Retry policy

Automax supports inherited retry policies on jobs, tasks, steps and substeps. The closest scope wins, so a substep can override a job-level default.

Retry is intended for transient operational failures: temporary package-manager locks, services that need another restart attempt, short-lived network glitches, or remote commands that can safely be retried.

Basic example

retry:
  attempts: 3
  delay: 5
  backoff: fixed

apiVersion: automax.io/v1
kind: Job
metadata:
  name: retry-demo
retry:
  attempts: 2
  delay: 1
tasks:
  - id: deploy
    targets: group:web
    steps:
      - id: restart
        substeps:
          - id: restart_service
            use: system.service.restart
            retry:
              attempts: 3
              delay: 2
              backoff: exponential
              max_delay: 10
            with:
              service: myapp
              sudo: true

Fields

Field	Default	Description
attempts	1	Total attempts, including the first execution.
delay	0	Delay in seconds before the next attempt.
backoff	fixed	fixed or exponential.
max_delay	unset	Maximum delay when exponential backoff is used.
retry_on_rc	unset	Optional list of return codes eligible for retry. Without it, every failed result can retry.

Camel-case aliases maxDelay and retryOnRc are accepted for operators who prefer that style. Public examples use the canonical snake_case fields.

Visual retry output

Text runs print retry attempts before the final result:

[RETRY] web01 task.deploy:step.restart:substep.restart_service attempt=1/3 rc=1 next=2 delay=2s systemctl failed
[RETRY] web01 task.deploy:step.restart:substep.restart_service attempt=2/3 rc=1 next=3 delay=4s systemctl failed
[OK] web01 task.deploy:step.restart:substep.restart_service rc=0

The final node output also records attempt metadata in the run state:

{
  "data": {
    "attempt": 3,
    "attempts": [
      {"attempt": 1, "ok": false, "rc": 1, "message": "systemctl failed"},
      {"attempt": 2, "ok": false, "rc": 1, "message": "systemctl failed"},
      {"attempt": 3, "ok": true, "rc": 0, "message": ""}
    ]
  }
}

Retry and failure policy

errorPolicy is evaluated before retry. This means an accepted non-zero return code that becomes a warning is not retried. A substep only reaches failurePolicy after all retry attempts are exhausted or after a return code that is not listed in retry_on_rc.