Why not just rewrite the monolith into microservices?

Because deadlines don’t care about architecture purity. A rewrite would have blown the date and multiplied failure modes. We carved one seam behind a façade and flag, which produced immediate risk reduction without destabilizing the rest.

Could we have skipped Istio?

You could use NGINX plus retries/timeouts, but the outlier detection and circuit breaking we needed were easier with Istio in this environment. We kept it minimal: no fancy multi-tenant mesh, just what protected the integration.

What if we don’t have Kafka or Debezium?

Start with read replicas and a queue. If CDC is too heavy for your org right now, create a pub/sub contract from the monolith first. The goal is decoupling, not tech for its own sake.

How do we apply this if we’re on ECS, not EKS?

Same playbook: GitOps via CodePipeline/Spinnaker or ArgoCD-on-ECS, canaries via ALB weighted target groups, and circuit breaking via service mesh proxies or Envoy sidecars. Tools change, principles don’t.

Case-studies · Nov 7, 2025 · 8 minute read

The Black Friday Launch That Our Legacy Stack Couldn’t Survive—Until We Modernized Just Enough

A retail logistics platform was staring down a contractual go‑live with a national big-box retailer. The stack was a Java 8 monolith, shared database, and a Jenkins snowflake. We had eight weeks. Here’s what we changed—and what moved the needle.

Back to all posts

The Black Friday Launch That Our Legacy Stack Couldn’t Survive—Until We Modernized Just Enough

Key takeaways

Implementation checklist