Inventory services and data first
List application processes, databases, uploads, scheduled jobs, certificates, DNS, external callbacks and allowed source addresses. Environment settings, permissions, service definitions and dependency versions outside source control also affect recovery. Store secrets appropriately rather than placing them in public artifacts or logs.
Budget destination space for live data, backup staging and restoration work. Advertised disk capacity is not a guarantee of usable application space, and sufficient memory does not establish software-architecture compatibility. Start with the 2-core, 4GB catalog if appropriate, then verify the actual dependencies.
Choose a consistency method before copying
File synchronization can move static content, but copying an actively written database directory does not establish a consistent backup. PostgreSQL’s pg_dump documentation checked on 2026-11-01 describes consistent exports and their scope. Choose logical backup, physical backup or replication according to the database, size and recovery objective.
The rsync manual distinguishes a dry run from a real operation and explains deletion behavior. Check source, destination, trailing slashes and preview output before executing a synchronization, with a usable backup already available. There is no universally safe delete-enabled command for every deployment.
Make the write path explicit during cutover
Test TLS, login, writes, queues and background jobs on the destination through a controlled entry point. During the migration window, pause writes or use a validated replication cutover, complete the final synchronization, then reopen service. Independent writes on both machines turn rollback into a data-reconciliation problem.
Reducing DNS TTL in advance can shorten normal cache retention, but it does not make every client refresh simultaneously. Observe old-server requests, scheduled tasks and callbacks after the switch. Avoid duplicate jobs that charge customers or send notifications twice.
Define rollback before the switch
Write down the failure conditions, responsible operator, data version to restore and treatment of writes already accepted by the new server. Changing DNS back does not automatically restore a consistent database state. Retain the old host for an acceptable window while preventing uncontrolled new writes there.
Retire old resources only after acceptance, backup and billing checks. The backup guide helps design a rehearsal. Successful migration means working business paths, data and background processes, not merely a homepage returning HTTP 200.