Feature Flags & Progressive Delivery

Feature flags decouple deployment from release. You deploy code to production that’s hidden behind a flag, then turn it on for 1% of users, then 10%, then 50%, watching metrics at each step. If something breaks, flip the flag off — instant rollback without a deployment. This is how Netflix, Google, and every high-performing engineering team ships safely.

---

Flag Types

Type	Purpose	Lifetime	Example
Release	Hide incomplete features	Days to weeks	New checkout flow
Experiment	A/B test variations	Weeks to months	Button color test
Ops	Circuit breaker, kill switch	Permanent	Disable recommendations under load
Permission	Entitlement-based access	Permanent	Premium-only features

---

Implementation

from feature_flags import FeatureFlags

flags = FeatureFlags(provider="launchdarkly")  # or Unleash, Flagsmith, etc.

# Simple boolean flag
if flags.is_enabled("new-checkout", user=current_user):
    return render_new_checkout()
else:
    return render_current_checkout()

# Percentage rollout
flags.create_flag("new-checkout", {
    "targeting": {
        "rules": [
            {
                "name": "internal-users",
                "match": {"attribute": "email", "endsWith": "@company.com"},
                "percentage": 100,  # All internal users
            },
            {
                "name": "beta-users",
                "match": {"attribute": "beta_opt_in", "equals": True},
                "percentage": 100,  # All beta users
            },
            {
                "name": "gradual-rollout",
                "match": "everyone",
                "percentage": 10,  # 10% of all other users
            },
        ],
        "default": False,
    }
})

---

Progressive Delivery

Step 1: Deploy to production (flag OFF for everyone)
        └── Verify: no errors, service healthy

Step 2: Enable for internal team (flag ON for @company.com)
        └── Verify: team tests new feature in production

Step 3: Enable for 5% of users
        └── Verify: error rate, latency, conversion rate

Step 4: Enable for 25% of users
        └── Verify: metrics stable, no performance regression

Step 5: Enable for 100% of users
        └── Monitor for 1 week

Step 6: Remove flag from code (clean up)

---

Flag Lifecycle

Phase	Duration	Action
Create	Day 0	Define flag, add to code behind flag
Test	Days 1-3	Internal testing, QA validation
Rollout	Days 3-14	Progressive rollout: 5% → 25% → 100%
Stable	Day 14+	Flag at 100%, monitoring confirms success
Cleanup	Day 30	Remove flag from code, delete from flag service

---

Anti-Patterns

Anti-Pattern	Problem	Fix
Flags never cleaned up	Thousands of stale flags, unmaintainable code	Expiration date on every flag, monthly cleanup
Nested flags	Complex interactions, untestable combinations	Minimize flag interactions, test combinations
Flags in hot loops	Performance overhead per-request	Cache flag values, evaluate once per request
No default value	Flag service down = feature broken	Always define sensible default (usually OFF)
Testing only with flag ON	Flag OFF path untested, breaks on rollback	Test both flag states in CI

---

Checklist

• Flag management platform selected (LaunchDarkly, Unleash, Flagsmith)
• Flag naming convention established
• Progressive delivery: rollout stages defined
• Kill switch flags for critical dependencies
• Cleanup policy: flags removed within 30 days of 100% rollout
• Testing: both flag states tested in CI
• Monitoring: flag evaluation metrics, rollout dashboards
• Default values: every flag has a safe default

---

:::note[Source] This guide is derived from operational intelligence at Garnet Grid Consulting. For feature flag consulting, visit garnetgrid.com. :::