An email-related DNS change produces three separate milestones: the record is published in the authoritative zone, the recursive resolvers you care about reflect that publication, and the mail service depending on it operates as intended. This checklist records all three, so that the decision to proceed rests on evidence rather than on the fact that someone edited the zone. Before modifying any zone, archive the complete current state and agree on objective verification criteria, because publication alone does not mean the change is complete.
Preparing the change record
Prepare the ticket and zone permissions, naming three specific people: the change owner, the DNS implementer, and the service owner. The change record requires these core inputs before editing any DNS zone:
- the current and proposed records,
- delegation and alias dependencies,
- evidence of the prior TTL,
- the service procedure,
- an authorized recovery candidate.
Also set the maintenance window, observation endpoints, abort criteria and review schedule before anyone edits the zone. If delegation authority or update permissions remain unclear, consult the DNS owner and keep publication on hold. These are recommended change controls, not RFC approval requirements.
Copy both blank tables below. Maintain one change record for every DNS name and resource type requiring an independent decision; grouped changes that cannot be separated belong in a single ticket, with each record documented in full. An RRset represents all records sharing the same name, type, and class. Preserve complete RDATA and TTL values alongside zone revision references, because revision numbers alone cannot show what changed (DNS record fields in RFC 1035). For long values, you may use an accessible, versioned verbatim attachment with a stable evidence ID. Where data is missing, write unknown — owner — next check time, or not applicable — reason when a field genuinely does not apply. Never treat an unknown as a passing result.
| Change field | Before / planned | After / observed |
|---|---|---|
| Change ID, purpose, affected mail identity/service | [fill] | [fill] |
| Change owner, DNS implementer, service owner, approval/time | [fill] | [fill] |
| DNS owner name, type/class, whole RRset/RDATA and TTL | [fill] | [fill] |
| Zone/account, delegated authorities, aliases/dependencies and version | [fill] | [fill] |
| Positive/negative cache evidence; application/policy cache and transit dependencies | [fill] | [fill] |
| Criteria, observation endpoints, allowed next action | [fill] | [fill] |
| Recovery candidate, its authorization/restrictions and owner | [fill] | [fill] |
| Decision, evidence IDs, decision owner/time and next check | [fill] | [fill] |
Add an observation row for every planned query endpoint and functional test. Each row captures the exact query, the endpoint hostname and role, the UTC timestamp, the status, the flags, returned answer records, and authoritative evidence including alias chains. Save the full dig transcript, because two kinds of absence mean different things: none means the section came back empty, while not captured means the evidence is missing. Value-only +short output cannot support that distinction; review the DNS response structure in RFC 1035 to see what a complete transcript contains. For query mechanics, see Verifying records with dig.
| Evidence ID / UTC | Question or functional input | Endpoint / role | Expected result / criterion | Full observed result and TTL/version | Result | Owner / next action |
|---|---|---|---|---|---|---|
| [fill] | [fill] | [fill] | [fill] | [fill] | [pass / hold / error / not checked] | [fill] |
Which DNS answer counts as evidence?
Verify that each authoritative name server queried belongs to the officially delegated name server set, because an unparented server can answer authoritatively without having delegation authority. A recursive result describes that resolver at that moment, not every receiver, so choose recursive test endpoints in coordination with the service owner (server roles in RFC 9499). Check the AA flag carefully: it confirms authoritative status for the queried name or initial answer record, not necessarily for a subsequent CNAME target (AA and response flags).
Tell a positive answer apart from NXDOMAIN, which means the name is absent, from NODATA, which means the requested type is absent, and from a referral. Follow CNAME alias chains through to their terminal name. SERVFAIL and REFUSED indicate operational errors, while network timeouts return no DNS status at all. If evidence is missing or truncated, keep the criterion on hold (response status and truncation).
