Day 20: Retrospective

Phase 7 · Oct 12, 2026

Agenda (2–3 hours)

• Review (60 min): Walk through each phase of the course; for each, identify one concept you now understand deeply and one you would revisit
• Write (60 min): Document the 5 most important lessons learned from building the integration project
• Reflect (60 min): Compare your understanding on Day 1 vs today; what would you do differently on the next distributed system you build?

Course Summary

Key Lessons

1. Design before code: the architecture review checklist (Day 7) caught real design gaps before implementation
2. Idempotency is load-bearing: every write path must be safe to retry; this is non-negotiable at scale
3. Observability is a first-class feature: trace_id in logs was the single highest-value debugging investment
4. Chaos before confidence: the worker crash test revealed a state machine bug that would have caused silent data loss in production
5. SLOs frame all trade-offs: every design decision can be evaluated against the error budget — is it worth 10% of this month's budget?

What to Revisit

• Raft: implement a simplified Raft log in Rust — understanding consensus unlocks everything else
• Single-table DynamoDB: read Alex DeBrie's book cover-to-cover; the access-pattern-first discipline changes how you think about all data modeling
• OTel sampling: tail-based sampling is complex but the only correct approach for production at scale
• Tower middleware: build a custom Tower layer from scratch — the Service trait abstraction is powerful once it clicks

Tomorrow: final challenge — the complete end-to-end system review.