// feature-flag.service.ts
import { Redis } from 'ioredis';

export class FeatureFlagService {
  private redis: Redis;

  constructor(redisUrl: string) {
    this.redis = new Redis(redisUrl);
  }

  async isEnabled(flagKey: string, context: Record<string, any> = {}): Promise<boolean> {
    // Production implementation should include caching and consistent hashing
    const value = await this.redis.get(`ff:${flagKey}`);
    if (value === 'true') return true;
    if (value === 'false') return false;
    
    // Fallback to default or percentage rollout
    return this.evaluateRollout(flagKey, context);
  }

  private async evaluateRollout(key: string, context: Record<string, any>): Promise<boolean> {
    // Implement percentage-based rollout logic here
    return false;
  }
}

// db-migration.helper.ts
import { Pool, QueryResult } from 'pg';

export class MigrationHelper {
  private pool: Pool;

  constructor(pool: Pool) {
    this.pool = pool;
  }

  // Phase 1: Expand - Add column safely
  async expandAddColumn(tableName: string, columnName: string, type: string): Promise<void> {
    const sql = `
      ALTER TABLE ${tableName} 
      ADD COLUMN IF NOT EXISTS ${columnName} ${type};
    `;
    await this.pool.query(sql);
  }

  // Phase 3: Backfill - Migrate data in batches to avoid locks
  async backfillData(
    tableName: string, 
    oldCol: string, 
    newCol: string, 
    batchSize: number = 1000
  ): Promise<void> {
    let affected = 0;
    do {
      const sql = `
        UPDATE ${tableName} 
        SET ${newCol} = ${oldCol} 
        W

HERE ${newCol} IS NULL AND ${oldCol} IS NOT NULL LIMIT ${batchSize} RETURNING id; `; const result: QueryResult = await this.pool.query(sql); affected = result.rowCount || 0;

  // Yield to event loop to prevent blocking
  await new Promise(resolve => setImmediate(resolve));
} while (affected > 0);

}

// Phase 5: Contract - Drop column only when safe async contractDropColumn(tableName: string, columnName: string): Promise<void> { // Verify no application instances reference this column // This check should be enforced by CI/CD or deployment gate const sql = ALTER TABLE ${tableName} DROP COLUMN IF EXISTS ${columnName};; await this.pool.query(sql); } }

