Skip to content
AdminKit

What is an SOA record, and what do its five timers actually control?

AdminKit explains the SOA record, the one record every DNS zone has exactly one of: it names the zone's primary name server and admin contact alongside five numeric timers — serial, refresh, retry, expire and minimum — that control how secondaries stay in sync and how long a "name doesn't exist" answer is cached before a stale serial causes trouble.

By AdminKit team, Netbase JSC · Updated

Key takeaways

  • Every DNS zone has exactly one SOA record. It names the primary name server and an admin mailbox, then lists five numbers: serial, refresh, retry, expire and minimum.
  • The serial is how secondaries know the zone changed. If it doesn’t go up after an edit, secondaries keep serving the old data.
  • Refresh, retry and expire only matter to secondary name servers. DNS NOTIFY usually triggers updates first, and refresh is the fallback.
  • Minimum no longer means “minimum TTL”. Since RFC 2308 it sets how long resolvers cache “this name doesn’t exist”, capped by the SOA record’s own TTL.
  • Most managed DNS providers set all of this for you. It matters most when you run your own authoritative servers.

What is an SOA record, and why does every zone have exactly one?

An SOA (start of authority) record is the header of a DNS zone, and a zone can’t have zero or several of them. Unlike A records or MX records, RFC 1035 requires exactly one per zone, marking, in the RFC’s own words, “the start of a zone of authority.” Query any domain’s SOA record with AdminKit’s DNS lookup tool (SOA is one of the record types it supports, alongside A, AAAA, MX, TXT, NS, CAA, PTR and CNAME) and you’ll get back two names and five numbers. In zone-file form, with comments, an SOA record looks like this:

example.com.  SOA  ns1.example.com. admin.example.com. (
                2026110100 ; serial
                7200       ; refresh
                3600       ; retry
                1209600    ; expire
                300 )      ; minimum

The two names are the primary (authoritative source of truth) name server and an email address for the zone’s administrator, written with the @ replaced by a . (so admin.example.com. means admin@example.com). The five numbers are what this guide is mostly about — see how to read a DNS lookup result for how SOA fits alongside the other record types AdminKit’s tool returns.

Most lookup tools, dig +short included, print the same seven values on one line, in the order RFC 1035 defines them. Example output (illustrative, documentation names):

$ dig +short example.com SOA
ns1.example.com. hostmaster.example.com. 2026111001 7200 3600 1209600 3600

What does each SOA field mean?

RFC 1035 §3.3.13 defines the two names and then five plain 32-bit values. Serial is a version number; the other four are measured in seconds.

FieldRFC 1035 nameWhat it controlsWho uses it
Primary serverMNAMEThe name server that holds the original copy of the zoneNOTIFY (RFC 1996), tools that look for the primary
Admin mailboxRNAMEContact address, @ written as .People reporting problems
SerialSERIALVersion number of the zoneSecondaries, before each transfer
Refresh / Retry / ExpireREFRESH RETRY EXPIREHow secondaries poll, retry and give upSecondaries only
MinimumMINIMUMTTL for negative answers (since RFC 2308)Recursive resolvers

Field by field:

  • Serial — “the unsigned 32 bit version number of the original copy of the zone.” A secondary name server compares this against its own stored copy before pulling a full zone transfer; a higher serial means the zone changed. Because the field can wrap around at its maximum value, RFC 1982 defines dedicated serial number arithmetic so comparisons stay correct even across that wraparound — plain numeric comparison alone would eventually misjudge a wrapped, smaller-looking serial as older than it is.
  • Refresh — “a 32 bit time interval before the zone should be refreshed.” This is how often a secondary checks the primary’s serial under normal, working conditions.
  • Retry — “a 32 bit time interval that should elapse before a failed refresh should be retried.” This only kicks in once a refresh attempt has already failed to reach the primary, and is typically shorter than refresh so a secondary recovers quickly once the primary is reachable again.
  • Expire — “the upper limit on the time interval that can elapse before the zone is no longer authoritative.” If a secondary can’t reach the primary for longer than expire, it stops answering queries for the zone altogether, on the assumption that its data is too stale to trust.
  • Minimum — originally “the minimum TTL field that should be exported with any RR from this zone,” but RFC 2308 §4 redefined this. See the section on minimum below — this is the field most legacy documentation still describes incorrectly.

