Skip to main content

1.7.3

Released 15 Nov 2026

Before you upgrade a multi-node installation

Upgrade every node before you create a Targets discovery source — it is refused while an older node is running. Do not bring an older node back afterwards: it does not know these sources, and it can remove certificates found only on servers from the inventory until the next scan finds them again. Rolling back to 1.7.2 is not supported once a Targets source exists.

A GlobalSign Atlas connection can be created only once every node runs 1.7.3; rolling back is not supported once GlobalSign certificates exist.

A node still on an earlier release does not know the A10 Thunder target: a distribution run or health check it picks up fails with "module type a10 not found". Finish the upgrade before creating A10 targets.

New​

  • Server certificate scan. A new discovery source type, Targets, reads the certificates stored on the SSH (Linux) and WinRM, IIS and Exchange (Windows) targets and groups you pick — files on every local disk and the Windows certificate stores — with each target's own credential. Root and intermediate CA certificates are listed apart under Inventory → CA certificates, without findings, risk or expiry alerts.
  • Private keys stay on the server. The scan decides on the server, by content, what may leave it: complete certificates only. Private keys, PKCS#12 files (even renamed ones) and anything unrecognised are only counted, never sent.
  • Where each certificate is. The source shows every file and store entry it found and each server's result, including folders the login could not read; a certificate's page lists the servers it was found on, and the inventory filters on Found on: Files / Windows stores.
  • Choosing servers out of hundreds. Servers are chosen in a side panel with search by name or address, type filters, Selected only and Select all; targets and groups you may not choose are greyed out with the reason.
  • Admin on every server it reads. Saving such a source needs the admin role on every chosen target, group and group member, while running it stays an operator action. Each run reads only the servers the admin who last saved it may manage.
  • Several DNS servers in one RFC 2136 credential. List a zone's DNS servers when you do not know which is the primary: CertAutoPilot writes each record on the one that applies updates and never on two servers. Every node must run 1.7.3 before such a credential is saved.
  • GlobalSign Atlas certificates. Certificates can now be issued from your own GlobalSign Atlas account under Settings → Certificate Authorities → CA Connectors. CertAutoPilot validates each domain on the account through its zone first and requests the certificate only after GlobalSign accepts it, so a validation that fails uses no quota.
  • The mTLS client certificate is made in CertAutoPilot. Generate CSR on the connection makes the key, keeps it encrypted and never shows it; paste the certificate the Atlas portal issues for the CSR and the connection switches to it.
  • Prove a GlobalSign domain yourself. Instead of letting CertAutoPilot write the validation record into your zone, you can publish the record yourself, serve a file on the domain's web server, or send an approval e-mail from the certificate page; nothing is bought until GlobalSign accepts the proof.
  • Stop a GlobalSign domain validation. Stop validation on the certificate's order card ends a domain validation that has ordered nothing, and such a certificate can now be deleted while it waits.
  • Check a GlobalSign validation now. Check now on the order card asks GlobalSign at once whether the domain is validated, instead of waiting up to 15 minutes for the next scheduled check; it buys nothing by itself.
  • Your GlobalSign account's rules, not fixed ones. Names per certificate, allowed name suffixes and key types (RSA or ECDSA) come from your Atlas account's validation policy, read live, and the wizard shows them. Wildcards are decided by GlobalSign; a refusal comes before anything is issued.
  • GlobalSign names in several zones. One GlobalSign certificate can carry names that live in different DNS zones; each validation record is written and removed through its own zone. A name the account has already validated needs no zone.
  • Rekey a GlobalSign certificate. Rekey asks GlobalSign for a copy of an active certificate with a new key, without domain validation; the current certificate stays valid. A popover beside the buttons explains how Renew, Re-issue and Rekey differ.
  • Accounts that require a signed public key. When an Atlas account's policy asks for the public key with a signature, CertAutoPilot sends it signed by the certificate key, which stays in CertAutoPilot.
  • Discard an unanswered GlobalSign request. When GlobalSign never answered a certificate request and CertAutoPilot cannot tell whether a certificate was issued, the order card says so and links to the account's issued certificates; a project admin can discard the request, after which the next Re-issue asks GlobalSign anew. Renew, Re-issue and bulk Renew on GlobalSign certificates say beforehand that they use a certificate from the quota.
  • Revoke GlobalSign certificates. Revoke on a GlobalSign Atlas certificate, or on an earlier version in its history, revokes it at GlobalSign.
  • The Atlas account at a glance. The connection's page shows the account's issuance policy, its validated domains and every certificate the account issued — also those issued outside CertAutoPilot — read live; a project admin can remove a domain validation from the account. The issue wizard offers only the key types the policy allows and says whether the domain is already validated there.
  • Add targets in bulk. Add in bulk on the Targets page creates up to 500 SSH, WinRM or IIS targets from a pasted list of servers — a name and an address per line — with every other setting filled in once. A preview shows what each line would do before anything is created, and an operator who needs approval raises one request for the whole list. With a group chosen, servers on the list that are targets already join the group too, so running a list again finishes one whose group add failed.
  • A10 Networks Thunder ADC. A new distribution target puts the certificate, its key and its chain on an A10 Thunder as one file and replaces that same file in place on every renewal, so the client-ssl templates and virtual ports that use it keep serving without dropping connections. It can name the certificate in client-ssl templates and SNI entries on the first deploy, works per partition, and offers live pickers and a deploy preview.
  • Renamed IIS sites are recognised. When a site a target names was renamed in IIS, deploy, dry run, preview and health check stop with IIS_SITE_RENAMED and name the new name, and Update to … opens the target with it filled in.
  • HSMs in FIPS mode. With the pkcs11 KEK provider, an HSM that generates the AES-GCM IV itself and refuses one from the application — a Thales Luna partition in FIPS mode — now works. CertAutoPilot finds this out when it opens the token and nothing is configured for it; certautopilot kek status shows IV source: hsm (or host when CertAutoPilot supplies the IV), and kek pkcs11-init prints which one its probe used. use_gcm_iv_from_hsm: true now means "only an IV from the HSM"; false stays the default. AWS CloudHSM remains unsupported. See supported HSMs.

