The new cloud is running. When can we shut down the old one?

What to check before shutting down your old cloud after migration, from remaining dependencies and rollback to restore tests and final sign-off.

Traffic has moved, monitoring is green, and the team is working on the new platform. Yet the old environment is still running. A nightly export might still depend on it. And if something fails tomorrow, nobody wants to discover that the only usable recovery option was there. How do you go from a successful cutover to a well-founded decision to shut down the old environment?

The last dependency rarely shows up on the dashboard

After cutover, it is easy to see whether key applications are reachable in the new environment. It is harder to tell whether the old one is still needed. A service account might write a weekly export to an old file share. A partner might call an endpoint at a hard-coded IP address. The month-end close might query a database that barely sees any traffic during normal operations.

Review the old environment while it is still running. Load balancer access logs, database connections, storage writes, and scheduled job runs can reveal remaining dependencies. Failed connections after cutover can also expose a forgotten consumer.

How long you need to observe depends on your workflows. A quiet week tells you little about a job that runs at month-end. Run infrequent but critical workflows in the new environment or wait for their next scheduled run. For every remaining access, establish where it comes from, what it is used for, and when it will move.

What happens to data written after cutover?

On migration day, the rollback plan can look straightforward. Switch traffic back and let the old service take over. Once customers place orders, upload files, or change accounts in the new environment, things get more complicated. The old database will only contain those changes if you can verify that they were copied back.

Agree on which environment is the authoritative source of data after cutover. Record how long rollback is possible, how post-cutover writes would be carried over, and how much data loss you could tolerate in an emergency. Consider queued messages and background jobs that have already been triggered. Running a job twice could, for example, send the same notification again.

Once returning to the old data is no longer viable, you need a tested recovery procedure for the new platform.

Test recovery while the old environment is still running

Take a recent backup from the new environment and restore it in isolation. Then verify a workflow that matters to the business. For an online shop, that might mean opening an order and viewing its receipt. This verifies that files, credentials, and application configuration work together with the restored database.

Measure the time from starting the restore to having a usable application, and check how recent the restored data is. Compare both results with your agreed recovery objectives.

Look for hidden dependencies on the old environment. If the backup catalog, keys, or credentials are available only there, move them before approving the shutdown.

Shut it down before decommissioning it

Schedule a shutdown window and assign responsibility. Decide who stops the old environment, who verifies the key workflows on the new platform, and who makes the call if anything looks wrong. Stop the old services in a controlled way first. That gives you a chance to investigate an overlooked dependency before permanently removing infrastructure and data.

Set a separate date for final decommissioning. By then, decide which data and logs you will retain, who can access them, and how to restore them if needed. You can then retire the old resources, service accounts, and connections.

The shutdown checklist

  • Dependencies accounted for. Remaining access has been traced to its source, and infrequent but critical workflows have been tested on the new platform.
  • Data ownership established. There is an authoritative source of data, and the team knows when rollback is possible and under what conditions.
  • Recovery demonstrated. A recent backup has been restored and a business workflow has passed its test.
  • Shutdown window agreed. Owners, verification steps, and criteria for stopping the procedure are in place.
  • Decommissioning planned. Data retention, access, and the date for final removal have been agreed.

The migration is complete when the new platform handles day-to-day operations and the team knows how to recover it after a failure. Then you can answer the last question with confidence. Yes, the old cloud can go.