3.1.1.9. Configuring Applications and Related Services to Support Multiple Availability Zones and Regions While Minimizing Downtime

3.1.1.9. Configuring Applications and Related Services to Support Multi-AZ

Deploying infrastructure across AZs is only half the work — your application code and dependent services must also handle multi-AZ operation correctly.

Application-level considerations:

• Session management: Don't store sessions in instance memory. Use ElastiCache (Redis), DynamoDB, or ALB sticky sessions (less preferred — creates AZ affinity).
• DNS caching: Set application-level DNS TTL low (60s) so failovers take effect quickly. Java's default is to cache forever.
• Connection retry logic: Database and cache connections should retry with exponential backoff when a failover occurs.
• Idempotency: During failover, requests may be retried. Ensure write operations are idempotent (using unique request IDs) to prevent duplicates.

Service-level multi-AZ configuration:

• RDS: Enable Multi-AZ in DB instance settings. Application uses the same endpoint — DNS transparently routes to the active instance.
• ElastiCache Redis: Enable Multi-AZ with auto-failover in the replication group. Application must handle brief connection drops during failover.
• EFS: Multi-AZ by default — mount targets in each AZ's subnet.
• SQS/SNS: Multi-AZ by default. No configuration needed.

# Connection retry pattern for RDS failover
import time
import psycopg2

def get_connection(max_retries=5):
    for attempt in range(max_retries):
        try:
            return psycopg2.connect(host=RDS_ENDPOINT, dbname=DB_NAME)
        except psycopg2.OperationalError:
            wait = 2 ** attempt  # Exponential backoff: 1, 2, 4, 8, 16s
            time.sleep(wait)
    raise Exception("Failed to connect after failover")

Exam Trap: EBS volumes are AZ-specific — they can't be shared across AZs. If an instance fails in AZ-A and is replaced in AZ-B, the new instance can't access the old EBS volume. For shared storage across AZs, use EFS (NFS) or move data to S3/DynamoDB.