Labels & Outputs

Worker Labels

Workers declare capability labels when they register. Jobs declare required labels in the workflow YAML.

jobs:
  build:
    run: cargo build
    labels: [linux, docker]

Matching rule: A job’s required_labels must be a subset of the worker’s labels. A job with labels: [linux, docker] will only be claimed by workers whose labels include both linux AND docker.

Jobs with no labels (the default) can be claimed by any worker.

Label Strategy Patterns

Pattern	Labels	Use Case
Environment	staging, production	Route deploys to the right env
Capability	docker, gpu, arm64	Match hardware/software requirements
Team	team-infra, team-ml	Isolate workloads per team
Region	us-east, eu-west	Run jobs near data

Example: Multi-Environment Deploy

jobs:
  build:
    run: cargo build --release

  deploy-staging:
    needs: [build]
    labels: [staging]
    run: ./deploy.sh staging

  deploy-prod:
    needs: [build]
    labels: [production]
    run: ./deploy.sh production

# Staging worker
WORKER_LABELS=staging cargo run -p workflow-graph-worker-sdk

# Production worker
WORKER_LABELS=production cargo run -p workflow-graph-worker-sdk

Upstream Outputs

When a job completes with outputs, downstream jobs receive them automatically via upstream_outputs.

How It Works

1. A job completes and reports outputs via POST /api/jobs/{lease_id}/complete
2. The DAG scheduler stores outputs in the ArtifactStore
3. When a downstream job’s dependencies are all satisfied, the scheduler fetches all upstream outputs
4. The downstream job receives them in job.upstream_outputs when claimed

Example

jobs:
  build:
    run: |
      cargo build --release
      echo "::set-output artifact_url=s3://bucket/build.tar.gz"
      echo "::set-output version=1.2.3"

  deploy:
    needs: [build]
    run: deploy.sh

When deploy is claimed, it receives:

{
  "upstream_outputs": {
    "build": {
      "artifact_url": "s3://bucket/build.tar.gz",
      "version": "1.2.3"
    }
  }
}

Reading Outputs in Workers

Python:

upstream = job["upstream_outputs"]
artifact_url = upstream["build"]["artifact_url"]
subprocess.run(["deploy.sh", artifact_url])

Rust:

let artifact_url = job.upstream_outputs
    .get("build")
    .and_then(|o| o.get("artifact_url"))
    .expect("build should produce artifact_url");