Launchdarkly Guarded Rollout

LaunchDarkly Guarded Rollouts

You’re using a skill that will guide you through configuring guarded rollouts in LaunchDarkly. Your job is to design rollout stages, select monitoring metrics, configure regression thresholds, and start the rollout.

Prerequisites

This skill requires the remotely hosted LaunchDarkly MCP server to be configured in your environment.

Required MCP tools:

• start-guarded-rollout – start a progressive rollout with monitoring
• get-flag – inspect the flag and its variations
• list-metrics – find metrics to monitor during the rollout

Optional MCP tools:

• stop-guarded-rollout – halt an active rollout immediately
• toggle-flag – ensure the flag is turned on before starting
• create-metric – create metrics if they don’t exist

Core Concepts

What Are Guarded Rollouts?

A guarded rollout progressively increases traffic to a new feature flag variation through a series of stages. At each stage, LaunchDarkly monitors selected metrics for regressions. If a regression is detected, the rollout can automatically pause and notify the team — or even roll back.

Key Components

Rollout Weight Units

Rollout weights use thousandths (basis points):

• 1000 = 1%
• 10000 = 10%
• 50000 = 50%
• 100000 = 100%

Monitoring Window

The monitoring window is specified in milliseconds:

• 3600000 = 1 hour
• 86400000 = 24 hours
• 604800000 = 7 days

Core Principles

1. Start Small: Begin with a low percentage (1-5%) to catch issues early
2. Monitor What Matters: Choose metrics that reflect user experience
3. Set Realistic Thresholds: Too tight = false alarms; too loose = missed regressions
4. Allow Time: Each stage needs enough monitoring time for signal to emerge
5. Have a Rollback Plan: Always configure at least notification on regression

Workflow

Step 1: Prepare

Before starting a guarded rollout:

1. Use get-flag to inspect the flag — note the variation IDs for test and control
2. Use list-metrics to find metrics suitable for monitoring
3. Ensure the flag is on in the target environment (use toggle-flag if needed)
4. Confirm there’s no active guarded rollout on this flag already

Step 2: Design Stages

Plan the rollout progression. A typical pattern:

Step 3: Configure Metrics

Select metrics that indicate problems:

Step 4: Start the Rollout

Use start-guarded-rollout:

{
  "projectKey": "my-project",
  "flagKey": "new-checkout-flow",
  "environmentKey": "production",
  "testVariationId": "variation-id-for-new-flow",
  "controlVariationId": "variation-id-for-current-flow",
  "randomizationUnit": "user",

Step 5: Verify

1. Use get-flag to confirm the guarded rollout is active
2. Check that the flag shows the rollout configuration in the environment
3. Monitor for any immediate regression notifications

Report results:

• Guarded rollout started with N stages
• M metrics being monitored
• First stage at X% traffic for Y hours

Stopping a Rollout

If issues arise or you need to halt the rollout:

{
  "projectKey": "my-project",
  "flagKey": "new-checkout-flow",
  "environmentKey": "production"
}

This immediately stops the progressive rollout and locks the flag at its current state.

Edge Cases

What NOT to Do

• Don’t start a guarded rollout on a flag that’s turned off
• Don’t skip the monitoring window design — rushing through stages defeats the purpose
• Don’t set regression thresholds to 0 — small fluctuations are normal
• Don’t forget to configure at least one metric — a rollout without monitoring is just a regular rollout