Operational Excellence, Security & Sustainability

Source: AWS Well-Architected Tool:

• Operational Excellence: OPS 1 – OPS 11 (11 Questions)
• Security: SEC 1 – SEC 11 (11 Questions)
• Sustainability: SUS 1 – SUS 6 (6 Questions)
  Core Concept: Ensuring systems are observable, safe to deploy, strictly protected at every layer, and environmentally efficient.

1. Operational Excellence (OPS 1 – OPS 11)

2. Observability Architecture: The 4 Golden Signals & Distributed Tracing

In an Amazon interview, saying "I will monitor the system with CloudWatch" is too generic. You must specify the exact metrics, aggregation methods, and distributed tracing strategies:

Synthesizing vector architecture diagram...

The 4 Golden Signals (Google SRE & Amazon Standard)

1. Latency:
   • Trap: Never monitor average (mean) latency! Averages mask severe P99 outliers.
   • Standard: Monitor P50, P90, P99, and P99.9 latency. For an e-commerce API, target P99 $<150\text{ms}$.
2. Traffic:
   • Measure request rate (Requests Per Second - RPS / QPS) and distinguish read vs. write volume.
3. Errors:
   • Track absolute error count and error rate: $\text{Error Rate} = \frac{\text{HTTP 5xx Count}}{\text{Total Requests}}$.
   • Set high-priority paging alarms if 5xx rate exceeds $1\%$ over a 2-minute rolling window.
4. Saturation:
   • Measure resource fullness: CPU utilization, memory pressure, database connection pool exhaustion, and message queue backlog (ApproximateNumberOfMessagesVisible).

3. Safe Deployment Pipelines: Canary vs. Blue/Green

Synthesizing vector architecture diagram...

When to Choose:

• Canary Deployment (Amazon's Primary Model):
  • Traffic is gradually shifted (e.g., 10% for 15 minutes, then 100%).
  • If the canary's 5xx error rate or P99 latency spikes compared to the stable fleet, deployment automatically rolls back before the majority of users are exposed.
• Blue/Green Deployment:
  • Good for database schema migrations where two versions cannot run concurrently, but requires double the compute infrastructure during the cutover window.

4. Security (SEC 1 – SEC 11)

5. KMS Envelope Encryption Deep Dive

Amazon interviewers love to ask: "How do you secure petabytes of user data without bottlenecking AWS KMS?"

If every database read or write made a direct call to the AWS KMS service over the network, KMS would throttle, latency would spike, and KMS API costs would explode. The solution is Envelope Encryption:

Synthesizing vector architecture diagram...

• Why it wins:
  1. Performance: Only one small KMS network call is needed per session/batch to get a data key. Millions of data blocks are encrypted locally in memory at hardware speeds.
  2. Security: Even if the database storage is compromised, the attacker cannot decrypt the data without calling KMS to decrypt the EDK, which is protected by strict IAM and CloudTrail auditing.

6. Sustainability Pillar (SUS 1 – SUS 6)