miniwdl-omics-run

This command-line tool makes it easier to launch WDL runs on the AWS HealthOmics workflow service. It uses miniwdl locally to register WDL workflows, validate command-line inputs, and start a run.

pip3 install miniwdl-omics-run

miniwdl-omics-run \
    --role {SERVICE_ROLE_NAME} \
    --output-uri s3://{BUCKET_NAME}/{PREFIX} \
    {MAIN_WDL_FILE} input1=value1 input2=value2 ...

Quick start

Prerequisites: Unix command line with Python & pip; up-to-date AWS CLI installed locally, and configured with full access to your AWS account.

First a few one-time account setup steps (S3 bucket, IAM service role, ECR repo), then launching a test workflow.

S3 bucket

Create an S3 bucket with a test input file.

AWS_ACCOUNT_ID=$(aws sts get-caller-identity --query Account --output text)
AWS_DEFAULT_REGION=$(aws configure get region)

aws s3 mb --region "$AWS_DEFAULT_REGION" "s3://${AWS_ACCOUNT_ID}-${AWS_DEFAULT_REGION}-omics"
echo test | aws s3 cp - s3://${AWS_ACCOUNT_ID}-${AWS_DEFAULT_REGION}-omics/test/test.txt

Service role

Create an IAM service role for your Omics workflow runs to use (to access S3, ECR, etc.).

aws iam create-role --role-name poweromics --assume-role-policy-document '{
    "Version":"2012-10-17",
    "Statement":[{
        "Effect":"Allow",
        "Action":"sts:AssumeRole",
        "Principal":{"Service":"omics.amazonaws.com"}
    }]
}'

aws iam attach-role-policy --role-name poweromics \
    --policy-arn arn:aws:iam::aws:policy/PowerUserAccess

WARNING: PowerUserAccess, suggested here only for brevity, is far more powerful than needed. See Omics docs on service roles for the least privileges necessary, especially if you plan to use third-party WDL and/or Docker images.

ECR repository

Create an ECR repository suitable for Omics to pull Docker images from.

aws ecr create-repository --repository-name omics
aws ecr set-repository-policy --repository-name omics --policy-text '{
    "Version": "2012-10-17",
    "Statement": [{
        "Sid": "omics workflow",
        "Effect": "Allow",
        "Principal": {"Service": "omics.amazonaws.com"},
        "Action": [
            "ecr:GetDownloadUrlForLayer",
            "ecr:BatchGetImage",
            "ecr:BatchCheckLayerAvailability"
        ]
    }]
}'

Push a plain Ubuntu image to the repository.

ECR_ENDPT="${AWS_ACCOUNT_ID}.dkr.ecr.${AWS_DEFAULT_REGION}.amazonaws.com"
aws ecr get-login-password | docker login --username AWS --password-stdin "$ECR_ENDPT"

docker pull --platform linux/amd64 ubuntu:22.04
docker tag ubuntu:22.04 "${ECR_ENDPT}/omics:ubuntu-22.04"
docker push "${ECR_ENDPT}/omics:ubuntu-22.04"

Run test workflow

pip3 install miniwdl-omics-run
wget https://raw.githubusercontent.com/miniwdl-ext/miniwdl-omics-run/main/test/TestFlow.wdl

miniwdl-omics-run TestFlow.wdl \
    input_txt_file="s3://${AWS_ACCOUNT_ID}-${AWS_DEFAULT_REGION}-omics/test/test.txt" \
    docker="${AWS_ACCOUNT_ID}.dkr.ecr.${AWS_DEFAULT_REGION}.amazonaws.com/omics:ubuntu-22.04" \
    --role poweromics --output-uri "s3://${AWS_ACCOUNT_ID}-${AWS_DEFAULT_REGION}-omics/test/out"

This zips up the specified WDL, registers it as an Omics workflow, validates the given inputs, and starts the workflow run.

The WDL source code may be set to a local filename or a public HTTP(S) URL. The tool automatically bundles any WDL files imported by the main one. On subsequent invocations, it'll reuse the previously-registered workflow if the source code hasn't changed.

The command-line interface accepts WDL inputs using the input_key=value syntax exactly like miniwdl run, including the option of a JSON file with --input FILE.json. Each input File must be set to an existing S3 URI accessible by the service role.

Advice

Omics can use Docker images only from your ECR in the same account & region.
- This often means pulling, re-tagging, and pushing images as illustrated above with ubuntu:22.04.
- And editing any WDL tasks that hard-code docker image tags to take them as inputs instead.
- Each ECR repository must have the Omics-specific repository policy set as shown above.
- We therefore tend to use a single ECR repository for multiple Docker images, disambiguating them using lengthier tags.
- If you prefer to use per-image repositories, just remember to set the repository policy on each one.
To quickly list a workflow's inputs, try miniwdl run workflow.wdl ?