RFC 2136 (Dynamic DNS)
Use RFC 2136 dynamic update to let CertAutoPilot add and remove the _acme-challenge TXT record on a DNS server you host yourself — BIND, PowerDNS, Knot, or a Windows DNS Server — instead of a cloud DNS API.
This is the answer to "my authoritative DNS is my own server; how do you write the challenge record?" You enable dynamic updates on the zone and point CertAutoPilot at the server.
How it works
For every DNS-01 challenge the worker sends a standard DNS UPDATE message (RFC 2136) to your DNS server: Present adds TXT _acme-challenge.<name>, the CA validates it, then CleanUp deletes it. Fully automatic on every renewal — no manual DNS edits.
Three details shape how you set the server up:
- The zone is found on your server. Before each update CertAutoPilot asks the configured server for the SOA of the challenge name, walking up until it finds the zone. The server must answer queries from the CertAutoPilot host (
allow-query), and the zone must be hosted there. - Adding a record replaces every TXT record at that name.
Presentfirst deletes the whole TXT set at_acme-challenge.<name>, then adds its own;CleanUpdeletes only its own value. Do not let another ACME client use the same challenge names on this server. CertAutoPilot itself never lets two of its certificates hold the same challenge name at once: a second certificate waits until the first has finished validating. The hold is renewed for as long as the first certificate is still validating, and is freed on its own if that worker stops. The waiting certificate keeps its worker busy, for at most the propagation timeout plus 11 minutes, then fails and is retried like any other failed issuance. On a fleet where some nodes still run a version before 1.5.54, those nodes do not take part in this, so upgrade every node. - Names are validated one at a time. Because of the replacement above, the names of one certificate are presented, validated and cleaned up in turn, with a 60-second pause between names. A certificate for
example.comand*.example.comtakes about a minute longer than one name; ten names take roughly ten minutes more.
Choose an authentication mode
Pick the Auth mode in the credential form; fill only that mode's fields.
| Mode | Fields | Use when | Security |
|---|---|---|---|
| TSIG (shared key) | nameserver + tsig_key + tsig_secret (+ algorithm) | BIND / PowerDNS / Knot and other RFC 2845 servers | Good — HMAC-signed |
| GSS-TSIG | nameserver + the tsig_gss_* fields | Windows AD-integrated zone ("Secure only") | Best on Windows — Kerberos, ties to an AD account |
| Nonsecure | nameserver only | A zone explicitly set to accept unsigned updates | Weak — firewall + a dedicated zone |
Windows DNS Server does not implement RFC 2845 shared-key TSIG for dynamic updates. On Windows use GSS-TSIG (for "Secure only" zones) or Nonsecure. Plain TSIG is for BIND/PowerDNS/etc. If you'd rather not set up Kerberos on the CertAutoPilot host, use the Windows DNS (WinRM) provider instead — it writes the record over WinRM/PowerShell with no TSIG or Kerberos.
What is TSIG / GSS-TSIG?
TSIG (RFC 2845) authenticates a DNS UPDATE by signing it with a shared secret key (an HMAC); the server only accepts updates carrying a valid signature. This is what BIND and most non-Windows servers use.
GSS-TSIG (RFC 3645) is the same idea but the key is negotiated via Kerberos instead of a pre-shared secret — this is what Windows calls "secure dynamic update": the client authenticates as an Active Directory account and the DNS server checks its permission. No static secret to distribute.
Nonsecure means no signature — the server accepts any well-formed UPDATE. Only safe on an isolated/firewalled network.
TSIG setup, step by step (BIND 9.11+)
This walkthrough is written for BIND 9.16 on RHEL / Rocky / Alma (package bind, the named service) — the named.conf parts apply to any BIND 9.11+ and, with the obvious file-path changes, to Debian/Ubuntu (/etc/bind/). BIND 9.11 (RHEL 8) spells type primary, type secondary and primaries as type master, type slave and masters; everything else is the same. CertAutoPilot's end-to-end test runs this setup on BIND 9.11.36 (RHEL 8), 9.16.23 (RHEL 9) and 9.18 (Ubuntu 24.04). PowerDNS and Knot use the same TSIG key format; consult their docs for the zone-side setting.
