ark retry

Retry failed pipeline steps in a live cluster, preserving steps that already succeeded - full flags reference for ark retry.

Reset failed steps back to Pending so the operator re-submits them. Steps that already succeeded are left untouched.

ark retry <team> [flags]

Flags

FlagDefaultDescription
-n, --namespace-Namespace of the ArkTeam (required)
--step-Retry only this specific step. Defaults to all failed steps.
--contextcurrent contextKubernetes context to use

Examples

# Retry all failed steps
ark retry research-pipeline -n my-org

# Retry a single step
ark retry research-pipeline -n my-org --step summarize

# Retry in a specific cluster
ark retry research-pipeline -n my-org --context prod-cluster

What retry does

ark retry is a surgical operation - it only touches failed steps:

  1. Resets phase of failed steps to Pending
  2. Clears message, output, and completionTime for those steps
  3. Sets status.phase to Running

Steps with phase: Succeeded are not modified. The operator picks up the change and re-submits only the reset steps on the next reconcile.

This is the right tool when a step failed due to a transient error (network timeout, rate limit, bad LLM response) and you want to resume without re-running the work that already completed.

Retry vs trigger

ark retryark trigger
Resets failed steps✅✅
Preserves succeeded steps✅✗
Accepts new --input✗✅
Use whenResuming after failureStarting fresh

See also