Skip to main content

​
The DAG

Jobs in a workflow form a directed acyclic graph (DAG) based on their needs: fields. Athanor uses Kahn’s algorithm to topologically sort jobs into levels. Jobs at the same level run concurrently.

​
Simple Chain

jobs:
  first:
    steps:
      - run: echo "first"
  second:
    needs: first
    steps:
      - run: echo "second"
  third:
    needs: second
    steps:
      - run: echo "third"
Execution: first → second → third (sequential).

​
Fan-Out / Fan-In

jobs:
  lint:
    steps:
      - run: echo "lint"
  test:
    needs: lint
    steps:
      - run: echo "test"
  build:
    needs: lint
    steps:
      - run: echo "build"
  deploy:
    needs: [test, build]
    steps:
      - run: echo "deploy"
Execution: 
lint ──┬── test  ──┬── deploy
       └── build ──┘
test and build run in parallel after lint completes. deploy waits for both.

​
Failure Propagation

If a job fails, all jobs that depend on it (directly or transitively) are skipped: 
lint (fails) ──┬── test  (skipped)
               └── build (skipped) ── deploy (skipped)
The workflow finishes with a failure status.

​
Cycle Detection

Circular dependencies are caught at parse time: 
jobs:
  a:
    needs: b
    steps:
      - run: echo "a"
  b:
    needs: a
    steps:
      - run: echo "b"
This produces an error: cycle detected in job dependencies.

​
No Dependencies

Jobs with no needs: field all run in the first level, concurrently: 
jobs:
  lint:
    steps:
      - run: echo "lint"
  test:
    steps:
      - run: echo "test"
  build:
    steps:
      - run: echo "build"
All three start at the same time.