Runtimezruntimez
runtimez / product / upgrade-readiness
product · upgrade readiness

Know what your next Kubernetes upgrade breaks — before you book the window.

Pick a target version and see every resource that breaks, ranked by blast radius. Each of the 793 upgrade rules quotes the vendor's release note and ships a read-only kubectl command, so you can check every verdict yourself.

one helm install · get, list, watch only · first findings in minutes

Upgrade Readiness overview: NOT READY for 1.35 with 2 blockers, findings by gate, a 1.35 / 1.36 / 1.37 comparison ladder and 398 days to the forced upgradeUpgrade Readiness overview: NOT READY for 1.35 with 2 blockers, findings by gate, a 1.35 / 1.36 / 1.37 comparison ladder and 398 days to the forced upgrade
793upgrade rules that quote the vendor
1.30 → 1.37hand-authored Kubernetes rule corpus
449config-breaker entries from kubelet and control-plane flags
11add-on catalogs, enabled on detection

Overview

Upgrading Kubernetes means reading release notes for the control plane, every add-on and every node image, then working out which of your objects they touch. Runtimez does that reading continuously against the live cluster. It groups what it finds into gates — add-ons, nodes and skew, APIs and CRDs, config, workload hygiene — and turns them into a plan you can run.

Works with
Amazon EKSGoogle GKEAzure AKSKarpenterTraefikIstioArgo CDcert-managerCoreDNSkube-proxy
Breakage radar

Every resource the next version breaks, with the reason

One list per target version, ranked by gate and blast radius. It replaces a week of release-note reading.

  • Removed and deprecated APIs, RBAC semantics that change (like create now needed for exec over WebSocket), kubelet flags and cgroup v1.
  • Deprecated APIs still called at runtime, detected from API-server metrics — not just objects stored in etcd.
  • Each finding names the resource, the version it breaks at and the objects affected.
What breaks at 1.35: Karpenter unsupported, a ClusterRole missing create for pods/exec, v1/endpoints still in use, kubelet cgroup v1 settings and workload hygiene findingsWhat breaks at 1.35: Karpenter unsupported, a ClusterRole missing create for pods/exec, v1/endpoints still in use, kubelet cgroup v1 settings and workload hygiene findings
Upgrade strategy

A plan in the order you will run it

Findings become a before-and-after checklist, each step traceable to the rule that raised it.

  • Before you upgrade: blocking steps first, like upgrading Karpenter to 1.9 or migrating cgroup v1 nodes.
  • After the upgrade: features the new version makes available, such as in-place pod resize and trafficDistribution.
  • Every step cites its source rule id, so reviewers can audit the plan line by line.
Upgrade strategy for 1.35: before-you-upgrade blocking steps for Karpenter and cgroup v1, and after-the-upgrade adoption opportunities, each with its source ruleUpgrade strategy for 1.35: before-you-upgrade blocking steps for Karpenter and cgroup v1, and after-the-upgrade adoption opportunities, each with its source rule
Add-on compatibility

The add-on breaks before Kubernetes does

716 release-note rules for Traefik, Istio, CoreDNS, Argo CD, Karpenter, KEDA, cert-manager, kube-proxy, the AWS Load Balancer Controller, the EBS CSI driver and the VPC CNI. Each catalog turns on when the add-on is detected.

  • Vendor compatibility matrices per add-on: Karpenter 1.6.8 is marked blocking for 1.35, with the version to move to.
  • Each add-on gets its own upgrade path and a release-note checklist for what cannot be seen from the cluster.
  • Ticking a checklist item never changes the score — nothing is assumed on your behalf.
Add-ons tab: Karpenter 1.6.8 to 1.9 blocking with 3 release-note rules, CoreDNS under review with 23 rules to verifyAdd-ons tab: Karpenter 1.6.8 to 1.9 blocking with 3 release-note rules, CoreDNS under review with 23 rules to verify
Maintain

What runs on top of the cluster reaches end of life too

Maintain lists the operating systems, language runtimes and data services in your running images, the date each one loses vendor support, and what a data-service upgrade asks of its clients. Read-only, like the rest of the product.