Base TTL planning on the actual prior state rather than arbitrary defaults (advance TTL planning in RFC 1034). Lowering a published TTL does not shorten copies already cached on remote resolvers. If upstream authoritative servers are down, resolvers running serve-stale (RFC 8767) may answer with expired records while presenting a fresh TTL counter. If you suspect serve-stale caching during an incident, check directly with the resolver operator (refresh and serve-stale conditions). For recommended values, see What DNS TTL should I use for email records?.
When adding a record where the name or type was previously absent, record the negative response and SOA record. The initial negative TTL is the smaller of the SOA TTL and the SOA MINIMUM field (negative-answer lifetime). Record this remaining negative cache lifetime separately from the newly published record TTL, because negative cache entries decrement continuously while cached (negative caching).
Which SPF, DKIM and MTA-STS dependencies does the change touch?
SPF: concatenate split TXT strings without inserted spaces. Require exactly one SPF policy record per domain: zero records produce a result of none, while publishing multiple records triggers an immediate permerror failure (SPF selection). Validate the IP, domain and sender inputs alongside syntax rules. Standard evaluation constraints apply: a global limit of 10 DNS lookups across include/a/mx/ptr/exists/redirect mechanisms, address query limits, and a recommended limit of two void lookups (evaluation limits). While mail in transit may still be checked, keep the legitimate old authorization in place (transition care).
DKIM: record the d= domain and s= selector values, which together form the selector lookup name. During routine key rotation, publish the new public key under a new selector while keeping the old public keys published through the relevant transit overlap; get evidence of the actual mail lifecycle before you retire a key (selector overlap).
MTA-STS: track four elements independently: the DNS TXT id, its DNS TTL, the HTTPS policy body and fetch evidence, and the policy max_age parameter. The HTTPS policy body SHOULD be updated before the TXT signal, and recipients SHOULD keep old policies workable during the update-discovery period governed by the TXT TTL (update order). Fetching senders must not use ordinary HTTP caching and must apply a valid cached policy if live discovery or fetch fails (policy-fetch rules). Deleting the DNS TXT record or HTTPS policy file does not remove protection immediately: removal has to account for every previously served policy and their overlapping lifetimes, so use a dedicated transport procedure (removal conditions).
These checks verify protocol dependencies; for full transition procedures, consult SPF maintenance, DKIM rotation, and MX/certificate changes. To validate your full authentication deployment, review How do I verify my SPF, DKIM, and DMARC setup?.
A worked decision: DNS-17
Synthetic example: the observations are invented, and no DNS queries or sends were performed. All timestamps represent UTC on 14 September 2026. Domain names and IP ranges derive from reserved example domains and documentation IPv4 blocks.
RRsets O and N represent complete single-record sets without additional TXT records at this host:
O: mail.example.com. 1800 IN TXT "v=spf1 ip4:192.0.2.25 -all"
N: mail.example.com. 600 IN TXT "v=spf1 ip4:192.0.2.25 ip4:198.51.100.25 -all"
In the observations below, the TTL shown is the observed one, not the nominal TTL in O or N. The new 600-second value is a scenario choice, not a recommendation. Documented parent delegation specifies authoritative name servers ns1.example.net. (192.0.2.53) and ns2.example.net. (192.0.2.54). Recursive resolvers are resolver-a.example.org (203.0.113.53) designated R1, and resolver-b.example.org (203.0.113.54) designated R2. No CNAME aliases exist in this path.
Anna acts as change owner, Ben implements DNS modifications, and Celia manages the mail service. Verification proceeds through three sequential gates:
- P: both documented authoritative endpoints return N with AA=1 and the anticipated zone revision.
- R: both designated recursive resolvers return N RDATA. This represents an internal operational gate, not global propagation.
- F: only after gate R passes may Celia send one controlled test message through the new service to
mx.receiver.example.org. Before activating traffic, capture receiver-side evaluation logs demonstrating SPFpassfor connecting IP 198.51.100.25, domainmail.example.com, and envelope senderbounce@mail.example.com. This test confirms the updated SPF dependency; it proves neither full message authentication nor inbox delivery.
