Skip to content

DNS Change Detection

Monitoring 15 min read
On this page

A DNS monitor records what your authoritative nameservers answer on its first sweep, then compares every later sweep against that recording. When an answer stops matching, Uptimia opens a change episode on that record and, depending on the record group, pages you. Setup and nameserver health belong to DNS Monitoring.

Before You Start #

  • You need a DNS monitor that has completed at least one sweep. Until then the Watched records table reads "The baseline sweep records the watched records within a minute."
  • Accept change and Mark dynamic need an Owner, Admin or Editor seat.
  • DNS monitors sweep every 5 minutes. The cadence is fixed: there is no frequency control on the editor.

How the First Sweep Sets Your Baseline #

The first sweep is a baseline sweep: it records an answer for every watched record and opens nothing.

  • Only authoritative answers count. A reply is used only if the server was reachable, the probe was trustworthy, and the answer carried the AA (authoritative answer) bit. Some servers are reachable but answer without AA: a lame secondary, or a resolver mistaken for a nameserver. Such a reply cannot seed the baseline, open a change or block a restore. It surfaces as a Not answering authoritatively health finding instead.
  • Values are compared as sorted sets. Round-robin order is normalized away, so four A records returned in a different order every sweep are not a change.
  • The baseline is per record. Each watched name and type (www A, @ MX, _dmarc TXT) carries its own values, TTL and the time it was set.

A record added later gets its baseline on the first sweep that answers it, with no episode. Removing a watched name deletes its baseline keys, so re-adding it starts clean. Changing the zone resets the baseline entirely.

What Counts as a Change #

An episode opens when any reachable, authoritative nameserver serves a value set that does not equal the baseline. One disagreeing server opens it; every reachable authoritative server must agree again before it closes. A record marked dynamic has its own rule, described under Mark Dynamic below.

Severity comes from the record's group, set on the Watched records card in the editor.

Group (editor label)Records comparedDefault severity
Apex & host addressesA, AAAA and CNAME for every watched nameCritical
MXMX at the zone apexCritical
NSNS at the zone apexCritical
Policy recordsTXT at the apex, TXT at _dmarc, CAA at the apexWarning

On a sending domain, watch the MX and Policy records groups. For how they sit alongside blacklist and uptime checks, see Monitoring Your Email Infrastructure.

Each group's select offers Critical, Warning and Info only. Critical and Warning both open an incident and alert your contacts; the alert reads CRITICAL or WARNING. Atlassian Statuspage posts nothing for a Warning change, because it describes your own zone edit rather than an outage. A Warning change shows on the monitor page as Warning · <record>, and its sweep log row carries an amber N records differ badge. Info only records the episode without opening an incident, and turning a group's toggle off does the same thing: the change is still recorded at info severity, so the history stays complete.

Three things are recorded and never page, whatever the group is set to:

  • A TTL-only change. If the values match and only the TTL moved, the episode is info. It never blocks a restore either: a record whose values are back but whose TTL still differs still closes.
  • A record that could not be read this sweep. A SERVFAIL, REFUSED or timeout is a check error. That record's state is frozen rather than judged, and the sweep row carries a "· N check errors" suffix.
  • A value change on a record marked dynamic, unless the answer set empties out.

An empty answer set is a legitimate value in both directions. A record that disappears shows as NOERROR, 0 answers — rrset deleted, a name that had nothing shows (no record) on the "was" side, and a CNAME'd name with an empty A rrset seeds a normal baseline rather than an error.

Reading a Change on the Watched Records Table #

Watched records is the first card on the monitor page. Its header names the sweep it reflects, then one of "as of the 14:20 sweep · all servers agree" ("both servers agree" on a two-nameserver zone), "changed from baseline" or "servers disagree". The roll-up on the right reads "1 changed · 10 at baseline".

Records sit in five groups: Apex addresses, Hosts, Mail routing, Policy records and Infrastructure. A group whose records are all at baseline collapses to its header row; one holding a changed, accepted or restored record starts expanded. Clicking a header overrides either way.

The columns are Record, Answer, TTL and Last changed. Answer shows what the nameservers serve now, with a struck-through was … sub-line underneath during and after an episode. Last changed carries the state:

Last changed readsStateWhat it means
Aug 12 — baselineAt baselineThe answer has matched since the baseline was set
changed 09:41Active episodeA server is serving something else. Accept change and Mark dynamic appear under it
Aug 12 · accepted ✓Closed by a personSomeone accepted the new answer as the baseline. The row says who, with any note
Aug 12 · restored 10:15 ✓Closed by sweepsThe values came back and the restore was verified

A restored row's was … line carries the window the wrong answer was live for, as in was 203.0.113.9 09:41–10:05. You can read the exposure straight off the table.

The Watched records card of a DNS monitor in Uptimia, with one changed A record showing its struck-through previous answer and the Accept change and Mark dynamic actions.Enlarge

Accepting a Change: Making the New Answer the Baseline #

Accept a change that was expected and that you intend to keep: a migration, a new mail provider, a corrected SPF record, a verification token. You or anyone on your team may have made it.