RFC 1912 §2.2 adds a practical note on the admin mailbox: if the local part of the address contains a dot (as in John.Smith), it must be escaped as John\.Smith, and the cleaner option is a generic hostmaster alias. It also warns that this address is used “for everything from reporting bad DNS data to reporting security incidents,” so it should reach a real person. Cloudflare’s own explainer on the SOA record walks through the same five fields a working DNS provider actually implements them as, independently confirming what each one does in practice rather than just in the RFC text.

How should the serial number be formatted?

Any format works as long as the number goes up after every change. The most common convention is a date plus a two-digit counter, YYYYMMDDnn. RFC 1912 recommends it and notes it “won’t overflow until the year 4294”, and RIPE-203, the RIPE DNS Working Group’s SOA recommendations, uses the same scheme with the first change of the day numbered 01.

SchemeExampleNotes
Date + counter2026111001Readable as “first change on 10 November 2026”. Allows 100 changes per day.
Plain counter1, 2, 3…Fine for software that bumps it on every save. Says nothing about when.
Unix timestamp1794268800Seconds since 1970 (here, midnight UTC on 10 November 2026). It is a smaller number than a date-based serial, so switching to it means lowering the serial.

RIPE-203 also warns not to rely on the date in a serial for debugging, because after setup many zones leave the number to the software’s auto-increment. RFC 1912 adds two smaller rules: don’t use decimal serials like 1.5, which BIND converts into unexpected integers, and RFC 1982 §7 cautions against setting the serial to zero, because some implementations have treated zero as a special case.

What happens when the serial wraps or goes backwards?

The serial is a 32-bit unsigned number, from 0 to 4294967295. RFC 1982 compares serials on a circle rather than a straight line: serial s2 is “newer” than s1 if it is ahead of it by less than half the circle, 2^31. That is why the largest increment allowed in one step is 2147483647 (2^31 − 1), and why RFC 1982 warns against increasing the serial by more than that within one expire period.

The rule matters most when someone makes a mistake. Lowering a serial makes every secondary ignore further changes, because the primary’s copy now looks older than theirs. The fix uses the wrap rule: increase the serial by up to 2^31 − 1, wait until every secondary has transferred it, then step to the value you want, which is now “ahead” on the circle. A worked example with a mistyped year:

Intended serial:          2026111001
Mistyped serial:          2062111001   (secondaries now hold this)

Step 1: add 2147483647 -> 4209594648   wait until all secondaries have it
Step 2: set              2026111001    ahead of 4209594648 by 2111483649 (< 2^31)

RFC 1982 says to “verify that ALL servers have correctly succeeded in following the primary server’s serial number changes, at each step.” Skipping that check can leave a secondary holding stale data under a serial it believes is newer.

How do refresh, retry and expire work together?

Under normal conditions a secondary asks the primary for its SOA every refresh interval. If the serial is higher, it transfers the zone. If the primary can’t be reached, the secondary tries again every retry interval. If that keeps failing until expire runs out, the secondary stops answering for the zone.

In practice, a primary doesn’t have to wait for a secondary’s own refresh timer to notice a change: DNS NOTIFY, defined in RFC 1996, lets the primary proactively tell every secondary right after an update. On receiving a NOTIFY, the secondary should “behave as though the zone … had reached its REFRESH interval”: it queries the SOA, and if the serial went up it starts a full transfer or an incremental one. That is what triggers most real-world zone transfers rather than secondaries polling on a fixed schedule — refresh is the fallback for when NOTIFY doesn’t arrive, not the primary sync mechanism on a well-configured zone.

