Feature Flags: Stale Flag Causes 15-Minute Outage

# Package: io.thecodeforge.python.devops

# Simplest possible feature flag — environment variable
import os

def get_recommendations(user_id: int):
    if os.getenv('ENABLE_ML_RECOMMENDATIONS', 'false') == 'true':
        return ml_recommendations(user_id)   # new ML-based system
    else:
        return rule_based_recommendations(user_id)  # old system

# Better: percentage rollout — test on a fraction of users
import hashlib

def is_flag_enabled(flag_name: str, user_id: int, percentage: float) -> bool:
    """Consistently assign users to buckets using hash — same user always gets same result."""
    hash_input = f'{flag_name}:{user_id}'.encode()
    hash_val   = int(hashlib.md5(hash_input).hexdigest(), 16)
    bucket     = (hash_val % 100) + 1  # 1-100
    return bucket <= percentage

# Roll out to 5% of users
def get_checkout_flow(user_id: int):
    if is_flag_enabled('new_checkout', user_id, 5.0):
        return new_checkout_flow(user_id)
    return old_checkout_flow(user_id)

# Package: io.thecodeforge.python.devops

# Using a feature flag service (LaunchDarkly, Unleash, Flagsmith)
import ldclient
from ldclient.config import Config

ldclient.set_config(Config(sdk_key='your-sdk-key'))
client = ldclient.get()

# Evaluate a flag for a specific user
def get_dashboard(user):
    context = {
        'key': str(user.id),
        'name': user.name,
        'email': user.email,
        'plan': user.subscription_plan,   # target premium users
        'country': user.country           # GDPR rollout by country
    }

    # Flag evaluated with user context — targeting rules in dashboard
    if client.variation('new-dashboard-v2', context, default=False):
        return render_new_dashboard(user)
    return render_old_dashboard(user)

# Package: io.thecodeforge.python.devops

# Naming convention for flag types
release_flag_variation = client.variation('release_new_checkout_v3', context, default=False)
experiment_flag_variation = client.variation('exp_checkout_button_color', context, default='blue')
ops_flag_variation = client.variation('ops_disable_payment_gateway', context, default=False)
perm_flag_variation = client.variation('perm_premium_dashboard', context, default=False)

# Package: io.thecodeforge.python.devops

# Canary with feature flag: even if the pod receives traffic, only a fraction of users get the new feature
import hashlib

def compute_bucket(user_id, flag_name, total_percent):
    hash_val = int(hashlib.md5(f'{flag_name}:{user_id}'.encode()).hexdigest(), 16)
    return (hash_val % 100) + 1 <= total_percent

# Canary: 5% of pods run new code, but only 20% of users on those pods get the feature
# That's effectively 1% of total users
if compute_bucket(user_id, 'new_recommendation_v2', 20):
    # This code only runs in the canary pods
    return new_recommendation_system(user_id)
else:
    return old_system(user_id)

# Package: io.thecodeforge.python.devops

# Example: automate flag cleanup detection in CI
# This would be a script that checks git for old flag references

import subprocess
import re

FLAG_PATTERN = r'client\.variation\([\'"]([\w-]+)[\'"]'

def find_old_flags(months: int = 3):
    # Get all flags used in codebase
    result = subprocess.run(['grep', '-roPh', FLAG_PATTERN, 'src/'], capture_output=True, text=True)
    flags = set(re.findall(FLAG_PATTERN, result.stdout))
    # Check each flag's metadata (would use API in real life)
    # For now, just list them
    return flags

# In CI, warn if a release flag is older than 2 weeks
# This helps reduce flag debt