Skip to main content

Dependencies and DAGs

Two ways to declare edges​

Husky creates DAG edges from:

  • depends_on
  • after:<job> frequency rules

depends_on​

Use depends_on when a job has one or more upstream prerequisites.

jobs:
  transform:
    description: "Transform raw events"
    frequency: manual
    depends_on: [ingest]
    command: "./scripts/transform.sh"

after:<job>​

Use after:<job> when a job should fire immediately after a successful upstream run.

jobs:
  generate_report:
    description: "Generate summary"
    frequency: after:transform
    command: "./scripts/report.sh"

Example pipeline​

jobs:
  ingest_events:
    description: "Download raw events"
    frequency: manual
    command: "./scripts/ingest.sh"

  transform_events:
    description: "Transform events"
    frequency: after:ingest_events
    command: "./scripts/transform.sh"

  generate_report:
    description: "Build report"
    frequency: after:transform_events
    command: "./scripts/report.sh"

DAG validation​

Husky rejects cycles before the daemon starts or reloads.

Example invalid chain:

  • job_a -> job_b -> job_c -> job_a

Runtime rules​

A downstream job only runs when required upstream jobs have succeeded.

Important effects:

  • failed upstream jobs block downstream dispatch
  • on_failure: stop halts the affected pipeline branch
  • output templates resolve at dispatch time using the current cycle_id

Inspecting the graph​

husky dag
husky dag --json
curl http://127.0.0.1:8420/api/dag

Fan-out and fan-in​

Husky supports:

  • fan-out: one upstream triggers multiple downstream jobs
  • fan-in: one job waits for multiple upstream jobs

The graph must remain acyclic.