Published guidance on the values differs, and each document says why:

TimerRFC 1912 §2.2 (1996)RIPE-203 (1999)Rule of thumb from both
Refresh20 min – 2 h, or 2 – 12 h on slow links; longer with NOTIFY86400 (24 h)Longer is fine when NOTIFY is working
Retry“some fraction of the refresh interval”7200 (2 h)Shorter than refresh
Expire2 – 4 weeks; greater than minimum and retry3600000 (1000 h)Longer than your worst realistic outage
Minimum1 – 5 days (as a default TTL, pre-RFC 2308)172800 (2 days), with SOA TTL 3600See RFC 2308 below

RIPE-203 explains its long expire value directly: “A value of a week or two has proven to be way too short.” Its long refresh assumes NOTIFY is in use, which “reduces the need for frequent SOA checks.”

Why doesn’t “minimum” mean “minimum TTL” anymore?

Because RFC 2308 changed it in 1998: minimum is now the negative caching TTL. RFC 1035 originally overloaded the field with three purposes, including acting as a floor under every record’s TTL in the zone. RFC 2308, also tracked in the IETF Datatracker, formally narrowed that down about a decade later: “the new defined meaning of the SOA minimum field” is how long a resolver is allowed to cache a “this name doesn’t exist” (NXDOMAIN) or “no data of this type” answer before it’s required to ask again. RFC 2308 states the floor meaning “has never in practice been used and is hereby deprecated,” and moves the default-TTL role to the $TTL zone-file directive. It no longer constrains the TTL of records that do exist in the zone; each record’s own TTL governs that, independent of what minimum says.

The exact rule from RFC 2308 §3: an authoritative server includes the SOA in the authority section of a negative answer, with its TTL set to the smaller of the SOA record’s own TTL and the minimum field. Example output (illustrative, documentation names), for a zone with SOA TTL 3600 and minimum 900:

$ dig nosuchhost.example.com A
;; ->>HEADER<<- opcode: QUERY, status: NXDOMAIN, id: 40112
;; AUTHORITY SECTION:
example.com.  900  IN  SOA  ns1.example.com. hostmaster.example.com. 2026111001 7200 3600 1209600 900

The 900 at the start of the authority line is how long a resolver may cache that NXDOMAIN. RFC 2308 §5 notes that negative caching values of “one to three hours have been found to work well” and that values over one day “have been found to be problematic.” A domain with a long minimum value and a typo in a newly-added hostname will have that typo’s “doesn’t exist” answer cached for longer than expected — a subtler version of the same caching delay covered in DNS propagation and TTL explained.

When does a wrong SOA setting actually cause a problem?

Most SOA misconfigurations are invisible until something goes wrong in a specific, narrow way:

MistakeWhat you’ll seeFix
Serial not increased after an editSome name servers return old records indefinitelyIncrease the serial and reload; check every server’s serial
Serial lowered, or set too high by a typoSecondaries ignore every later changeUse the two-step wrap procedure above
Expire too shortA secondary stops answering when a primary outage outlasts itSet expire well above refresh + retry, in weeks
Minimum very highA new hostname still “doesn’t exist” for hoursLower minimum (or SOA TTL) before adding names
MNAME repeats the zone nameAnything that looks up the primary from the SOA reaches the wrong hostName the actual primary; RIPE-203 §4.1 calls the zone name a mistake here unless it also reaches the primary

A serial that didn’t actually increment after a zone edit means secondaries never notice the change happened — the single most common SOA-related support issue. A high minimum extends how long a genuine typo or a freshly added record’s “doesn’t exist” answer sticks around in resolver caches, which looks identical to a propagation delay from the outside — see DNS propagation and TTL explained for how to tell the two apart.

How do you check that all your name servers agree?

