Kubernetes outage history
Kubernetes had 53 incidents in the last 12 months.
- Incidents, 12 months
- 53
- 16 critical or high
- Median time to fix
- n/a
- 0 resolved
- Longest
- n/a
- Never on the status page
- n/a
- no status page
Kubernetes incidents per month
View as table
| Month | Incidents |
|---|---|
| 2026-11 | 49 |
| 2026-10 | 4 |
| 2026-09 | 0 |
| 2026-08 | 0 |
| 2026-07 | 0 |
| 2026-06 | 0 |
| 2026-05 | 0 |
| 2026-04 | 0 |
| 2026-03 | 0 |
| 2026-02 | 0 |
| 2026-01 | 0 |
| 2025-12 | 0 |
| 2025-11 | 0 |
Every Kubernetes incident in the last 12 months
| Date | Incident | Severity | Time to fix | Status page |
|---|---|---|---|---|
| Kubernetes Topology Manager Leaks container_aligned_compute_resources_count on Retries Cause: The topology manager admit handler executes too early during pod updates (before PredicateAdmitHandler), admitting the update and incrementing the container scope counter for every container on every retry. | minor | ongoing | ||
| Kubernetes StatefulSet Controller Memory Leak from Undeleted Metrics | medium | ongoing | ||
| k8s-mcp-server Unauthenticated RCE/Tool Dispatch Vulnerability Cause: The SSE server in main.go is created without an authentication option or context function, and the POST /message handler accepts loadable sessionId values without checking authorization or Origin headers prior to dispatching tools/call. | high | ongoing | ||
| Kubernetes Downward API Volumes Ignore Pod-Level CPU and Memory Limits | medium | ongoing | ||
| Kubernetes kubectl get -o custom-columns-file Ignores --no-headers | minor | ongoing | ||
| Kubernetes EndpointSlice Auto topology hints stay stale after CPU changes Cause: CPU-only Node updates do not refresh the cached zone capacity ratios for services using automatic topology hints. | medium | ongoing | ||
| Kubernetes WebSocket Exec Heartbeat Times Out After 61 Seconds Under Backpressure | medium | ongoing | ||
| Kubernetes Scheduler AddGenericPodGroup Can Expose Partially Populated PodGroup Cause: PriorityQueue.AddGenericPodGroup makes the root group poppable while it holds only one member when observed after its pods, and podGroupMemberPods.clear builds the pod list by ranging over a map leading to random member order. | medium | ongoing | ||
| Kubernetes kubectl Silently Drops Keys on Lines Exceeding 64 KiB in Env Files Cause: cmdutil.AddFromEnvFile uses a default bufio.NewScanner(f) with a 64 KiB token limit and returns nil after the scanner loop without verifying scanner.Err(). | medium | ongoing | ||
| Kubernetes Kubelet Restart Kills Running Pods Using Device Plugin Resources Cause: Pods utilizing device plugin resources through pod overhead fail re-admission checks when the kubelet restarts. | high | ongoing | ||
| Kubernetes E2E Test Failure in Job Pod Resource Updates | minor | ongoing | ||
| Kubernetes Scale Update Drops Spec.Replicas Manager Ownership Cause: The first scale write incorrectly splits the kubectl-create entry of managedFields into two entries with the same manager, operation, and apiVersion, altering field ownership tracking. | medium | ongoing | ||
| Kubernetes Kubectl Silently Shows Endpoints: None on EndpointSlice List Failure Cause: ServiceDescriber.Describe in staging/src/k8s.io/kubectl/pkg/describe/describe.go ignores the error returned when calling d.DiscoveryV1().EndpointSlices(service.Namespace).List(...). | minor | ongoing | ||
| Kubernetes kubectl v1.36.6 Bundles Vulnerable Go Packages with CVEs Cause: The kubectl release binaries include outdated versions of golang.org/x/net and golang.org/x/text libraries that contain multiple CVEs. | high | ongoing | ||
| Kubernetes apiserver Watch Cache Silently Drops Events with Identical ResourceVersions Cause: In pkg/apiserver/watchcache (dispatchEvent), events from the same etcd transaction sharing a resourceVersion are incorrectly filtered, causing only the first event to be dispatched while the rest are discarded. | high | ongoing | ||
| Kubernetes VolumeBinding Filter Cost Scaled with Nodes and CSIStorageCapacity Cause: The VolumeBinding filter evaluates capacity checks per node by listing all cluster-wide CSIStorageCapacity objects and iterating through label selectors iteratively. | medium | ongoing | ||
| Kubernetes 1.38 CRD CEL Validation Rules Failing After cel-go Upgrade Cause: An upgrade to cel-go v0.33.0 in Kubernetes 1.38 increased the estimated cost of certain CEL validation rules, such as indexing split() results and calling linear-cost string functions, causing them to exceed limits. | high | ongoing | ||