  1. Click Review change in the alert email, or Accept change on the changed row. With several changes open, Accept all N in the card header handles them in one decision. The Was this change expected? modal opens with the values that moved at the top: added values marked +, removed ones struck through.
  2. Type a note if you want one. Notes are optional, capped at 500 characters, and included in the notice your team receives.
  3. Click Accept — new baseline (Accept N changes — new baseline when you accept several).
The Was this change expected? modal in Uptimia, showing the DNS values that moved, the optional note field and the Accept — new baseline button.Enlarge

If you did not expect the change, leave it open. When the old value comes back, the alert closes by itself.

The accept button stays disabled in two cases, and the modal says why:

  • A nameserver still serves a different value. You can accept once the change has reached every nameserver; otherwise that server would alert in reverse on the next sweep.
  • The record changed again since you opened the modal. Cancel and review the current value.

Accepting is permanent:

  • The episode's new values become that record's baseline. Every later sweep is compared against them.
  • The episode closes as accepted, stamped with your email address and your note.
  • If nothing else alertable remains open, the incident closes and everyone who received the alert gets a notice saying who accepted it. The email includes your note. Atlassian Statuspage and WhatsApp messages do not name the person. Turning off Alert when records are restored and verified turns this notice off too.
  • A DOWN alert still held back by a recipient's alerting hours is delivered together with the notice.
  • If a manual incident is open on the monitor, accepting moves the baseline and the record stops counting as changed, but the manual incident stays open and no one is notified.

Accepting works only on an open record change. It does nothing on a health finding, or on an episode that has already restored or been accepted.

In the root-cause block of the alert email, an accepted close prints Restore verification: Closed by acceptance — the new answer is now the baseline, never "restored after N matching sweeps".

Mark Dynamic: CDN Pools, GeoDNS and Round-Robin Answers #

Use Mark dynamic when the values are supposed to rotate: a CDN edge pool, a GeoDNS answer, a load-balanced address set. Accepting such a record only re-arms the same alert on the next rotation.

Clicking Mark dynamic opens the Mark this record dynamic modal. Confirming adds the record to the monitor's dynamic list and accepts the open change at the same time, with the fixed note marked dynamic — values are allowed to rotate.

The Mark this record dynamic modal in Uptimia, explaining that dynamic records are compared by presence and type only and that marking it also accepts the open change.Enlarge

From then on the record is compared by presence and type only. It opens an episode in two cases only: when its answer set becomes empty (the record disappears), and when it comes back after being empty. Value and TTL differences are ignored. The row picks up a dynamic — presence only, values rotate freely badge, and the Records stat cell counts it in its "N changed · M dynamic" sub-line.

Some DNS providers answer a record to most regions and answer it empty to a few, so one location's empty answer does not prove a deletion. When a sweep sees a dynamic record appear or disappear, Uptimia asks the monitor's other selected locations, one at a time, and never locations outside your selection. Five in a row seeing the same thing (all of them, on a monitor with fewer than five other locations) open the episode, dated at the first sweep; the first location that still answers as the baseline does ends the check and nothing opens. If the other locations give no usable answer, the episode opens as it would on a normal record. While the check runs, the row carries the note The last check from this location differs from the baseline. A change is recorded only if this monitor's other locations agree.

Closing works the same way: once the record answers as the baseline again, three other locations must confirm it (all of them, if there are fewer than three), or six clean sweeps in a row close it. A monitor with a single location has nobody to ask, so three sweeps in a row open an episode and six clean ones close it.

Because only your locations are asked, a record that disappears from just part of the world may not alert. With all locations selected, losing it in 75% or more of them opens an episode within a few sweeps and losing it in half can take a few hours; losing it in a tenth does not open one. If every selected location stops seeing a record, an episode opens even if other regions still get it.

Warning: marking a record dynamic cannot be undone from the control panel. The editor has no dynamic-records control, there is no button to un-mark a record, and re-saving the monitor preserves the list. Removing a watched name drops the dynamic markings for that name's address records, but the five zone-level records (@ MX, @ NS, @ TXT, _dmarc TXT, @ CAA) are always watched. Clearing a marking on one of those can only be done from a script (see Getting Started with the Uptimia API).

Verified Restores: Why a Change Takes Several Sweeps to Clear #

A change episode closes itself when the values come back. By default it does not close on the first matching sweep, because a half-propagated zone or one stale secondary would close it early.

On each sweep where every reachable authoritative nameserver answers the baseline value for that record, the clean streak advances by one. When the streak reaches the monitor's setting, the episode is marked restored and verified. A sweep that still differs resets the streak to zero; a sweep that could not read the record neither advances nor resets it.

Set the depth on the Sensitivity card in the editor (Actions → Edit from the monitor page).

