Skip to content

v2.58.0 (certbot gets certificates from CertMate's ACME server, the inventory counts the certificates CertMate manages, and a failed issuance says what to check)

Latest

Choose a tag to compare

@fabriziosalmi fabriziosalmi released this 17 Nov 23:22
11d98bd

v2.58.0 (certbot gets certificates from CertMate's ACME server, the inventory counts the certificates CertMate manages, and a failed issuance says what to check)

A minor release. The API contract moves to 2.68: a failed issuance job carries its error code and a hint (#1495), and the inventory lists the certificates CertMate manages (#1505). Most of it comes from the two audits of v2.56.1 and the product decisions taken on them; a few of their smaller findings are still open. Every change has a test that was run before it and failed.

  • certbot gets a certificate from CertMate's ACME server (#1504, closes #1484). The server sent the leaf alone, and certbot refuses a chain of one certificate: it registered, proved its challenge and finalized, then wrote nothing ("less than 2 certificates in chain"). The download now carries the root after the leaf, which a TLS server does not need and may send. lego, acme.sh and cert-manager, which accepted the leaf alone, get the root too. certbot 5.8.0 now runs end to end in the test suite; the interoperability runs with the other clients have not been repeated.
  • The inventory counts the certificates CertMate manages (#1505). It showed "Managed 0, Total 0" on an instance that managed 14: a managed certificate entered it only when discovery, which is opt-in, probed its host. Each one is now read from disk at startup and after every issuance and renewal, recorded as managed, and a renewal marks the certificate it replaced as superseded. GET /api/inventory lists them (contract 2.68).
  • A failed issuance says what to check (#1495). The dashboard issues in the background, and a failed job carried certbot's output and nothing else, so the hint the API gives (#1460) was never shown. The job now carries the same error_code and hint (contract 2.67), and the dashboard leads with the hint. A CA directory that cannot be reached is named with its URL, and a private CA whose certificate cannot be verified points at its trust bundle, where the hint used to say to check DNS credentials.
  • A backup made from the settings page can restore the instance when backups are encrypted (#1498). Every backup made there was a configuration snapshot without credentials or private keys, while the page said they held them. With CERTMATE_BACKUP_PASSPHRASE set, the page offers a complete, encrypted backup (on by default); without it, it says what its backups hold and how to make complete ones.
  • The first-run wizard can be finished with HTTP-01 (#1503). Its second step could not be passed without a DNS provider. It now asks how names are validated, and HTTP-01 needs no DNS credentials. The form that opens at the end reads the CA list again, which it used to show as "Configure a CA in Settings".
  • Settings asks before leaving with changes nobody saved (#1494). A threshold changed and a click elsewhere lost the change in silence. Each part with its own Save button is tracked; parts that apply at once (users, API keys) are not.
  • Force renew asks first (#1496). It skips the due check and counts against the CA's rate limits, and was one click away. Plain Renew does not ask.
  • The PFX export password can be removed (#1496). A blank field keeps it, as the page now says; a box removes it, and a line says whether one is saved.
  • Smaller fixes (#1496, #1497): the Users tab loads when it is opened, instead of showing local authentication as off for four seconds; a DNS card says "1 account"; a missing account field is named by its label rather than its placeholder; Cache TTL describes the deployment-status cache it sets; the empty dashboard no longer says that authentication is off by default.