When your domain already enforces MTA-STS, an MX migration begins with an overlap policy rather than an immediate DNS cutover. Keep your legacy inbound mail server running until the new route is proven and your rollback window has closed. The reason comes down to caching: publishing a new policy today cannot erase copies external senders cached earlier, so you have to plan around every active policy revision still held across the internet. RFC 8461 details both the policy cache lifetime and the hazards of overlapping historical caches that this procedure works around.
MTA-STS focuses strictly on securing the SMTP connection between servers, as defined in the RFC 8461 transport security scope. It doesn’t authenticate the author of an email and doesn’t encrypt message contents end to end, boundaries that RFC 3207 spells out for STARTTLS itself.
Which policy expiry sets the earliest switch date?
Start by writing down exactly what senders might still have cached: current DNS MX records, every policy served recently, the last moment each version was fetched, and each policy’s max_age. Every expiration equals last_served + max_age, and the date you plan against is the latest expiration among all policies that exclude your new mail server. Shortening today’s max_age changes nothing for copies senders fetched earlier, following the historical-cache removal rule in RFC 8461 section 8.3. This is a conservative migration recommendation that I derive from the RFC’s historical-cache removal rule, not a step the RFC states in these words. If you don’t know your policy history or when versions were served, hold the change until you can establish a defensible upper bound.
DNS caching runs on its own schedule, completely separate from policy lifetimes. A DNS TTL dictates how long resolvers cache records, whereas policy max_age counts from the moment an MTA fetched the policy file; lowering a TTL today doesn’t rewrite answers resolvers already hold. Track both previous and replacement TTLs for your MX and _mta-sts TXT records, keeping DNS cache timing distinct from policy lifetime. If you’re deciding how low to set your TTLs ahead of the migration, check What DNS TTL should I use for email records?
Confirm the inbound route with your new provider instead of assuming it. As of 15 September 2026, Microsoft distinguishes an MX pointing directly to Exchange Online from one that points at an intermediary service, whose requirements you verify separately, and its multi-provider example explicitly requires the other providers’ MX hosts as well. A provider wildcard that fits one topology should not be copied into a different one (Microsoft routing and multi-provider guidance).
Two certificate surfaces, checked endpoint by endpoint
For every MX involved (old, new, and backup), test each configured endpoint, including IPv6 where advertised. A successful handshake on one IP tells you nothing about the rest of the server pool. Verify that STARTTLS responds properly, the certificate has not expired, chains to a trusted root, and lists the MX hostname in its Subject Alternative Name. These checks follow the STARTTLS and policy matching rules and SMTP certificate validation requirements in RFC 8461, as well as Google’s inbound server requirements. Testing endpoint by endpoint is an operational recommendation, not a protocol requirement.
The policy host is the second surface: mta-sts.example.net needs its own trusted, unexpired HTTPS certificate matching its hostname, and must return the policy with HTTP 200 without redirects. A functioning SMTP certificate on your MX proves nothing about the web endpoint (HTTPS policy requirements, RFC 8461 section 3.3).
If you are only renewing a certificate and the MX hostnames stay the same, your MTA-STS policy doesn’t need any edits as long as the new certificate meets all standard trust and SAN requirements. MTA-STS does not pin a certificate fingerprint. That is a conclusion drawn from SMTP certificate validation, not permission to skip the endpoint checks. That said, you still need to test every endpoint directly before going live. Keep the previous certificate deployment usable for rollback only for as long as it remains valid and uncompromised.
A worked overlap schedule from old to new MX
This exercise is invented for example.net; it is not a deployment observation and nothing was sent. It assumes a complete policy history, reliable UTC serving timestamps, normal DNS expiry, and no DANE or separate partner TLS constraints. Both MX services can deliver to the same intended mailboxes throughout. Suppose your domain example.net uses primary mail server mx-old.example.net under policy A: version: STSv1, mode: enforce, mx: mx-old.example.net, and max_age: 604800 (seven days), last served on 1 October at 09:00 UTC. Policy B allowed the same host with max_age: 86400 (one day) and was last served on 3 October at 09:00. Policy A remains valid in remote caches until 8 October at 09:00, while B expires on 4 October at 09:00. If you only waited out B’s 24 hours, you would leave four days where senders still hold policy A and reject your new server.
