feature flags with instant rollback and percentage rollout

If you want feature flags with instant rollback and percentage rollout, the core idea is:

• Feature flag = turn a feature on/off without redeploying
• Instant rollback = disable the flag immediately if something goes wrong
• Percentage rollout = expose the feature to only a small % of users first, then gradually increase

Minimal architecture

1) Flag definition

Store flags in a central service or config store:

{
  "new_checkout": {
    "enabled": true,
    "rolloutPercent": 10,
    "targets": ["beta-users", "internal"]
  }
}

2) Evaluation logic

When a request comes in:

1. Check if flag is globally disabled
2. Check if user is in an allowlist/segment
3. Otherwise, hash a stable user key and map it to 0–99
4. Compare to rollout percentage

Example:

• user hash bucket = 23
• rollout = 10%
• result = off

3) Instant rollback

Keep flag evaluation remote or cache-based with short TTL:

• Toggle flag to false
• Invalidate cache or push update via pub/sub
• All app instances stop serving it quickly

Deterministic percentage rollout

Use a stable hash so a user consistently sees the same version.

Pseudo-example:

def is_enabled(user_id, flag_name, rollout_percent):
    bucket = hash(f"{flag_name}:{user_id}") % 100
    return bucket < rollout_percent

This ensures:

• same user always gets same treatment
• no flickering between requests

Best practices

• Use stable identifiers like user_id, not session ID
• Add kill switch flags for critical features
• Keep short cache TTLs or push updates instantly
• Log flag decisions for debugging
• Separate:
  • global on/off
  • user targeting
  • percentage rollout

Common rollout flow

1. Release code behind flag
2. Enable for internal users
3. Roll out to 1%
4. Increase to 5%, 10%, 25%, 50%, 100%
5. If issues occur, set flag off immediately

Tools you can use

Managed options:

• LaunchDarkly
• Unleash
• Flagsmith
• ConfigCat

Build-your-own:

• Redis/Postgres-backed flag store
• CDN/config service + pub/sub invalidation
• Deterministic hashing for rollout

If you want, I can also provide:

• a Node.js/Python implementation
• a database schema for flags
• a production-ready rollout algorithm