cert-manager is the easiest way to automatically manage certificates in Kubernetes and OpenShift clusters.
cert-manager 1.21 brings ACME Renewal Information (ARI) support, AWS IAM authentication for the Vault issuer, several security hardening changes, and continued improvements to Gateway API integration and cainjector. There are three breaking changes related to Helm chart RBAC and metrics values โ review them carefully before upgrading.
Known Issues
- Controller crash-loops when a Certificate sets
renewal.policy: Disabled: the new Certificate renewal policies feature (#8258) causes a nil pointer dereference panic in the trigger controller whenever a Certificate's spec.renewal.policy is set to Disabled โ pki.RenewalTime() returns (nil, nil) for that policy, but the caller unconditionally dereferences the result. This crashes the controller process (crash-loop) for any cluster with such a Certificate. Workaround: do not set renewal.policy: Disabled on any Certificate until this is fixed; remove the field (or set a different policy) from any Certificate that already has it, and restart the controller if it is currently crash-looping. See #9031 for details.
- Log spam for non-cert-manager-labelled Secret events: the typed predicates refactoring (#8407) causes
filteredEventHandler type assertion failures ("OnAdd missing Object", "OnUpdate missing ObjectOld", "OnDelete missing Object") for every non-cert-manager-labelled Secret event, multiplied by 7 certificate sub-controllers. This is cosmetic only โ the affected controllers only need events from cert-manager-labelled Secrets (which arrive via the typed informer); the metadata informer events were always filtered out by predicates in previous versions. Issuer and ClusterIssuer controllers are not affected. See #8994 for details.
- Issuer/ClusterIssuer can get stuck at
Ready: False, Reason: InvalidSolver and never self-correct: new eager validation of ACME solver Secrets (#8255) means an Issuer/ClusterIssuer referencing a solver Secret (e.g. a DNS01 provider credential) that doesn't exist yet will correctly report Ready: False, but creating the missing Secret afterwards does not trigger re-reconciliation โ the controller's Secret-watch logic was never updated to recognise solver Secrets. It will only recover on the next 10-hour informer resync, a change to the Issuer/ClusterIssuer's own spec, or a controller restart. Workaround: after creating the missing Secret, make a trivial edit to the Issuer/ClusterIssuer spec (or delete and recreate it) to force reconciliation. See cert-manager/cert-manager#9036 for details and a fix proposal.
Major Themes
Default tokenrequest RBAC removed from Helm chart
โ ๏ธ Breaking change
The Helm chart no longer creates a default Role and RoleBinding granting the cert-manager controller permission to create tokens for its own ServiceAccount (serviceaccounts/token: create). No documented workflow requires this RBAC โ the Route53 docs section that motivated it was removed in 2024.
If you use serviceAccountRef.name pointing at the controller ServiceAccount, you must now either create your own Role/RoleBinding granting serviceaccounts/token: create, or migrate to a dedicated ServiceAccount (recommended โ see the Vault or Route53 documentation).
Restrict Challenge and Order RBAC in cert-manager-edit ClusterRole
โ ๏ธ Potentially breaking change
The cert-manager-edit aggregate ClusterRole no longer grants create for challenges.acme.cert-manager.io or create, patch, update for orders.acme.cert-manager.io (GHSA-8rvj-mm4h-c258). These resources are internal to cert-manager's ACME workflow. Challenge patch and update are retained because users may need them to remove stuck finalizers.
This change was already shipped in v1.20.3 and v1.19.6, so if you are running one of those versions this will not be a breaking change. If you have tooling that creates Challenge or Order resources directly, you will need to grant those permissions explicitly.