Improved​

  • HSM connection recovers on its own. With the pkcs11 KEK provider, a lost HSM session, logout or library restart is reconnected automatically, and each node checks the HSM every minute: certautopilot kek verify shows every node's HSM state and the certautopilot_kek_healthy metric reports it. A refused PIN is not retried until a restart, so the token's user PIN is never locked by CertAutoPilot.
  • Each KEK version carries a key check value. kek pkcs11-init and kek reinstate refuse a token that holds a different key under the expected label, and a node started with the wrong key (HSM or env variable) refuses to start instead of writing data the other nodes cannot read.
  • A refused login is held across the cluster. For FortiGate, FortiWeb and A10 targets, a credential the device refused is not sent to that device again for about 70 seconds by any node of the installation — not only by the one that was refused — so several nodes no longer add up to the device's login lockout. A node the device accepted the credential from in the last 10 minutes is not held, and entering the secret again is tried at once.
  • One zone screen for every issuer. A CA connector certificate now picks its zone on the same screen as an ACME certificate: the zone that covers the name is suggested and can be changed.
  • Choosing targets on a module and a group. A module's targets and groups, a new group's first targets and Add Targets on a group are now chosen in the same side panel as a server scan's servers — search by name or address, Selected only, Select all — instead of a tag list; a module shows how many targets it reaches in total.
  • Finding a target by its address. The target search — on the Targets page and in the API — also matches the address a target dials (host name or IP), and the API lists take several types at once and a list of ids.
  • Sturdier against a misbehaving server. On every Windows connection, a reply that is not a WinRM answer, a reply sent a byte at a time and a TLS handshake that never completes now end the request within its time limit — such a server could stop or exhaust CertAutoPilot before. SSH deploy commands keep their output within bounds too.
  • Server scans read less and keep more. A binary certificate file sends only its certificates and each file is read once, at most 1 MiB; a scan that does not know a certificate's chain no longer erases the chain a network scan recorded.
  • Server scans share a memory limit. One worker process reads at most 18 servers at once and keeps their answers within 1 GiB, whatever the number of sources and runs; both are settings (discovery.target_scan_max_parallel_hosts, discovery.target_scan_memory_mb). Servers holding tens of thousands of chained certificates keep their chains, and the targets list pages large projects without sorting every target.