Throughout, replace example.com with your zone and 10.0.0.5 with your BIND server's IP. Every command runs on the BIND server as root unless stated otherwise.
Step 1 — Generate the TSIG key
tsig-keygen -a hmac-sha256 acme-update. > /etc/named/acme-update.key
chown root:named /etc/named/acme-update.key
chmod 640 /etc/named/acme-update.key
cat /etc/named/acme-update.key
The file looks like this — you will need the three highlighted values in Step 7 (case and the trailing dot of the key name do not matter to CertAutoPilot):
key "acme-update." { ← key NAME (tsig_key)
algorithm hmac-sha256; ← ALGORITHM (tsig_algorithm)
secret "K2b9Qx…/Zt4c="; ← SECRET (tsig_secret)
};
Use hmac-sha256 (or -sha512). Do not use hmac-md5 — CertAutoPilot rejects it. -sha1 works but is legacy.
Step 2 — Load the key in named.conf
Add one line at the top level of /etc/named.conf (outside options { } and outside any zone { } block):
include "/etc/named/acme-update.key";
Step 3 — Move the zone file to the dynamic directory
On Debian and Ubuntu (package bind9) the zone files of your own zones live in /var/lib/bind/, which bind can already write — keep them there, nothing to move. The other steps are the same with Debian paths: the key in /etc/bind/acme-update.key (chown root:bind), the include line in /etc/bind/named.conf, and the zone blocks in /etc/bind/named.conf.local. Verified on Ubuntu 24.04 (BIND 9.18).
A zone that accepts dynamic updates must live in a directory named can write to, because BIND keeps a journal (<zone>.jnl) next to the zone file. On RHEL-family systems that directory is /var/named/dynamic/ (already owned by named and SELinux-labelled named_cache_t). A zone file left under /var/named/ gives SERVFAIL on every update and journal open failed: permission denied in the log.
mv /var/named/example.com.zone /var/named/dynamic/example.com.zone
chown named:named /var/named/dynamic/example.com.zone
restorecon -Rv /var/named/dynamic
Step 4 — Allow the key to write challenge records
Edit the zone's block in /etc/named.conf (or the file it is included from): point file at the new location and add an update-policy.
Before:
zone "example.com" IN {
type primary; // "master" on older configs — same thing
file "example.com.zone";
};
After:
zone "example.com" IN {
type primary;
file "dynamic/example.com.zone"; // relative to options { directory "/var/named"; }
update-policy {
grant acme-update. zonesub TXT CNAME;
};
};
What the one grant line means: the key acme-update. may add/delete TXT and CNAME records only, at any name inside this zone, and nothing else — it cannot touch A, MX, NS or the zone apex's other records. TXT covers every ACME DNS-01 challenge CertAutoPilot can request from this zone: example.com, *.example.com, app.example.com, deep.sub.example.com (the challenge names are _acme-challenge.example.com, _acme-challenge.app.example.com, …). CNAME is needed only for a CA connector whose vendor validates with a CNAME: a commercial CA's domain-validation record is a CNAME or a TXT, chosen by the vendor (GlobalSign Atlas uses a TXT at _globalsign-domain-verification.<domain>, so TXT is enough for it). If you will never issue through such a connector from this zone, grant acme-update. zonesub TXT; is enough — a CNAME write is then REFUSED, which the credential test's opt-in DCV leg shows you up front.
Several zones on the same server need the update-policy block in each zone's block. One key and one CertAutoPilot credential then serve all of them; in CertAutoPilot each zone is listed under Settings → Zones (Step 8), and one zone entry can list several domains.