Ask each authoritative server directly for the SOA and compare serials. dig +nssearch does this for every server listed in the zone’s NS records. Example output (illustrative, documentation addresses):

$ dig +nssearch example.com
SOA ns1.example.com. hostmaster.example.com. 2026111002 7200 3600 1209600 900 from server 192.0.2.53 in 11 ms.
SOA ns1.example.com. hostmaster.example.com. 2026111001 7200 3600 1209600 900 from server 198.51.100.53 in 24 ms.

Here the second server is one change behind. If that persists past a refresh interval, the secondary isn’t receiving NOTIFY or can’t transfer the zone, which the guide on zone transfers (AXFR) covers. You can only run this against servers you’re allowed to query; AXFR itself should stay restricted to your own secondaries.

Most managed DNS providers compute and update the serial automatically and set sane defaults for the other four fields, so this mostly matters if you run your own authoritative name server, or if a zone transfer between your own primary and secondaries is behaving unexpectedly. Either way, AdminKit’s DNS lookup tool is the fastest way to see what your zone’s SOA record currently publishes, without needing shell access to the name server itself. It queries through a public recursive resolver, so the TTL it shows may be lower than the zone’s configured value while the answer sits in cache. This guide is part of AdminKit’s DNS cluster of DNS record explainers and tools.

This guide is general information, not a security audit of your specific network — see our disclaimer for what AdminKit's guidance does and does not commit to.

Frequently asked questions

What is an SOA record for, in one sentence?

It identifies the zone's primary name server and administrator contact, and carries five timing fields that tell secondary name servers how often to check for updates, how long to keep answering if the primary goes unreachable, and how long a negative ("doesn't exist") answer should be cached. RFC 1035 §3.3.13 defines it as marking "the start of a zone of authority."

Why does the serial number matter, and why isn't it just incremented by 1 each time?

A secondary name server compares its own copy of the serial against the primary's before pulling a zone transfer — a higher number means something changed. Many operators use a date-based format like YYYYMMDDnn (e.g. 2026110100) instead of a plain counter, so the serial is also a human-readable "last changed" timestamp. Because the field wraps at its 32-bit maximum, RFC 1982 defines serial number arithmetic specifically so a secondary compares serials correctly even across that wraparound, rather than treating a wrapped, smaller-looking number as older.

What's the difference between refresh and retry?

Refresh (RFC 1035 §3.3.13) is how often a secondary checks the primary's serial under normal conditions — "a 32 bit time interval before the zone should be refreshed." Retry is how soon it tries again specifically after a refresh attempt failed — "a 32 bit time interval that should elapse before a failed refresh should be retried" — so retry only matters once refresh has already failed to reach the primary, and is typically set shorter than refresh.

Does the minimum field still mean "minimum TTL for every record in the zone"?

No, and this is the most commonly misread field. RFC 1035 originally defined minimum as a floor applied to every record's TTL in the zone, but RFC 2308 §4 formally narrowed that: "the new defined meaning of the SOA minimum field" is the TTL used for negative (NXDOMAIN or no-data) responses — how long a resolver caches "this name doesn't exist" before asking again. Any documentation still describing it as a blanket minimum TTL is describing the pre-1998 definition, not current practice.

Can I edit my own domain's SOA record?

Only indirectly. Your DNS provider or registrar's control panel manages the primary name server and admin-contact fields, and usually recalculates the serial automatically on every change — manual edits to the raw record are uncommon outside of self-hosted authoritative DNS software like BIND or PowerDNS, where an operator edits the zone file directly.

Want AdminKit to watch this for you?

Free checks tell you the state today. Protective DNS and monitoring tell you when it changes.

Huy Nguyen (David), Founder and CEO of Netbase JSC

Talk to an expert

Huy Nguyen (David), Founder & CEO, Netbase JSC

Ask whether this fix applies to your setup, or get a second pair of eyes on your DNS and mail records.