#### 3. Graceful Shutdown and Health Checks

Zero-downtime requires the application to handle SIGTERM signals and deregister from load balancers before terminating. This prevents in-flight requests from being dropped.

```typescript
// server.ts
import express from 'express';
import { FeatureFlagService } from './feature-flag.service';
import { MigrationHelper } from './db-migration.helper';

const app = express();
let isShuttingDown = false;

// Health check endpoint
app.get('/health', (req, res) => {
  if (isShuttingDown) {
    return res.status(503).json({ status: 'draining' });
  }
  res.status(200).json({ status: 'healthy' });
});

// Readiness check for load balancer
app.get('/ready', async (req, res) => {
  // Check dependencies
  const dbOk = await checkDatabase();
  const cacheOk = await checkCache();
  
  if (dbOk && cacheOk) {
    res.status(200).json({ status: 'ready' });
  } else {
    res.status(503).json({ status: 'not_ready' });
  }
});

// Graceful shutdown handler
const shutdown = async () => {
  if (isShuttingDown) return;
  isShuttingDown = true;

  console.log('Shutting down gracefully...');
  
  // Stop accepting new requests
  // In K8s, this allows Service to remove endpoint from rotation
  
  // Wait for in-flight requests (timeout protection)
  await new Promise(resolve => setTimeout(resolve, 5000));
  
  // Close connections
  await db.pool.end();
  process.exit(0);
};

process.on('SIGTERM', shutdown);
process.on('SIGINT', shutdown);

app.listen(3000, () => {
  console.log('Server running on port 3000');
});

4. Dual-Write Pattern

During the cutover phase, V2 must write to both schemas to ensure data consistency.

// user.repository.ts
import { FeatureFlagService } from './feature-flag.service';

export class UserRepository {
  constructor(
    private db: any, 
    private flags: FeatureFlagService
  ) {}

  async saveUser(user: any) {
    const useNewSchema = await this.flags.isEnabled('user_new_schema');

    // Always write to old schema for backward compatibility
    await this.db.query(
      'INSERT INTO users (id, name, email) VALUES ($1, $2, $3)',
      [user.id, user.name, user.email]
    );

    // Dual-write if flag is enabled
    if (useNewSchema) {
      await this.db.query(
        'INSERT INTO users_v2 (id, name, email, metadata) VALUES ($1, $2, $3, $4)',
        [user.id, user.name, user.email, JSON.stringify(user.metadata)]
      );
    }
  }

  async getUser(id: string) {
    const useNewSchema = await this.flags.isEnabled('user_new_schema_read');
    
    if (useNewSchema) {
      return this.db.query('SELECT * FROM users_v2 WHERE id = $1', [id]);
    }
    return this.db.query('SELECT * FROM users WHERE id = $1', [id]);
  }
}

Architecture Rationale

• Backward Compatibility: Every deployment must be backward compatible with the previous version. V2 must work with V1 data; V1 must not break if V2 writes new data.
• Atomic Cutover: Feature flags provide an atomic switch for the read path. Once the backfill is complete, flipping the flag transitions traffic to the new schema instantly without deployment.
• Stateless Design: Applications must not store session state in memory. Sessions must be externalized to Redis or cookies to support instance rotation.
• Health Check Granularity: Separate /health (liveness) and /ready (readiness) endpoints. The load balancer should only route traffic to pods passing /ready. During shutdown, /ready fails immediately, stopping traffic routing while /health remains true until termination.

Pitfall Guide

1. Breaking Database Migrations

Mistake: Running ALTER TABLE DROP COLUMN or renaming columns in a single deployment. Impact: V1 instances crash when querying the missing column. Fix: Always use Expand/Contract. Add new columns, dual-write, backfill, cutover reads, then drop old columns in a subsequent deployment.

2. Sticky Sessions and In-Memory State

Mistake: Assuming stateless deployment works with in-memory sessions or caches. Impact: Users lose sessions or see stale data during rolling updates. Fix: Externalize state to Redis, DynamoDB, or distributed caches. Implement session affinity only if unavoidable, and ensure the load balancer handles session transfer.

3. Health Check Misconfiguration

Mistake: Using a single health endpoint or checking only process uptime. Impact: Load balancer routes traffic to pods that are starting up or failing dependencies. Fix: Implement startup probes, liveness probes, and readiness probes. Readiness should check database connectivity and cache status. Configure drain timeout to allow in-flight requests to complete.

4. Feature Flag Leakage

Mistake: Leaving feature flags enabled indefinitely without cleanup. Impact: Code complexity increases, testing matrix explodes, and dead code causes performance degradation. Fix: Treat feature flags as technical debt. Set expiration dates. Automate flag cleanup in CI/CD pipelines. Monitor flag usage and remove unused flags.

5. Rollback Strategy Absence

Mistake: Deploying V2 without a tested rollback path. Impact: If V2 fails, rolling back to V1 causes downtime or data loss due to schema incompatibility. Fix: Ensure V1 can run alongside V2 data. Test rollback procedures in staging. Automate rollback triggers based on error rate thresholds.

6. DNS Propagation Delays

Mistake: Using DNS-based routing with high TTL for Blue/Green switches. Impact: Users experience downtime during DNS propagation. Fix: Use low TTL values (e.g., 60 seconds) for deployment switches. Prefer load balancer-level routing over DNS for critical switches.

7. CI/CD Pipeline Fragility

Mistake: Manual gates or untested deployment scripts. Impact: Human error introduces downtime; deployments are too slow to support frequent releases. Fix: Automate all deployment steps. Use infrastructure as code. Implement automated rollback on failure. Test deployment strategies in staging with production-like data.

Production Bundle

Action Checklist

• Verify database backward compatibility for all schema changes using Expand/Contract pattern.
• Implement separate /health and /ready endpoints with dependency checks.
• Configure load balancer or orchestrator with drain timeout and readiness checks.
• Use feature flags to decouple deployment from release for all logic changes.
• Implement graceful shutdown handling (SIGTERM) to complete in-flight requests.
• Externalize session state and caches to distributed storage.
• Automate rollback triggers based on error rate and latency thresholds.
• Test deployment strategy in staging with production data volume.

Decision Matrix

Scenario	Recommended Approach	Why	Cost Impact
Stateless microservice, config change	Rolling Update	Low complexity, safe for stateless apps.	Low
Database schema migration	Expand/Contract	Only pattern safe for breaking schema changes.	Medium
High-risk feature, A/B testing	Canary + Feature Flags	Granular control, instant kill switch, traffic splitting.	High
Multi-region deployment	Blue/Green per region	Isolation, fast rollback, region-level safety.	High
Legacy monolith, infrequent deploys	Blue/Green	Simplifies rollback, reduces risk for large changes.	Medium

Configuration Template

Kubernetes Deployment with Rolling Update strategy, readiness probes, and graceful shutdown configuration.

apiVersion: apps/v1
kind: Deployment
metadata:
  name: zero-downtime-app
spec:
  replicas: 4
  strategy:
    type: RollingUpdate
    rollingUpdate:
      maxSurge: 1      # Allow 1 extra pod during update
      maxUnavailable: 0 # Never reduce available pods
  template:
    spec:
      terminationGracePeriodSeconds: 30 # Time for graceful shutdown
      containers:
      - name: app
        image: myapp:latest
        ports:
        - containerPort: 3000
        readinessProbe:
          httpGet:
            path: /ready
            port: 3000
          initialDelaySeconds: 5
          periodSeconds: 5
          failureThreshold: 3
        livenessProbe:
          httpGet:
            path: /health
            port: 3000
          initialDelaySeconds: 15
          periodSeconds: 20
        lifecycle:
          preStop:
            exec:
              command: ["/bin/sh", "-c", "sleep 10"] # Delay SIGTERM

Quick Start Guide

1. Add Health Endpoints: Implement /health and /ready endpoints in your application. /ready must check database and cache connectivity.
2. Configure Graceful Shutdown: Add SIGTERM handler to stop accepting requests, wait for in-flight requests, and close connections. Configure orchestrator drain timeout.
3. Implement Feature Flags: Integrate a feature flag service. Wrap all new logic in flag checks. Ensure flags can be toggled without redeployment.
4. Set Up Database Migrations: Use the Expand/Contract pattern for all schema changes. Write migration scripts that add columns, dual-write, backfill, and cutover reads safely.
5. Automate CI/CD: Configure your pipeline to run health checks, validate database compatibility, and trigger rollback on error rate spikes. Test the deployment strategy in staging.