If you want a tighter rule that lists the exact names instead:
update-policy {
grant acme-update. name _acme-challenge.example.com. TXT; // example.com AND *.example.com share this name
grant acme-update. name _acme-challenge.app.example.com. TXT; // one line per additional hostname
grant acme-update. name _acme-challenge._acme-challenge-test.example.com. TXT; // the probe record the credential test with a domain (Step 7) writes
};
The last line matters: the credential test does not write to the real challenge name (that could clobber an issuance in flight) but to _acme-challenge._acme-challenge-test.<the domain you type in the test>. With a name-only policy that omits it, the test is REFUSED while real issuance works — it looks like a broken key when it is not. With the tight policy, type the zone apex (example.com) into the test field so the probe lands on the granted name. The zonesub form above covers everything. A name-only policy also has to list a CA connector's validation names (GlobalSign Atlas: _globalsign-domain-verification.<domain>); a vendor that names its record with a fresh per-order token cannot be served by a name-only policy at all — use zonesub for a zone that issues through one.
The trailing type list in every grant line is the safety net for the rest of your zone. CertAutoPilot only ever writes TXT records at _acme-challenge names (adding one replaces the TXT set at that name, deleting names the exact record — name, type and value) and, for a CA connector, the vendor's validation record at the name the vendor gives, but the zone should not have to trust that: with the types restricted, the key cannot change an A, MX or NS even if the client were compromised. Do not write grant acme-update. zonesub ANY; or omit the type — both give the key full write access to the zone — and do not use allow-update { any; };, which is the "Nonsecure" mode and lets anyone on the network rewrite it.
Step 5 — Check the config and reload
named-checkconf
named-checkzone example.com /var/named/dynamic/example.com.zone
rndc reconfig # or: systemctl reload named
Make sure the CertAutoPilot host can reach the server on both UDP and TCP port 53. TSIG and unsigned updates go over UDP, GSS-TSIG updates over TCP, and the zone lookup retries over TCP when an answer is truncated. The server must also answer queries from the CertAutoPilot host: if the zone or options block restricts allow-query, add the CertAutoPilot host to it, otherwise the zone lookup fails with could not find zone.
firewall-cmd --permanent --add-service=dns && firewall-cmd --reload
Step 6 — Prove it works with nsupdate before touching CertAutoPilot
Copy /etc/named/acme-update.key to any Linux box with bind-utils (the CertAutoPilot host is a good choice — it proves the network path too) and run:
nsupdate -k acme-update.key <<'EOF'
server 10.0.0.5
zone example.com
update add _acme-challenge.example.com. 60 TXT "tsig-test"
send
EOF
dig @10.0.0.5 _acme-challenge.example.com TXT +short # expect: "tsig-test"
nsupdate -k acme-update.key <<'EOF'
server 10.0.0.5
zone example.com
update delete _acme-challenge.example.com. TXT
send
EOF
nsupdate printing nothing means success. If it prints an error, fix it here — CertAutoPilot will hit the same one:
nsupdate says | Cause | Fix |
|---|---|---|
update failed: REFUSED | No update-policy on the zone, the key is not in it, or the wrong zone block was edited | Step 4 |
update failed: NOTAUTH / BADSIG / tsig verify failure | Key name, secret or algorithm does not match the server's key file | Step 1–2 (nsupdate needs the key name exactly as in the file, trailing dot included) |
update failed: SERVFAIL | named cannot write the journal | Step 3 (directory ownership / SELinux) |
; Communication with 10.0.0.5#53 failed: timed out | Firewall or wrong IP/port | Step 5 |
On the server, journalctl -u named -f shows the update as updating zone 'example.com/IN': adding an RR at '_acme-challenge.example.com' TXT.
Then remove the copied key file from the test box: shred -u acme-update.key.