On 3 October at 09:00, publish overlap policy C: version: STSv1, mode: enforce, separate mx lines for mx-old.example.net and mx-new.example.net, and max_age: 86400. Once every web endpoint returns policy C, publish a new DNS TXT id; updating HTTPS before TXT is the exact order required by RFC 8461 section 8.1. In this walkthrough, assume the previous TXT TTL was 3,600 seconds and the MX TTL was 7,200 seconds, with no longer TTLs active.
| Decision point, UTC | Planned action | Hold or rollback condition |
|---|---|---|
| 3 October, 10:00 | The TXT TTL has expired; keep the old DNS MX and overlap policy C active. | Policy A still excludes the new MX: do not cut over yet. |
| After 8 October, 09:00 | Policies A and B have expired. Recheck both mail servers, then update the DNS MX to the new host while keeping policy C active. | Any TLS or mailbox routing failure: hold on the old route. |
| No earlier than 8 October, 11:00 | Wait two hours after changing DNS (assuming a 09:00 switch). Verify inbound mail delivery while keeping both servers and policy C live. | If the new route fails, restore the old DNS MX; keep the new endpoint working for cached new answers while the rollback propagates. |
| After observation window closes | Remove the old authorization only after old DNS answers have drained and the old route is no longer needed. | Never publish a new-only policy while you still need the option of rolling back to the old MX. |
These are the earliest decision times under the stated assumptions, not automatic go signals. If your deployment takes longer or one server continues serving an older policy file, recalculate based on that endpoint’s actual last-served time. The observation period is yours to choose: it should cover the senders that matter to you and the time your team needs to respond, which is why the example deliberately assigns no universal duration.
What to verify before you retire the old service
Before retiring the old mail server, cross-check three sources of evidence: your direct TLS tests, confirmed inbound delivery in user mailboxes, and TLS-RPT reports. The success counters in TLS-RPT track policy-compliant TLS handshakes rather than inbox placement, and reports generally arrive hours after the UTC day closes, as explained in the RFC 8460 timing and session counter definitions. An absence of error reports doesn’t prove every sender or backup route is working cleanly. If you haven’t set up aggregate reporting yet, review the guide chapter on TLS-RPT (RFC 8460).
Switching your policy to testing or none doesn’t purge cached enforcement rules from remote MTAs. During recovery, restore a compatible working route that satisfies the cached policy; deleting your DNS discovery record won’t revoke an active policy senders already hold (cached policy on discovery failure, RFC 8461 section 3.3).
Track the broader migration steps in the DNS change checklist. If your domain doesn’t have an MTA-STS policy yet, start with the MTA-STS rollout procedure instead.
Frequently asked questions
Do I have to update the MTA-STS policy when I only renew an MX certificate?
No. If your MX hostnames remain unchanged, you do not need to update your policy file as long as the renewed certificate meets all standard SMTP and HTTPS validation rules. MTA-STS checks hostnames, not certificate fingerprints.
Does shortening max_age today speed up the change?
No. Senders that already cached your policy keep it until its original max_age runs out. Reducing max_age today only influences senders fetching the policy after your update.
Which policy expiry sets the date for the switch?
The latest expiration among all policies that exclude the new MX host sets your earliest safe cutover date. Calculate each expiration as last_served plus max_age. This is a conservative migration recommendation derived from the RFC’s historical-cache removal rule, not a step the RFC states in these words.
Does switching the policy to testing or none clear cached policies?
No. Switching to testing or none doesn’t purge policies that senders already cached. Remote mail servers continue enforcing the cached policy until its max_age expires.
Does a working SMTP certificate on the MX cover the policy host?
No. The HTTPS endpoint at mta-sts.example.net requires its own web certificate and must return HTTP 200 without redirects. A valid SMTP certificate on your MX host proves nothing about your web server.