| Kubelet Incorrect Memory Accounting With Systemd 259 cgroup v2 Cause: Systemd 259+ enables memory_hugetlb_accounting combining hugepages and memory into the same accounting pool, which conflicts with kubelet tracking them separately. | high | ongoing | ||
| Kubernetes PodGC Force Deletes Replacement Pods Due to Missing UID Precondition Cause: markFailedAndDeletePodWithCondition in pkg/controller/podgc/gc_controller.go performs deletions by name without a UID precondition or grace period, and skips the UID-pinned status patch for pods already in a terminal phase. | high | ongoing | ||
| Kubernetes 1.37 DRA Pods Wait ~5 Minutes for Devices from Deleted Pods | medium | ongoing | ||
| Kubernetes kube-scheduler: DRA Extended Resource Scoring Uses Stale Caches Cause: DRACaches deviceMatchCache and related caches store results in sync.Maps without an invalidation mechanism over the lifetime of the scheduler process. | medium | ongoing | ||
| Kubernetes DynamicResources Plugin Assigns Wrong ResourceClaims to Pods | medium | ongoing | ||
| Kubernetes APF queueSet Panics Due to Premature Removal of Lingering Requests Cause: The queueSet implementation removes a request from the executing requests set immediately upon finishing its main service, even though the request still holds seats that are only released later via a deferred EventAfterDuration callback. | high | ongoing | ||
| Kubernetes Kubelet Fails to Reconcile Partial Cgroup Configuration Failures Cause: Kubelet gates pcm.EnsureExists on !pcm.Exists(pod) during SyncPod. If cgroup configuration fails after the directory is created, subsequent syncs only check if the cgroup exists and skip reapplying the configuration. | medium | ongoing | ||
| Checkmk Kubernetes Special Agent Argument Injection Vulnerability (CVE-2026-121374) Cause: Argument injection via crafted node roles in the 'Exclude nodes based on their role' rule option in the Kubernetes special agent allows arbitrary file writes. | critical | ongoing | ||
| Kubernetes Kubelet Stops Negotiating HTTP/2 When ReloadKubeletClientCAFile is Enabled Cause: In NewMainKubelet (pkg/kubelet/kubelet.go), the implementation for ReloadKubeletClientCAFile overrides or misconfigures the TLS config handling such that ALPN is no longer negotiated. | medium | ongoing | ||
| Kubernetes PodReadyToStartContainers Condition Lacks Sandbox Creation Error Details Cause: The PodReadyToStartContainers condition reports the state of the sandbox (e.g., 'no pod sandbox exists') but fails to include the underlying error message from the failed sandbox creation attempt. | minor | ongoing | ||
| Kubernetes Pod with Ephemeral Local Volume Preempted During PreBind Stays Pinned Cause: The scheduler preempts a victim in PreBind in memory and cancels the binding cycle without deleting the pod, while the VolumeBinding PreBind API writes (such as the selected-node annotation on the PVC) are not reverted because Unreserve only reverts the assume cache. | medium | ongoing | ||
| Kubernetes Pods Rejected by Missing CSI Driver Not Requeued Promptly on Registration Cause: When preventPodSchedulingIfMissing is enabled (VolumeLimitScaling, enabled by default in 1.37), NodeVolumeLimits rejects pods referencing an unregistered CSI driver, and subsequent driver registration fails to trigger a pod retry, leaving them stuck until the unschedulable pods duration timeout expires. | medium | ongoing | ||
| Kubernetes Scheduler Cache Loses Bound Pod After Failed Bind Response | high | ongoing | ||
| Kubernetes apiserver cacher DATA RACE in integration tests | minor | ongoing | ||
| Kubernetes cluster/log-dump Fails to Report GCE Node Failures Due to Deprecated Logs Cause: detect_node_failures in cluster/log-dump/log-dump.sh queries the legacy Compute Engine activity log (compute.googleapis.com%2Factivity_log), which is deprecated and results in host errors never being reported. | minor | ongoing | ||
| Kubernetes Kubelet Reports Zero Resource Request for Pods Using Pod-Level Resources Cause: GetResourceRequestQuantity in pkg/api/v1/resource/helpers.go skips container-level resources whenever any pod-level resource is set, resulting in a zero resource request calculation. | high | ongoing | ||
| Kubernetes Device Plugin NUMA Hints Ignore Pending Device Requests Cause: Device plugin NUMA hints fail to account for pending device requests from admitted pods when PodLevelResourceManagers is enabled with pod-scoped Topology Manager. | medium | ongoing | ||
| Kubernetes Device Plugin PreStartContainer Fails with Pod-Level Resources Cause: The Device Plugin PreStartContainer fails before deferred allocation occurs when using pod-level resources and pod scope in the Topology Manager, because the container's device allocation is not yet present in the local cache. | high | ongoing | ||