Step 7 — Add the credential in CertAutoPilot
Settings → DNS Credentials → Add Credential, give it a Name, provider RFC 2136 (Dynamic DNS):
| Form field | Value | From |
|---|---|---|
| Nameserver | 10.0.0.5:53 | Your BIND server IP (port 53 unless you changed it) |
| Other servers of the zone (optional) | 10.0.0.6, 10.0.0.7:53 | The zone's other DNS servers, when you do not know which one is the primary — see Several servers in one credential |
| Auth mode | TSIG | — |
| TSIG Key Name | acme-update. | The key "…" line (the trailing dot is optional here) |
| TSIG Secret | K2b9Qx…/Zt4c= | The secret "…" line, without the quotes |
| TSIG Algorithm | HMAC-SHA256 | The algorithm …; line — must match the key |
Press OK to save it. Then press Test on the credential's row: the Test window takes an optional domain. Left empty, for TSIG it only checks the form (the mode is proved by an actual update), so type a name in the zone and press Run Test — that sends a real signed UPDATE (add, then delete) through the same path an issuance uses, to the probe name _acme-challenge._acme-challenge-test.<what you typed>. If it fails, the error is one of the four rows in Step 6; a wrong secret shows up as dns: bad authent (BIND logs tsig verify failure (BADSIG)).
This exact setup — BIND 9.16.23 as shipped by RHEL 9, this key, this zonesub policy — is what CertAutoPilot's own end-to-end test issues a certificate against, so the steps above are verified, not transcribed.
Step 8 — Add the zone in CertAutoPilot
The credential says how to write; a zone says for which names. Every name on a certificate must fall under a zone in CertAutoPilot, or the issue form stops at that name.
Settings → Zones → Add Zone:
| Form field | Value |
|---|---|
| Domains | The zones this BIND serves that you will issue from, one per line (e.g. example.com). One entry can list several domains of the same server. |
| Name | Optional label |
| DNS Credential | The RFC 2136 credential from Step 7 |
Press Add. Certificates for example.com, *.example.com or app.example.com now validate through your BIND automatically, on every renewal too.
Step 9 — If your BIND has secondaries
The CA validates the record against the zone's public name servers, so the TXT record must reach every secondary before validation. See Primary and secondaries below.
Afterwards: editing a dynamic zone by hand
Once a zone is dynamic, never edit the zone file directly while named is running — the journal would overwrite your change. Freeze it first:
rndc freeze example.com # flushes the journal into the zone file
vi /var/named/dynamic/example.com.zone # bump the serial
rndc thaw example.com
Or skip the file entirely and make the change with nsupdate using a key that has broader rights.
More than one BIND server
Only one server of a zone applies a dynamic update: its primary. A secondary refuses it, or passes it on to the primary when it is set up to forward updates. A credential can name just that server (Nameserver), or list the zone's servers and let CertAutoPilot find the one that takes the update.
Several servers in one credential
Put one server in Nameserver and the others in Other servers of the zone (up to 9; host or host:port, port 53 when omitted). For every record CertAutoPilot:
- finds the zone by asking the listed servers for its SOA, skipping a server it cannot reach;
- tries the server the zone's SOA names as primary first, when that name (or its address) is one of the listed servers — the address is asked of the server that answered the SOA (so an internal
ns1.<zone>the public resolvers do not know still counts), then of this host's resolver. The SOA only orders the list; the servers' answers decide; - sends the signed update over TCP (with one server it is UDP, as before), and goes to the next server only when the update certainly did not happen there:
- the connection could not be made (refused, unreachable, the name does not resolve), or
- the server answered REFUSED (a secondary that does not forward) or NOTAUTH (a server that does not hold the zone), or
- the server refused the TSIG key — BADKEY (it does not know the key), BADSIG (its secret differs) or BADTIME (the clocks differ). A server that refuses the TSIG does not process the update, so the key only has to be on the server that applies updates (usually the primary); the secondaries do not need it;
- stops at the first server that answers that it applied the update. A secondary that forwards updates counts: it passes the update to the primary and answers for it.
