Why Migrations Fail at the DNS Step, Not the Data Step
Most dedicated server migrations copy data correctly and then lose hours at the cutover, because the cutover is the only part that depends on a third party you do not control: resolver caches. A DNS record with a 24-hour TTL can keep a meaningful share of visitors pointed at the old server for a full day after you flip it, regardless of how clean the rsync was. The fix is scheduling, not technology.
The timeline below is built backwards from a target cutover hour. Every step has a measurable completion criterion, so the migration can be aborted at any checkpoint rather than discovered to be broken after the fact.
| Day | Action | Completion criterion |
|---|---|---|
| T-14 | Lower TTL to 300 s | Record shows 300 in authoritative zone |
| T-7 | Build new server, baseline config | Stack installs cleanly, ports open |
| T-3 | Full data sync + dry-run restore | Checksums match, app boots |
| T-1 | Delta sync, smoke tests on hostfile | All checkpoints pass over IP |
| T-0 | Final delta, freeze, flip A record | New IP serving, old read-only |
| T+1d | Verify propagation, retire old | Old IP traffic below 1% |
Step One: Drop the TTL Two Weeks Early
Lower the TTL on every record that will change — A, AAAA, and any CNAMEs in the chain — at least one full old-TTL period before the cutover. If the current TTL is 86,400 seconds, that means 14 days ahead, because caches retain the old value for the full original duration. Lowering TTL on the day of the move accomplishes nothing.
Verify the change from an external resolver rather than the provider control panel, since panels often show the intended value while the authoritative servers still serve the old one. Query the record’s authoritative nameserver directly and confirm the TTL in the answer section.
Step Two: Build the Target Before You Touch the Source
Provision the new dedicated server with the same OS, package versions, and firewall rules as the source, and record the differences deliberately rather than discovering them during cutover. Install the stack, open the required ports, and confirm TLS certificates can be issued for the new host before DNS moves — if your certificate uses HTTP-01 validation, that validation needs to reach the new server, which usually means either a DNS-01 challenge or issuing the certificate after the flip.
Do not skip the reverse DNS and IP reputation check. A new IP with poor reputation in mail or anti-fraud databases causes failures that look like application bugs. Send a test message and check the new IP against the major blocklists before committing.
Step Three: Sync Data and Prove the Restore
Run a full sync, then restore the application from the synced copy on the target and confirm it boots. A backup that has never been restored is not evidence of anything. Record checksums on both sides for the critical data set so the delta sync can be verified as complete rather than assumed.
For databases, this is the step where the plan usually breaks. A dump-and-restore takes a consistent snapshot but costs downtime proportional to size; binary log replay gives near-zero downtime but requires the target to be configured as a replica before the freeze. Decide which before T-3, because switching approaches mid-cutover is how migrations overrun.
Step Four: The Freeze, Final Sync, and Record Flip
Enter write freeze, run the final delta sync, verify checksums, and flip the A record. The order matters: flipping before the final sync means writing to a server that is about to be replaced. Keep the old server running and reachable by IP so rollback is a record change, not a rebuild.
Test immediately by forcing your own resolver to the new IP — editing the local hosts file is faster and more reliable than waiting for propagation. Run the full smoke-test list against the new server by IP before the TTL change has time to spread. If any check fails, revert the record; the window for a clean rollback is measured in minutes.
Our dedicated server comparison table is useful here for confirming the new host’s IP allocation, port configuration, and whether additional IPv4 addresses are available for staged service-by-service cutover.
Step Five: Verify Propagation and Handle Stragglers
Poll public resolvers and your own monitoring for traffic arriving at the old IP. Watch for two categories of straggler: resolver caches honouring an unexpectedly long TTL from a record you missed, and hardcoded IPs in client configurations, mobile apps, or third-party integrations. Hardcoded addresses are the most common cause of a “migration” that generates tickets for weeks.
Keep the old server answering with a redirect or a maintenance page for at least one full TTL period. Do not decommission on cutover day. Confirm old-IP request volume has fallen below 1% of normal before powering down, and retain the disk for another two weeks.
Rollback Criteria and the Post-Migration Review
Define rollback triggers in writing before the window opens: error rate above X percent for more than ten minutes, latency above Y milliseconds at the 95th percentile, or any data write failure. Without numeric thresholds, the decision defaults to whoever is most tired at 3 a.m.
After the window closes, review what the timeline actually looked like against the plan, and record the deltas — final sync duration, propagation time, and any missed record. Those numbers make the next migration shorter. Teams that run this checklist twice typically cut the second window by half, largely by removing the guesswork around sync timing and TTLs. Comparing dedicated server plans with migration support included can also remove the rebuild step entirely for smaller workloads.




Leave a Reply
You must be logged in to post a comment.