Skip to main content

Guarded rollouts, workflows, and pipelines

Release controls turn a flag change into a measurable, reviewable process. They remain separate resources so the same flag can have different release policy in Test and Production.

Guarded rollouts​

Open Project → Guarded rollouts → Create guarded rollout. Select a flag and environment, then configure:

  • the context kind and bucket attribute used for stable allocation;
  • stages such as 10% for 60 minutes, 50% for four hours, then 100%;
  • one or more guardrail metrics and their acceptable thresholds;
  • observation windows and minimum sample requirements;
  • whether regression pauses the rollout or automatically rolls back;
  • the fallback variation and notification destinations.

Statuses are draft, scheduled, running, paused, completed, rolled_back, and failed. A stage advances only after its duration and health criteria are satisfied. Automatic rollback restores the configured safe allocation and records the reason in the resource payload and audit log.

LaunchDarkly guarded-release and release-policy payloads are retained as lossless JSON during migration. This includes provider fields not yet shown as dedicated form controls.

Metrics​

Open Project → Metrics. Metrics can count custom events, measure numeric values, calculate conversions, or act as guardrails. Configure the event key, randomization unit, aggregation, direction (higher or lower is better), inclusion of units without events, measurement window, and any filters. Experiments and guarded rollouts reference metric keys; changing a key after data collection breaks continuity.

Workflows​

Open Project → Workflows → Create workflow. A workflow is an ordered release schedule. Stages may request approval, wait until a date, change targeting, increase a percentage, send a notification, or wait for a manual continuation. Each stage stores its status and schedule so a paused or failed workflow can be inspected without losing intent.

Typical progressive workflow:

  1. approval gate;
  2. enable targeting at 10%;
  3. wait one hour and check health;
  4. increase to 50%;
  5. wait four hours and check health;
  6. release to 100% or roll back.

Release pipelines and policies​

Open Project → Release policies to recommend progressive or guarded delivery for matching environments, tags, views, context kinds, or stages. A policy stores traffic progression, metrics, regression thresholds, auto-rollback, and minimum sample size. Policies are ordered by priority; a fallback policy applies when no scoped policy matches.

Open Project → Release pipelines to define reusable lifecycle stages such as Development, Test, Limited availability, and Production. A pipeline can store required approvals, environment mappings, stage order, allowed transitions, and automation. Use pipelines for organizational consistency; use a workflow for the concrete execution attached to a flag.

Keys are unique within the same project, environment, and resource kind. The Advanced configuration JSON editor is intentionally lossless and is the canonical place for fields not represented by a first-class editor.