Implement Progressive Delivery Steps

Roll out a change in controlled increments with automated analysis between steps so a bad release is caught early.

~40 mindeploymentopsreliability

Task

Define progressive delivery steps with automated metric analysis gates between increments.

When to use this

A release is rolled out fully at once, so any regression affects all users immediately.

Prerequisites

A progressive delivery controller such as Argo Rollouts or Flagger
Metric queries for success rate and latency
A deployment pipeline

Steps

1. Define incremental steps with pauses.

steps:
  - setWeight: 10
  - pause: { duration: 5m }
  - setWeight: 50
  - pause: { duration: 5m }

Expected output: Traffic increases in steps, with time to analyse each increment.

2. Attach automated analysis that aborts on regression.

analysis:
  templates:
    - templateName: success-rate
  args:
    - name: service
      value: app

Expected output: The rollout pauses or aborts automatically when success rate drops below threshold.

Troubleshooting

Problem

Metrics too sparse at 10 percent

Solution

Use longer pauses or compare against a canary baseline rather than absolute thresholds.

Problem

Rollback leaves partial traffic

Solution

Confirm the controller resets weight to zero and drains canary pods on abort.

Limitations

Requires reliable metrics.
Canary analysis adds release time.
Not a substitute for pre-release testing.

What this recipe does

Define progressive delivery steps with automated metric analysis gates between increments.

A release is rolled out fully at once, so any regression affects all users immediately.

Prerequisites

A progressive delivery controller such as Argo Rollouts or Flagger
Metric queries for success rate and latency
A deployment pipeline

Procedure

Define incremental steps with pauses.

steps:
  - setWeight: 10
  - pause: { duration: 5m }
  - setWeight: 50
  - pause: { duration: 5m }

Expected result: Traffic increases in steps, with time to analyse each increment.

Attach automated analysis that aborts on regression.

analysis:
  templates:
    - templateName: success-rate
  args:
    - name: service
      value: app

Expected result: The rollout pauses or aborts automatically when success rate drops below threshold.

How to confirm it worked

Run the commands above and confirm every expected result in order. If a step does not produce the stated output, use the troubleshooting notes before continuing. A passing run shows a simulated regression triggers an automatic abort during a pause instead of reaching full traffic.

Troubleshooting

Metrics too sparse at 10 percent — Use longer pauses or compare against a canary baseline rather than absolute thresholds.

Rollback leaves partial traffic — Confirm the controller resets weight to zero and drains canary pods on abort.

Limitations

Requires reliable metrics.
Canary analysis adds release time.
Not a substitute for pre-release testing.