Metrics port name and path Helm values removed
โ ๏ธ Breaking change
The Helm values prometheus.servicemonitor.targetPort, prometheus.servicemonitor.path, and prometheus.podmonitor.path have been removed. The controller Service metrics port has been renamed from tcp-prometheus-servicemonitor to http-metrics. Because the Helm values schema uses additionalProperties: false, users who still have any of the removed keys in their values overrides will see a schema validation error on upgrade โ remove them before upgrading. (#8952)
ACME and Certificate Management
- ACME Renewal Information (ARI): experimental support for RFC 9773 behind the
ACMEUseARI feature gate. When enabled, cert-manager queries the ACME server's renewalInfo endpoint for the recommended renewal window, allowing servers like Let's Encrypt to proactively prompt renewal during mass revocations or CA key rollovers. (#8798)
waitInsteadOfSelfCheck solver option: skip cert-manager's own self-check and instead wait a configured duration before asking the ACME server to validate. An escape hatch for split-horizon DNS and NAT hairpin environments. See configuration details. (#8858)
- AWS IAM authentication for Vault: the Vault issuer now supports IRSA, EKS Pod Identity, and ambient EC2/ECS credentials, removing the need for long-lived AWS Secrets. (#8422)
- Certificate renewal policies: a new
renewalPolicies field on the Certificate API provides more expressive control over renewal scheduling, complementing renewBefore and renewBeforePercentage. (#8258)
- Configurable CertificateRequest retry backoff: the new
--certificate-request-maximum-backoff-duration flag (default: 32 hours) caps the exponential backoff for failed CertificateRequests, useful for environments with scheduled CA maintenance windows. (#8893)
- Modern2026 PKCS#12 profile: a new FIPS 140-3 compatible encoding profile using AES-256 + SHA-256 KDFs instead of legacy 3DES/RC2. (#8841)
- Webhook certificate renewal after system suspend: the webhook now detects missed certificate renewals after system suspend (S3/S4) or VM live migration by polling wall-clock time, recovering within one minute of resume. (#8464)
Gateway API and cainjector
- HTTP01 ListenerSet parentRef fallback: the
acme.cert-manager.io/http01-parentreffallback: "true" annotation causes cert-manager to use the parent Gateway for solver HTTPRoutes instead of the ListenerSet, enabling TLS-only ListenerSets to use a shared HTTP listener for ACME challenges. (#8749)
cert-manager.io/ignore-tls-listeners annotation: exclude specific Gateway TLS listeners from certificate management. (#8727)
- Additional listener protocols: configurable listener protocols beyond the default set. (#8683)
enableGatewayAPI configuration restructure: enableGatewayAPI and enableGatewayAPIListenerSet are deprecated in favor of gatewayAPI.enabled / gatewayAPI.enableListenerSet. The old fields continue to work. (#8732)
CAInjectorMerging promoted to GA: unconditionally enabled; will be removed in a future release. (#8583)
- cainjector server-side apply unconditional: the
ServerSideApply feature gate is deprecated. (#8692)
- cainjector
--ignore-namespaces flag: skip specified namespaces when watching Secrets for injection. (#8614)
Deployment and Observability
- Venafi OAuth token observability: a new
AuthFailed Issuer condition reason distinguishes bad credentials from transient errors. PANW NGTS is now supported as a Venafi backend. (#8808, #8779)
runtimeClassName support: configurable for cert-manager components and ACME HTTP01 solver pods. (#8791, #8976)
startupapicheck.ttlSecondsAfterFinished: opt-in automatic cleanup of the startupapicheck Job. (#8523)
--acme-http01-solver-extra-labels: propagate global.commonLabels to dynamically-created ACME HTTP01 solver resources. (#8761)
Notable Bug Fixes
- Integer overflow in
renewBeforePercentage: Certificates with durations longer than approximately 3 years were incorrectly rejected or assigned incorrect renewal times. (#8947)
- Infinite re-issuance loop: cert-manager no longer loops when an issuer returns an already-expired certificate. (#8610)
- ACME transient network errors: challenges no longer permanently fail on TLS handshake timeouts, DNS resolution failures, or context cancellation during nonce fetches and authorization waits. (#8760)
- DNS-over-HTTPS response body cap: response body reads are now bounded at 128 KB to prevent potential OOM. (#8803)
- Vault path traversal: the Vault issuer webhook now rejects
.. path segments, preventing path.Join from silently resolving relative segments. (#8930)
- DNS issuer secrets validated before ready: prevents silent misconfiguration. (#8255)
Community
As always, we'd like to thank all of the community members who helped in this release cycle, including all below who merged a PR and anyone that helped by commenting on issues, testing, or getting involved in cert-manager meetings. We're lucky to have you involved.
A special thanks to:
- @Copilot
- @FelixPhipps
- @Peac36
- @SebTardif
- @apkatsikas
- @bitloi
- @dap0am
- @figaw
- @immanuwell
- @jabbrwcky
- @jnohlgard
- @jsoref
- @ltwongaa
- @lunarwhite
- @mateenali66
- @onurmicoogullari
- @putongyong
- @seanorama
- @texasich
for their contributions, comments and support!
Also, thanks to the cert-manager maintainer team for their help in this release:
โฆ(truncated)