  • End-of-life dates: 39 products from endoflife.date, including Kubernetes and EKS, eight Linux distributions, seven language runtimes (Node.js, Python, .NET, Go, Ruby, PHP, Java) and 14 data services. PostgreSQL 14 reaches end of life on 12 November 2026 and .NET 8 on 10 November 2026. Dates are updated regularly.
  • Upgrade steps you can't skip: 18 stops the vendors state for Elasticsearch, OpenSearch, MongoDB, MySQL, Kafka, RabbitMQ and Cassandra, each citing the vendor's upgrade notes. Elasticsearch 9.0 needs 8.18 first (9.1 and later need 8.19); MySQL 5.7 goes to 8.0, then 8.4, before 9.0.
  • Minimum driver versions: 145 minimum client versions across 8 data services, each with the upstream source it rests on, tested against the client libraries in the images that use the server. PostgreSQL 14 stores new passwords as SCRAM-SHA-256, so pgjdbc before 42.2.1 cannot log in. A driver with no stated minimum is marked for review, never ready.
runtimez / maintainread-only
what the running images carry · dates from endoflife.date
EOL
PostgreSQL 14 · 12 November 2026End of life for the release line the image runs.
EOL
.NET 8 · 10 November 2026End of support for the runtime in the image.
PATH
Elasticsearch 8.10 → 9.0Stop at 8.18 first. Elastic: 9.0.x is reached from 8.18.x.
PATH
MySQL 5.7 → 9.0Stop at 8.0, then 8.4. A bugfix or LTS series cannot be skipped.
DRIVER
pgjdbc below 42.2.1Cannot log in once PostgreSQL 14 stores passwords as SCRAM-SHA-256.
Illustrative example · the dates, stops and minimums are the real catalog values
Fleet view

Know which cluster to upgrade first

Every connected cluster side by side, with an attention queue that answers “what do I do today?”.

  • Fleet risk, open findings, CVE posture and releases in one row.
  • Clusters ranked by breakage risk and forced-upgrade deadline.
  • The deadline comes with the extended-support cost of waiting, next to the effort to upgrade.
Fleet home: fleet risk 84, clusters, 893 open findings, CVE posture and releases, with a needs-your-attention queueFleet home: fleet risk 84, clusters, 893 open findings, CVE posture and releases, with a needs-your-attention queue
Open source

The same rules, as a free CLI

kube-upgrade-check runs the same rule catalogs, except the AWS Load Balancer Controller's, as a free CLI. No account, no agent.

  • Install with brew install runtimez-com/tap/kube-upgrade-check.
  • Apache-2.0; the whole rule catalog is JSON in the repository.
  • Point it at a live cluster with kubectl upgrade-check --target 1.35; --fail-on high exits non-zero for a pipeline gate. It does not scan manifests, so keep Pluto in CI for those. See the CLI →
kube-upgrade-check running in a terminal and listing upgrade findings

More in Upgrade Readiness

Everything else that ships in this product today.

Config breakers

Reads kubelet /configz and control-plane flags to catch settings the target release removes — invisible to manifest scanners.

CRD blast radius

Counts the custom resources that will be orphaned, not just which CRD version is old.

Webhook and node-runtime skew

Admission-webhook and node-runtime checks plus 5 version-skew rules: the two things that stall a control-plane roll halfway.

Gate ledger

Pass or fail per rule, with the objects named and the command to re-check it yourself.

Coverage self-assessment

Tells you what it could not see — like provider-managed control-plane flags — so a green verdict means something.

End-of-life and driver floors

End-of-life dates for 39 products, 18 upgrade stops you can't skip, and 145 minimum client versions for data-service upgrades.

Export report

Share the readiness verdict for the selected target with the people who approve the maintenance window.

How it compares

vs pluto / kubent

They read manifests. Runtimez also reads CRDs, webhooks, kubelet flags, node runtimes and 11 add-ons, and keeps watching after the scan.

Frequently asked questions

Still curious? hello@runtimez.io

Last updated

Which Kubernetes versions are covered?

The hand-authored corpus covers Kubernetes 1.30 through 1.37 with 72 rules and 5 version-skew rules. Add-on catalogs follow each vendor's own release notes.

Does it work on EKS, GKE and AKS?

Yes. It runs on Amazon EKS, Google GKE, Azure AKS and any conformant cluster. On managed control planes it says which flags it cannot inspect instead of reporting them as clean.

Does Runtimez need write access to my cluster?

No. The agent only uses get, list and watch, collects secret names but never values, and uninstalls with one Helm command.

How is this different from pluto or kubent?

Those tools read manifests for deprecated APIs. Runtimez also checks APIs still called at runtime, kubelet and control-plane config, CRDs and instance counts, admission webhooks, node runtimes and add-on compatibility, and keeps re-evaluating as the cluster changes.

Does it tell me about end-of-life software, not just Kubernetes?

Yes. Maintain tracks end-of-life dates for 39 products from endoflife.date, including Linux distributions, seven language runtimes and 14 data services, and matches them to what your running images contain. For data services it also lists the upgrade steps the vendor says you can't skip and the minimum client versions the new server version needs. Dates are updated regularly, and like the rest of Runtimez it is read-only.

Can I try the rules without installing the agent?

Yes. kube-upgrade-check runs the same rule catalogs, except the AWS Load Balancer Controller's, from your laptop against a live cluster, with no account and no agent.

See it on your own cluster in under an hour.

Free for your first cluster. Read-only by default. Uninstall is one helm command.