follows:

dx list developers bwa_mem_gatk_exome_demo

To remove developers:

dx remove developers bwa_mem_gatk_exome_demo user-bob

Create a new version

We can create a new version of the workflow by updating the dxworkflow.json and using dx build --globalworkflow. In order to create the new version, we need to explicitly update the version string, e.g. to "0.0.2".

If we loose the source dxworkflow.json we can download it any time with dx get, for example to get version 0.0.2:

dx get globalworkflow-bwa_mem_gatk_exome_demo/0.0.2

Any developer of the workflow can download its source code and build new versions. Authorized users that have access to published versions can only download or run it.

Delete

We can mark a global workflow version as deleted, which will make the version unrunnable. It will still be possible to describe the workflow for provenance purposes. For example:

dx api globalworkflow-xxxx delete

Please, use this route with caution as deleting a global executable can break users' reproducibility requirements.

Deleting all workflow versions will not release the workflow name. That is, it will not be possible to reuse the name for a different global workflow by another user.

Search and discover global workflows

Search all available versions

The command dx find globalworkflows is useful for browsing global workflows that are available to us.

dx find globalworkflows

The above command will print a list of published global workflows. To list the workflows that are not published:

dx find globalworkflows --unpublished
TIP: By default, "dx find globalworkflows" lists one version per each available workflow - the version marked as "default". To print the whole version history for a workflow, add the "--all" flag, for example:
dx find globalworkflows --unpublished --all

Search by category

The --category parameter can be used to restrict the search to workflows from a specific category. Common categories are available as tab completions. For example:

dx find globalworkflows --category Variation\ Calling

To view all available categories that we can search by:

dx find globalworkflows --category-help

Summary

In the table we summarize the main steps or stages we went through to create and publish our workflow.

Object name ID prefix Namespace Access Definition and purpose
Workflow workflow Project based on project permissions A file-like workflow object stored in a project; used for private, light-weight development
Global workflow unpublished globalworkflow Global developers A development version of a workflow that is not visible to users but is to developers; used for development and testing in multiple regions, release management, preparation for publishing to users
Global workflow published globalworkflow Global users, developers A version of a workflow that is available to authorized users
Global workflow deleted globalworkflow Global users, developers A version (previously published or unpublished) that is not runnable but can be described

Next steps

  • A detailed description of the global workflow object is included in the API documentation.

Last edited by Aleksandra Zalcman, 2018-05-23 00:40:55

 Feedback