It never moves on when the update may have been applied: no answer in time, a dropped connection, SERVFAIL (a forwarding secondary answers that when the primary is down, and it cannot say whether the primary applied the update), or an answer without the TSIG signature. The attempt then fails with the servers and their answers in the error, and the job's normal retry sends the update again in the same order. Sending one update twice to the same server is harmless, so a record is never left on two servers without CertAutoPilot knowing. When every listed server refuses the TSIG key, the error says so — check the key on the primary and the CertAutoPilot host's clock.
Removing the record after validation follows the same rule, so it reaches the server that holds the record. Test with a domain writes and removes its probe record the same way and names the server that applied it.
What a list does not change:
- A primary that is down cannot be replaced by a list. Only the primary applies updates; a forwarding secondary answers SERVFAIL while it is down. A list helps when the primary is not known, when the primary role moves between servers, when some listed servers are unreachable from CertAutoPilot, and with servers that all accept updates.
- The propagation check is unchanged. Before the CA is asked, every name server in the zone's NS set must return the record (see Primary and secondaries).
- GSS-TSIG (Windows secure updates) keeps one server; use the Windows DNS (WinRM) provider for several domain controllers.
- Every node must run 1.7.3 or later before a list can be saved: an older node would write to the first server only.
Changing the list, like changing the nameserver, needs the TSIG secret entered again — unless you are the person who entered the secret, or an organization owner.
What a server entry may be. A host name or an IP address (IPv6 in brackets: [fd00::53], [fd00::53]:5353), with an optional port 1–65535 — no scheme (http://), path, user, query, spaces or control characters, at most 270 characters. A link-local or cloud-metadata address (169.254.0.0/16, fe80::/10, 169.254.169.254, 100.100.100.200), 0.0.0.0 / :: and multicast addresses are refused when saved (400). With Other servers of the zone set, the same addresses are also refused when an update is sent — whatever the name resolves to then — and the next listed server is tried. Private and loopback addresses are allowed: that is where internal DNS lives. A server saved before this rule keeps working until it is changed.
The end-to-end test runs this against BIND 9.16.23 and 9.11.36 with a primary, a secondary that refuses updates, one that forwards them and one that does not know the TSIG key: a list given as [refusing secondary, server that is down, primary] issues a certificate through the primary and the record reaches the secondaries; the secondary without the key answers BADKEY and the record is written on the primary; a server that takes the connection and never answers stops the update there. (A server that does not know the key answers REFUSED with the TSIG error BADKEY — measured on BIND 9.11, 9.16 and 9.18; CertAutoPilot reads the TSIG error, not the reply code.)
Zones on different servers
Create one RFC 2136 credential per server, each with that server's address. The same TSIG key may be installed on all of them, or each can have its own. Then, in Settings → Zones, bind each zone to the credential of the server that hosts it.
Primary and secondaries
- Send updates to the primary — or list the servers and let CertAutoPilot find it (above). A secondary refuses updates unless it forwards them. To keep a secondary as the address CertAutoPilot talks to, add
allow-update-forwarding { key acme-update.; };to the zone on the secondary; it then passes the signed update on to the primary. - Every name server in the zone's NS set is checked. With the default strict pre-check, CertAutoPilot asks each name server listed in the zone's NS records, directly on port 53, and waits until all of them return the new record. So:
- the CertAutoPilot host must reach every listed name server on UDP and TCP 53 — one it cannot reach makes every issuance wait for the whole propagation timeout and fail;
- zone transfers must reach every secondary within the propagation timeout (Settings → General, 30–600 seconds). Keep
notify yes;(the default) on the primary, and addalso-notify { <secondary IPs>; };for secondaries that are not listed as NS records.
- A hidden primary is fine. A primary that is not in the NS set is written to but not checked; the listed secondaries are.
- If some listed name servers are unreachable from CertAutoPilot by design, turn on Allow Partial Propagation and set DNS Resolvers to resolvers that see the zone's public answers. One of them returning the record is then enough. See Propagation settings.
Views (split horizon)
With view blocks, an update lands in the view that matches the CertAutoPilot host's address or key, while the CA reads the external view. If those are different views with separate copies of the zone, the CA never sees the record. Either:
- route the key to the external view by putting
match-keys { acme-update.; };first in that view's match list, or - keep a single copy of the zone and share it with
in-viewin the other views.
