Moving a production platform from GCP to AWS
For a cybersecurity company, I led the move of several production applications from Google Kubernetes Engine to Amazon EKS, including databases, object storage and encrypted customer data.
- Client
- Cybersecurity company
- Role
- Lead engineer
- Year
- 2026
The challenge#
The client's products ran across several Google Kubernetes Engine clusters, with Cloud SQL databases, Cloud Storage buckets and customer secrets protected by a key-management service. The business wanted everything on AWS, next to the rest of their platform, without losing data and without a long outage.
What I did#
I led the migration from discovery to cutover. I mapped every component and dependency, agreed what to retire instead of moving, and rebuilt the applications into a shared namespace on the client's existing EKS clusters using their Terraform module patterns. Rather than carrying over the old service mesh and deployment tooling, the applications joined the platform's existing Vault, Argo Workflows, logging and error tracking.
The hardest part was the data. One application stored customer fields encrypted under keys that lived in the old cloud. I wrote a re-encryption job that decrypted each value with the old key and encrypted it with the new one, then verified that every live field was converted and that every customer's data decrypted correctly on AWS.
How it works#
Each application moved through dev and then production with a written runbook. The production cutover followed a fixed order: take a database snapshot as the rollback point, lower DNS cache times, scale the old workloads to zero to freeze writes, restore the databases from a fresh export, copy the buckets with size checks and a sampled checksum, re-encrypt secrets, bring the new services up, then point old addresses at new ones with permanent redirects.
Results#
- All five applications in scope went live on AWS in one planned cutover, with health checks green and zero restarts.
- Every encrypted field on a live record was converted and verified. The only fields left as they were belonged to records customers had already deleted.
- The old cloud was scaled to zero and kept as a fallback until decommissioning.
- The team that runs the platform received a DevOps and SRE guide written for them.
What I learned#
A migration is mostly a data project with some Kubernetes attached. The rehearsal on dev, the rollback snapshot and the boring checklist were what made the production cutover uneventful.