At 09:56 Celia documented an offline policy review, E0. She evaluated each candidate source IP against domain mail.example.com and sender bounce@mail.example.com, comparing static policies O and N rather than active DNS lookups.
| E0 source IP | O result | N result |
|---|---|---|
| 192.0.2.25 | pass | pass |
| 198.51.100.25 | fail | pass |
| 203.0.113.25 | fail | fail |
Reconstruct table E0 through exact IP matching, followed by the matching all mechanism across left-to-right evaluation and qualifiers. Neither policy uses any DNS-querying mechanism, so each costs 0 lookups and 0 void lookups; the initial TXT retrieval does not count against evaluation limits (term counting). These are calculations on paper, not SMTP observations.
| Change field | Before / planned | After / observed |
|---|---|---|
| Change ID, purpose, affected mail identity/service | DNS-17: add approved source 198.51.100.25 for bounce@mail.example.com; new service disabled; approved old source 192.0.2.25 active | Publication completed; same purpose; new service remains disabled; old source remains active |
| Change owner, DNS implementer, service owner, approval/time | Anna / Ben / Celia; Anna approved publication only in plan-1 at 09:50 | Ben published at 10:00; Anna reviewed at 10:03; no activation approval |
| DNS owner name, type/class, whole RRset/RDATA and TTL | mail.example.com. TXT IN; O, TTL 1800 | N, authoritative TTL 600; observations E1–E4 |
| Zone/account, delegated authorities, aliases/dependencies and version | Example DNS / zone example.com; ns1/ns2 as defined above; no aliases; revision 2026091401 | Same account, zone and authorities; revision 2026091402; no other DNS records changed |
| Positive/negative cache evidence; application/policy cache and transit dependencies | B1–B4: positive O replies; negative-cache observation not applicable because no absence observed; no HTTPS-policy or DKIM change; old source retained | E4: R2 returns O; cause and insertion time unknown; no policy-cache or transit retirement claimed; old source/authorization retained |
| Criteria, observation endpoints, allowed next action | P then R then F; ns1/ns2/R1/R2 and controlled receiver above; F permitted only after R | P passes; R holds at R2; F not run; do not activate new service |
| Recovery candidate, its authorization/restrictions and owner | O only while new source remains disabled and unsent; Anna approves recovery; Ben republishes and repeats P/R and service checks | No rollback requested: retain N and old source. If new source has sent, review transit before removing its authorization; escalate compromised/withdrawn authorization |
| Decision, evidence IDs, decision owner/time and next check | Anna’s publication-only approval, plan-1, 09:50; E0 added 09:56; review planned 10:03 | Anna, 10:03: HOLD activation; P=pass/R=hold/F=not checked; E1–E4 and F0; Ben/Celia recheck 10:20 and escalate unresolved evidence to the relevant operator |
In the test table below, authoritative queries (ns1, ns2) return AA=1/RD=0/RA=0 (flag set A), while recursive queries (R1, R2) return AA=0/RD=1/RA=1 (flag set R). All responses return RCODE=NOERROR, TC=0, 1 answer RR, zero CNAME aliases, and none in the authority and additional sections for query symbol Q (mail.example.com. IN TXT). Reconstruct responses from these parameters, records O or N, and the observed TTL. Revisions come from the change log at the same time, not from the TXT reply; recursive zone revisions are not observed.