When CertAutoPilot's own resolvers only see the internal view, also turn on Skip Propagation Pre-check, because the pre-check cannot see what the CA sees.
Several primaries without zone transfer
If the same zone is loaded on several independent servers (for example anycast nodes fed by configuration management), an update reaches only the one server CertAutoPilot writes to, and the CA may ask another. This layout is not supported for the zone itself — a list of servers does not change that, since the update is applied on one of them only. Use a primary with secondaries, or delegate the challenge names to a single server as below.
Delegate challenge names to one server (recommended for complex setups)
Point each _acme-challenge name at a small zone that lives on one dedicated BIND server, and let CertAutoPilot write only there. Your production zones then never accept updates at all.
-
Create a zone such as
acme.example.neton the dedicated server, delegate it publicly, and give it the TSIG setup above (grant acme-update. zonesub TXT;). -
In each production zone, add a CNAME per certificate name — one covers a name and its wildcard:
_acme-challenge.example.com. CNAME example-com.acme.example.net._acme-challenge.app.example.com. CNAME app-example-com.acme.example.net. -
Create one RFC 2136 credential pointing at the dedicated server, and bind the production zones to it in Settings → Zones.
CertAutoPilot follows the CNAME before writing, using Settings → General → DNS Resolvers (the public set when empty), so those resolvers must see the CNAMEs. It then writes the TXT record into acme.example.net, and the CA follows the same CNAME. For the credential test, type the delegated zone (acme.example.net) so the probe record lands there.
GSS-TSIG setup (Windows secure dynamic update)
For a Windows AD-integrated "Secure only" zone. Uses an AD account's username + password — the CertAutoPilot host does not need to be domain-joined.
-
In DNS Manager keep the zone at Secure only. Grant an AD service account (e.g.
svc-acme) Write on the zone (zone → Security tab), or delegate a_acme-challenge.<domain>sub-zone and grant it only there. -
In the credential form choose Auth mode = GSS-TSIG and enter the Realm (AD domain, UPPERCASE), Username, Password.
-
On the Linux host running CertAutoPilot, provide a minimal
/etc/krb5.conf(Kerberos is pure-Go here — no libgssapi/cgo needed):[libdefaults]default_realm = CORP.EXAMPLE.COMdns_lookup_kdc = true[realms]CORP.EXAMPLE.COM = { kdc = dc1.corp.example.com }Requirements: the host can reach a KDC (domain controller) on port 88, clock skew < 5 min with the DC, and the account can update the zone. No domain-join.
KRB5_CONFIGoverrides the location — and when it is set but points at a file that cannot be read, that is an error rather than a fall back to/etc/krb5.conf. -
Press Test on the credential. With no test domain given it checks the Kerberos configuration this mode depends on: that the file exists and parses, and that the realm you entered either has a
[realms]entry or can be discovered throughdns_lookup_kdc = true. It contacts no KDC and writes no record. A missing or realm-lesskrb5.confused to pass this test and fail at the first issuance instead.
If setting up krb5.conf / KDC access on the Linux host is undesirable, the Windows DNS (WinRM) provider achieves the same result over your existing WinRM admin access — and its own Kerberos option needs no host setup at all.
This is the only Kerberos path in CertAutoPilot that reads a file. Every other one — WinRM, IIS, Exchange and the Windows DNS provider — builds its Kerberos configuration in memory and finds the KDC through a DNS SRV lookup, so it needs nothing on the host. GSS-TSIG is different because the negotiation happens inside the ACME library, which offers no way to hand it a configuration; it reads KRB5_CONFIG or /etc/krb5.conf itself.
Enable dynamic updates on a Windows zone
DNS Manager → zone → Properties → General → Dynamic updates:
- Secure only — AD-integrated zones. Use GSS-TSIG.