SettingValuesDefaultUnit text beside it
Treat a change as restored after1, 3 or 5 matching sweeps3 matching sweepsspans 10 min at this cadence · every server must agree; at 1, restored at the first matching sweep

At the fixed 5-minute cadence, 3 matching sweeps span 10 minutes of agreement and 5 span 20. The row records both moments: when the values came back, and when verification completed.

The Sensitivity card of a DNS monitor in Uptimia, with the restore depth set to 3 matching sweeps, spanning 10 minutes at the fixed 5-minute cadence.Enlarge

The incident closes once every alertable episode has restored or been accepted. A verified restore sends the recovery alert if Alert when records are restored and verified is on in the monitor's Alerting card, and on chat and webhook channels it reads UPTIMIA ALERT (UP): DNS monitor "<name>" restored — records back at baseline (verified) after <duration>. SMS is compressed to UPTIMIA ALERT (UP): DNS restored - verified: <zone> since <date>.

The GeoDNS Trap: Rotation Makes Regional Answers Look Like Changes #

Each sweep queries your nameservers from one probe location, rotating through the monitor's selected locations. For a zone that answers the same everywhere, one probe per sweep is enough, and the rotation adds geographic coverage without multiplying query volume against your servers.

A zone that answers by client region breaks that. The sweep from Amsterdam sees the European address, the next from Singapore sees the Asian one, and the single baseline flips between changed and restored. The symptom is a monitor that alerts and self-restores repeatedly, with a Changes · 30 days count in the double digits.

Best fix first:

  1. Mark the regional records dynamic. Presence-and-type comparison catches a record disappearing everywhere or from most of your locations, and ignores which pool member you were handed. An empty answer from one region is checked against your other locations first, so it does not open an episode on its own.
  2. Narrow the locations to one region. On the Locations tab, switch to Custom selection and pick probes in a single region. A DNS monitor needs at least 6 DNS-capable probes, and a smaller selection is refused. Uptimia intersects a saved profile built for the general fleet with the DNS-capable fleet first. The editor then says what it dropped: N of M probes in this profile are DNS-capable — applied those. If that intersection is empty, or leaves fewer than 6, the save is refused rather than silently widened. Check what the profile resolves to first (see Location Profiles).

Snoozing Change Alerts During Planned Work #

Before a migration, snooze the change alerts rather than pausing the monitor. Pausing stops the sweeps, so you lose the record of what happened, while snoozing keeps sweeping and keeps recording. For the other knobs that quiet a noisy account, see Getting Fewer, Better Alerts.

On the Sensitivity card, set Planning DNS work? Snooze change alerts for to 4 hours, 24 hours or 7 days. The default is off. While it runs:

  • Changes are still detected, recorded as episodes, shown on the Watched records table with their diffs, and counted in the sweep log. Only the paging is held.
  • Health alerts are unaffected. An unreachable nameserver, serial drift or a delegation mismatch still pages at full severity.
  • An incident already open before you snoozed stays open. The snooze stops new change episodes from paging; it closes nothing.

Editing a monitor mid-snooze shows a disabled on — 3 h 20 min left entry in the select. Choosing off ends the snooze on save; choosing a window restarts it from now.

Where DNS Changes Show Up #

SurfaceWhat you see
Watched records cardThe live diff, the state per record, and the row actions
Sweep log cardThe last 10 sweeps: time, result badge (at baseline, N records differ in red, or amber when only Warning records differ, health issue, partial), probe location, servers answered and queries. A row opens the full per-nameserver transcript, and Raw answers in the page header shows the last sweep in full
Logs in the sidebarThe account-wide Monitoring Logs page carries DNS sweeps as their own type. See Monitoring Logs
Incident detail pageThe sweep's reason sentence as the headline error, the status, Type, Started, Ended, Duration, the Detecting checkpoints list and the alerts that were sent
Alert email (root-cause block)The zone, the discovered nameservers, the sweep interval, Detected ("Between two sweeps (window ≤ 5 minutes)"), the active change and health counts, and Restore verification
ExportsThe incident's Export menu: Download PDF, Download HTML, and (for Owner, Admin and Editor seats) Create public link

Only Critical and Warning episodes reach the incident record. An info episode, such as a TTL-only delta, a group set to Info only or a change on a disabled group, lives on the Watched records table and never appears in Incidents.

Who Can Accept or Mark Dynamic #

RoleSees the change and the diffCan Accept / Mark dynamic
Owner✓✓
Admin✓✓
Editor✓✓
Read-only✓—
Accounting / Billing——

The buttons are hidden from Read-only and Accounting / Billing seats, and the same rule applies to a script: an API key created by a team member carries that member's role, so a Read-only member's key cannot accept either. Only an owner-authority key skips the role check.

Was this helpful?

Your answer helps the team improve this article.

Still need help?

Uptimia Bot answers right away and passes you to the team when it can't.

Source: https://help.uptimia.com/help/a/dns-change-detection-baselines-acknowledge