Pattern: Feature flags for safe deployments

Deploying new features is all-or-nothing. Rolling back requires a new deployment. Can't do gradual rollouts or A/B tests.

Use feature flags to decouple deployment from release:

# Simple implementation
class FeatureFlags:
def __init__(self, config_source):
self._flags = config_source.load()

def is_enabled(self, flag_name, user=None):
flag = self._flags.get(flag_name)
if not flag:
return False

# Global kill switch
if not flag.get('enabled', False):
return False

# Percentage rollout
if 'percentage' in flag and user:
return hash(f"{flag_name}:{user.id}") % 100 < flag['percentage']

# User allowlist
if 'allowed_users' in flag and user:
return user.id in flag['allowed_users']

return flag['enabled']

# Usage in code:
if features.is_enabled('new_checkout', current_user):
return new_checkout_flow()
else:
return legacy_checkout_flow()

Best practices:

• Clean up old flags regularly (tech debt)
• Log flag evaluations for debugging
• Have a kill switch for every flag
• Test both paths in CI

Feature flags let you deploy code without releasing features, enabling gradual rollouts, instant rollbacks, and experimentation.