| Kubernetes Kubelet Env File Parser Violates KEP-3721 on Duplicate Keys Cause: ParseEnv in pkg/kubelet/util/env/env_util.go immediately returns upon encountering the first instance of a key rather than continuing to process subsequent definitions. | minor | ongoing | ||
| Kubernetes metav1.ParseToLabelSelector Silently Drops Conflicting Equality Requirements Cause: The function keeps only the last provided equality requirement when duplicate keys are present, unlike labels.Parse which retains both. | medium | ongoing | ||
| Kubernetes Default Preemption Plugin Evicts Partial Gang-Scheduling Members Causing Deadlock Cause: The scheduler identifies victim pods based purely on node-local priority evaluations, PDBs, and single-node disruption constraints without accounting for atomic gang-scheduling dependencies. | high | ongoing | ||
| Kubernetes Device Manager Checkpoint State Desynchronizes on Admission Failure Cause: Container admission failure occurs after devicemanager.Allocate() has already updated and persisted the device allocation checkpoint state to disk without rolling back the checkpoint upon admission failure. | medium | ongoing | ||
| Kubernetes Concurrent Watch Decoder Goroutine Panic Handling Issues Cause: Workers recovering from a panic via HandleCrash failed to publish results, causing callers to wait indefinitely without notification of failure. | medium | ongoing | ||
| Kubernetes List Requests with Limit and Field Selector Return Latest Revision Instead of Exact ResourceVersion | medium | ongoing | ||
| Kubernetes Server-Side Apply Status Subresource Incorrectly Acquires Root Field Ownership Cause: Server-side Apply on the /status subresource acquires ownership of additional root fields that are restored by the CRD status strategy from the old object. | medium | ongoing | ||
| Kubernetes ResourceClaim Updates Fail Due to Non-Canonical JSON Opaque Parameters Cause: The client sends non-canonical JSON bytes that are stored by etcd, but the apiserver re-encodes them upon return or patch application. Because the immutable spec check compares raw bytes, the mismatch causes the update to be rejected. | medium | ongoing | ||
| Kubernetes Kubelet: Node /etc/hosts Entries Shadow hostAliases in hostNetwork Pods Cause: Kubelet writes the node's /etc/hosts file content to the buffer before appending hostAliases entries from the spec, causing standard resolvers to return the node IP first for matching hostnames. | medium | ongoing | ||
| Kubernetes Uncached OpenAPI v3 Schemas Cause Excessive Heap Allocations Cause: Uncached OpenAPI v3 JSON schemas are repeatedly unmarshaled multiple times per resource during patch and verification operations. | medium | ongoing | ||
| Kubernetes Image Volumes with PullPolicy: Always Cause Excessive Pulls During CrashLoops | medium | ongoing | ||
| Kubernetes 1.36 and 1.37 PodResources API Breaks Device Allocation at Sandbox Creation Cause: Device Manager does not support pod-level resource allocation in these versions, making AllocatePod a no-op so that the device plugin's Allocate RPC only fires when the container requesting the device is created. | high | ongoing | ||
| Kubernetes kubectl port-forward Exits and Closes Local Listener on Server Workload RST | high | ongoing | ||
| Kubernetes Kubelet Static Pod URL Source Leaks Goroutines Cause: NewSourceURL in pkg/kubelet/config/http.go uses wait.NeverStop and unbuffered channel sends that block permanently when nothing reads from the channel after the context is canceled. | medium | ongoing | ||
| Kubernetes Kubelet Re-Runs Terminated Containers After Runtime Container Removal | medium | ongoing | ||
| Kubernetes Kubelet Re-runs Terminated Ephemeral Containers Cause: The kubelet decides whether to start an ephemeral container solely by checking if the runtime still has a container with that name. When an exited container is removed from the runtime, the next pod sync triggers the ephemeral container to start again. | medium | ongoing | ||
| Kubernetes Kubelet Panics on PodCertificateRequest Tombstone Deletion Cause: An unhandled type assertion in the PodCertificateRequest issuing manager fails when processing cache.DeletedFinalStateUnknown instead of a direct *v1.PodCertificateRequest object. | high | ongoing | ||
| Kubernetes Kubelet Image-Pull Errors Obscure NotFound and Anonymous Fallback Failures | minor | ongoing |
Compare Kubernetes's reliability
Questions
- How often does Kubernetes go down?
- Kubernetes had 53 incidents in the last 12 months.