- Nonsecure and secure — allows unsigned updates. Use Nonsecure mode, and only with the hardening below.
- None — disabled; RFC 2136 cannot work.
PowerShell: Set-DnsServerPrimaryZone -Name "example.com" -DynamicUpdate "Secure" (or "NonsecureAndSecure").
Add the credential
- Settings → DNS Credentials → Add Credential → a Name → provider RFC 2136 (Dynamic DNS).
- Nameserver — the DNS server as
host:port, e.g.10.0.0.5:53. - Auth mode — Nonsecure / TSIG / GSS-TSIG, then fill that mode's fields.
- OK to save, then Test on the credential's row → type a domain in the zone → Run Test (a real add-and-delete of a probe TXT record).
- Settings → Zones → Add Zone → the domains + this credential (Step 8).
Config fields
| Field | Mode | Purpose |
|---|---|---|
nameserver | all | DNS server accepting updates, as host:port. |
auth_mode | all | nonsecure / tsig / gss. Required when updating a credential through the API — it tells CertAutoPilot to clear the other mode's fields. Omitting it leaves stale credentials in the stored config, so a TSIG → nonsecure switch would keep signing. |
tsig_key | TSIG | TSIG key name (e.g. acme-update.). |
tsig_secret | TSIG | Base64 TSIG secret. |
tsig_algorithm | TSIG | Pick to match your key (default hmac-sha256.). |
tsig_gss_realm | GSS-TSIG | Kerberos realm, e.g. CORP.EXAMPLE.COM. |
tsig_gss_username | GSS-TSIG | AD account allowed to update the zone. |
tsig_gss_password | GSS-TSIG | That account's password. |
Secrets (tsig_secret, tsig_gss_password) are encrypted at rest and never returned by the API.
Security hardening
- Prefer TSIG or GSS-TSIG over nonsecure. If you must use nonsecure, firewall port 53 to the worker IPs.
- Delegate the challenge names to a separate zone and grant update rights only there — never on your production zone. See Delegate challenge names to one server.
- Least-privilege: the key / AD account should be able to write only
TXT _acme-challenge.*.
Propagation
The record is written straight onto the authoritative server, so on a single server propagation is near-instant. With secondaries, the default check waits until every name server in the zone's NS set returns the record — see Primary and secondaries and Propagation settings.
Troubleshooting
- "bad signature" / NOTAUTH — TSIG key name/secret/algorithm mismatch (must match the server's key exactly), or a plain-TSIG update sent to a Windows "Secure only" zone (use GSS-TSIG or WinRM).
- "REFUSED" — dynamic updates not enabled on the zone, or the mode doesn't match ("Secure only" needs GSS-TSIG), or the key/account lacks write permission.
- GSS-TSIG can't get a ticket — the host can't reach the KDC (port 88), krb5.conf realm wrong, or clock skew > 5 min. Press Test first: it rules out a missing, unparseable or realm-less
krb5.confwithout contacting anything, leaving only reachability and clock skew. - Timeout — firewall blocking UDP/TCP 53, or the wrong
nameserverhost:port. - "could not find zone" — the server does not answer the CertAutoPilot host's queries (
allow-query), or it does not host the zone the name belongs to. - Issuance waits for the full propagation timeout, then fails with "NS … did not return the expected TXT record" — a secondary has not received the update, or CertAutoPilot cannot reach one of the listed name servers. See Primary and secondaries.
- "another certificate is still validating the DNS-01 record …" — a second certificate for the same name waited the propagation timeout plus 11 minutes and the name was still in use. It is retried like any other failed issuance.
- "locking DNS-01 challenge name …" — CertAutoPilot could not reach its own database to coordinate the challenge name for that whole wait. Check the database connection; the issuance is retried like any other failure.
See also
- Windows DNS (WinRM) — Windows secure zones without Kerberos
- DNS providers overview
- Manual DNS
Video tutorial
Short, hands-on videos showing this feature working for real. English audio, English and Turkish subtitles. All video tutorials →