Our Playbook for a Zero-Downtime Cloud Migration

The scariest sentence a client can say to us is “this system can’t go down.” It is also the most common. Order systems, payment flows, patient records — the things most worth migrating are exactly the things you cannot switch off. So we have built our migration practice around one promise: your users should not be able to tell it happened.

We map before we move

Before a single resource is provisioned, we spend time understanding what actually talks to what. Hidden cron jobs, that one report nobody owns, the integration a partner relies on — these are what break migrations. We document the real dependency graph, not the one in the outdated wiki.

We run old and new in parallel

Rather than a big-bang cutover, we stand the new environment up alongside the old one and route a trickle of real traffic to it. We compare outputs, watch the dashboards, and grow the share only as our confidence grows. If anything looks wrong, rolling back is a routing change, not an emergency.

• Replicate data continuously so the new system is always warm.
• Shift read traffic first — it is the safest to reverse.
• Shift writes behind a feature flag, percentage by percentage.
• Keep the old system on standby until the new one has earned trust.

“A migration nobody noticed is the highest compliment our infrastructure work can get.”

Done this way, the cutover itself is anticlimactic — which is exactly the point. By the time we flip the last switch, the new environment has already been quietly serving production for days. We just turn the old one off.