| Evidence ID / UTC | Question or functional input | Endpoint / role | Expected result / criterion | Full observed result and TTL/version | Result | Owner / next action |
|---|---|---|---|---|---|---|
| B1 / 09:55 | Q | ns1.example.net. / authoritative | O baseline | O, TTL 1800; A flags; revision 2026091401 | Baseline saved | Ben / compare after publication |
| B2 / 09:55 | Q | ns2.example.net. / authoritative | O baseline | O, TTL 1800; A flags; revision 2026091401 | Baseline saved | Ben / compare after publication |
| B3 / 09:55 | Q | R1 / recursive | Save baseline view | O, remaining TTL 1200; R flags; revision not observed | Baseline saved | Celia / compare same endpoint |
| B4 / 09:55 | Q | R2 / recursive | Save baseline view | O, remaining TTL 1500; R flags; revision not observed | Baseline saved | Celia / compare same endpoint |
| E1 / 10:02 | Q | ns1.example.net. / authoritative | N / P | N, TTL 600; A flags; revision 2026091402 | Endpoint passes | Ben / retain evidence |
| E2 / 10:02 | Q | ns2.example.net. / authoritative | N / P | N, TTL 600; A flags; revision 2026091402 | Endpoint passes | Ben / P complete |
| E3 / 10:02 | Q | R1 / recursive | N RDATA / R | N, remaining TTL 590; R flags; revision not observed | Endpoint passes | Celia / retain evidence |
| E4 / 10:02 | Q | R2 / recursive | N RDATA / R | O, remaining TTL 1080; R flags; revision not observed | Hold | Ben / recheck 10:20; ask operator if unresolved |
| F0 / 10:03 | Intended MAIL FROM bounce@mail.example.com; source 198.51.100.25; domain mail.example.com | mx.receiver.example.org / controlled receiver | After R, receiver-side SPF pass for these inputs / F | Not run: R unmet; no SMTP or receiver observation; DNS TTL/version not applicable | Not checked | Celia / keep service disabled |
Artifacts E1 and E2 satisfy gate P, but E4 shows R2 continuing to return legacy record O, leaving gate R unfulfilled despite E3 passing. Test F remains blocked, and Anna places service activation on hold. Log E4 does not reveal why resolver R2 continues returning record O. The 10:20 appointment is a staffed recheck, not a prediction of cache expiry or global convergence.
Holding, recovering or closing
Prior to publication, reasons to hold include unapproved tickets, unexpected baseline values, or incomplete dependency or preflight evidence. Following publication, keep service activation or record retirement on hold whenever:
- authoritative servers disagree,
- published values differ from the approved plan,
- records are unexpectedly absent or return DNS errors,
- recursive resolvers display unresolved inconsistencies,
- required service tests fail or remain unperformed.
Escalate raw evidence directly to the owner of the unresolved component, whether that is the DNS owner, the resolver owner, or the mail or HTTPS owner.
Roll back only to a freshly approved recovery candidate that remains safe to run, repeating the full verification process. Compromised keys or revoked sender permissions go to the incident owner, because previous use is not permission to restore them. Maintain the documented baseline and operational decision history in your worksheet. When any required check fails or cannot be executed, record HOLD with an assigned owner and an explicit recheck time. Mark the change closed only when all agreed verification criteria pass and the accountable change owner formally approves completion.
Frequently asked questions
Does lowering the TTL shorten copies that are already cached?
No. Lowering a published TTL does not shorten copies that are already cached. Plan TTL preparation from the actual prior state rather than from a universal setting.
What is the difference between NXDOMAIN and NODATA?
NXDOMAIN means the queried domain name does not exist in the zone. NODATA means the domain name exists, but carries no records of the requested type.
Does an old value with a positive TTL prove the resolver served stale data?
No. An old value returned with a positive TTL proves neither freshness nor serve-stale behavior. When authoritative servers are unreachable, resolvers with serve-stale enabled may return expired records while presenting a fresh positive TTL.
Does deleting the MTA-STS TXT record remove protection immediately?
No. Deleting the TXT record or the HTTPS policy does not remove protection immediately. Removal has to account for every previously served policy and their overlapping lifetimes, so use a dedicated transport procedure.
When can the change be closed as successful?
Only when all agreed verification criteria pass and the accountable change owner formally approves closure. If any required check fails or remains unperformed, mark the change on hold with an assigned owner and an explicit recheck time.