Changed behaviour​

  • IIS: a renamed excluded site is caught. With every site selected, an excluded site that was renamed in IIS now stops the run with IIS_SITE_RENAMED before anything is imported, instead of being swept up; an excluded site that no longer exists only raises the warning IIS_EXCLUDED_SITE_NOT_FOUND.
  • Windows: output limit for CertAutoPilot's own commands. A command CertAutoPilot runs itself (file transfer, checks) stops at 64 MiB of output with WINRM_OUTPUT_LIMIT; your ActionSet commands still run to their end.
  • Windows: a non-WinRM answer is not retried. For IIS and WinRM targets, an HTTP 200 that is not a WinRM reply (another web server or proxy on the port) now fails at once instead of being retried; Exchange still retries it.
  • IIS: a missing site fails with the right code. A single-site deploy whose site does not exist now reports IIS_SITE_NOT_FOUND (was IIS_BINDING_NOT_FOUND); the health check reports a renamed site as unhealthy and a missing one as a warning.
  • Server scans of IIS and Exchange targets use the target's timeouts. The connect and command timeouts set on the target now apply to its server certificate scan as they apply to its deploy; a command timeout shorter than the scan's Time per server ends the scan there.
  • Malformed text in a request is refused. A query or path that is not valid UTF-8 now answers 400 INVALID_QUERY_ENCODING, and a search longer than 200 characters is cut to 200 instead of being refused or sent to the database whole.
  • Unknown target types are refused. Filtering the target or target-group list by a type that does not exist answers 400 instead of an empty list.
  • Discovery source edits name what does not apply. Changing a field that does not apply to the source's type (for example a network setting on a Targets source) answers 422; DNS zone transfer and DNS provider record errors on save are 422 now (were 400). Empty lists of a Targets source come back as [].
  • RFC 2136 server addresses are checked. A DNS server entry must be a host name or address with an optional port; link-local and cloud-metadata addresses are refused when saved, and a credential listing several servers also refuses them when it connects (a single-server credential connects as before).
  • A delete waits for its dependency check. When what depends on a resource cannot be checked in time, the delete answers 503 DEPENDENCY_CHECK_UNAVAILABLE with nothing deleted; retry it. An approved delete that meets this can be retried too.
  • Server scan locations are kept for 180 days. A file or store entry a scan no longer finds stays listed as gone for 180 days, then is removed.
  • pkcs11 KEK provider: new format for wrapped keys. A key wrapped by the pkcs11 provider now records the length of its IV, so one written by 1.7.2 or earlier cannot be read by 1.7.3. An installation on the pkcs11 provider must not be upgraded in place: install 1.7.3 anew and re-import the data, as for changing the KEK provider. Installations on the env provider are not affected.

Fixed​

  • "None" in the overrides drawer. FortiGate placements set to none over the API were turned back into "the target's placements" when the drawer was saved; the drawer now offers None beside the target's value and a value set here, for FortiGate placements and A10 templates and SNI entries.

  • An HSM that refuses an encryption no longer stops the process. When the HSM refused to encrypt (a wrong mechanism parameter, a lost session), the backend or the kek command ended with a crash trace; it now reports the HSM's error.

  • FortiGate deploy after a new intermediate. A deployment that imports a new intermediate no longer fails when FortiGate resets the management connection a few seconds later: reads and value-setting writes are sent once more on a new connection.

  • IIS site names with an apostrophe or brackets. A site such as Bob's Shop broke preview, dry run and deploy for that site and every "all sites" target on the same server, and a site name with brackets skipped its sub-applications' pools when recycling; any site name now works.