Why not just lift‑and‑shift the entire monolith to Kubernetes?

Because moving a bottleneck doesn’t remove it. Under an eight‑week freeze, we optimized the critical path. Containerizing the whole monolith would burn time and create more moving parts without hitting the business SLOs.

Why Debezium instead of dual‑write or batch ETL?

Dual‑write risks inconsistency and code churn; batch ETL adds staleness. Debezium gave near‑real‑time sync with minimal app changes and a clean rollback story.

Did you consider a service mesh like Istio?

We did, and skipped it for timeline/complexity. NGINX Ingress + Resilience4j covered our needs. If you already run Istio well, it can add policy and telemetry; don’t add it during a crunch unless it’s muscle memory.

How did you prevent a canary from hurting users?

Strict guardrails: pre‑agreed abort thresholds tied to SLOs, Argo Rollouts analysis with Prometheus queries, and OpenFeature targeting so early exposure hit low‑blast‑radius segments first.

What about the AI‑generated code you found?

We audited hotspots, removed O(n²) eligibility logic and speculative null guards, added property‑based tests, and wrapped critical calls with circuit breakers. We’ve formalized this as our vibe code cleanup and AI code refactoring service.

Case-studies · Nov 30, 2025 · 10 minute read

The Promo Engine That Blocked a Holiday Launch — And the 6‑Week Modernization That Freed It

An anonymized retail engagement where a strangler-fig carve‑out, GitOps, and a sane rollout plan unblocked a make‑or‑break launch without a big‑bang rewrite.

Back to all posts

The Promo Engine That Blocked a Holiday Launch — And the 6‑Week Modernization That Freed It

Key takeaways

Implementation checklist