cert-manager v1.21.0-beta.0

[!NOTE] ⚠️ This is a pre-release. For testing only!

For full release notes including breaking changes, upgrade notes, major themes and community credits, see the v1.21 release notes.

Changes since v1.21.0-alpha.1

Feature

• Add certificateRequestMaximumBackoffDuration controller configuration option to cap retry backoff time for failed CertificateRequests. Configurable via config file, --certificate-request-maximum-backoff-duration CLI flag, or Helm value config.certificateRequestMaximumBackoffDuration. Defaults to 32 hours for backward compatibility. (#8893, @lunarwhite)
• Add an optional waitInsteadOfSelfCheck field to ACME HTTP01 and DNS01 solvers so cert-manager can skip its own self-check and ask the ACME server to validate after a configured wait. (#8858, @wallrj)
• Add configurable runtimeClassName support for cert-manager components and ACME HTTP01 solver pods. (#8791, @jsoref)
• Added ARI support through the ACMEUseARI feature gate. (#8798, @hjoshi123)
• Added AWS IAM authentication support for Vault issuer, including IRSA (IAM Roles for Service Accounts) and ambient credentials (EC2/ECS). (#8422, @bitloi)
• Adds support for the Modern2026 go-pkcs12 profile and FIPS 140-3. (#8841, @seanorama)
• A new flag --ignore-namespaces was added to the cainjector binary. It can be used to filter out namespaces from being watched for secrets to use for injectables. (#8614, @figaw)
• Disabled client side rate-limiting if AP&F is enabled. (#8757, @hjoshi123)
• Processed annotations cert-manager.io/alt-names, cert-manager.io/ip-sans to Certificates generated from ingress like objects in cert-shim controllers. (#8927, @jabbrwcky)
• When using ACME HTTP-01 with a ListenerSet, setting the annotation acme.cert-manager.io/http01-parentreffallback: "true" causes cert-manager to use the parent Gateway as the solver HTTPRoute parentRef instead of the ListenerSet. This enables TLS-only ListenerSets to rely on a shared Gateway HTTP listener for ACME challenges. (#8749, @apkatsikas)

Bug or Regression

• BREAKING: The Helm chart no longer ships a default Role and RoleBinding granting the cert-manager controller ServiceAccount permission to create tokens for itself (serviceaccounts/token: create). This RBAC was added in v1.16 (#7213) but no documented workflow requires it, and the motivating Route53 docs section was removed in Oct 2024. If you rely on serviceAccountRef.name pointing at the controller ServiceAccount (an undocumented pattern), you must now create your own Role and RoleBinding granting serviceaccounts/token: create on that ServiceAccount, or migrate to one of the documented patterns (IRSA ambient, or a dedicated ServiceAccount with its own RBAC). (#8931, @wallrj-cyberark)
• ACME challenges no longer terminally fail on transient network errors (TLS handshake timeouts, DNS failures, context cancellation) during nonce fetches and authorization waits. The challenge controller returns the error and lets the workqueue retry with backoff. (#8760, @texasich)
• Fix webhook serving certificate not being renewed after system suspend. (#8464, @Peac36)
• Fixed a rare panic in the trigger controller when a Certificate is deleted from the informer cache while a reconcile is in progress (e.g. during namespace teardown). (#8962, @hjoshi123)
• Fixed an integer overflow in renewBeforePercentage calculations that caused Certificates with durations longer than approximately 3 years to be incorrectly rejected by validation or assigned incorrect renewal times. (#8947, @ThatsMrTalbot)
• Fixed potential OOM in DNS-over-HTTPS client by bounding response body read with io.LimitReader (128 KB cap). (#8803, @SebTardif)
• Fixed validation of timezone-prefixed renewal window cron specs without a schedule. (#8813, @immanuwell)
• Harden ACME Challenge and Order resources: reject user-created Challenges without Order ownership, enforce Order spec immutability, and detect pre-placed same-name Challenges with mismatched specs. (#8948, @wallrj-cyberark)
• Remove ACME Challenge create and Order create/patch/update from the cert-manager-edit aggregate ClusterRole to prevent direct manipulation of these internal resources (GHSA-8rvj-mm4h-c258). (#8958, @wallrj-cyberark)
• Update logic to identify and preserve the secret matching nextPrivateKeySecretName. (#8577, @putongyong)
• Vault Issuer webhook validation now rejects ... path segments in spec.vault.path and auth mount path fields, preventing path.Join from silently resolving relative segments before constructing the Vault API request. (#8930, @wallrj-cyberark)

Other (Cleanup or Flake)

• Remove Helm values prometheus.servicemonitor.targetPort, prometheus.servicemonitor.path, and prometheus.podmonitor.path. The metrics path is always /metrics and the target port is always http-metrics. Rename the controller service metrics port from tcp-prometheus-servicemonitor to http-metrics for consistency with other workloads. Users must remove these keys from their value overrides before upgrading. (#8952, @erikgb)
• Update base images to Debian 13. (#8849, @ltwongaa)