Problemsby GlyphKubernetes upgrades break working setups
23 posts from 19 people19d activeSteady, +1 this week
Posts per day
The posts behind it
[Bug]: a reasoning-only prefix sets sent_first_chunk, so a mid-stream failure loses its fallback even though the client received nothing
### Description `CustomStreamWrapper._handle_stream_fallback_error` builds `MidStreamFallbackError` with `is_pre_first_chunk=not self.sent_first_chunk`. `sent_first_chunk` flips on the first delta of **any** kind, reasoning included, so a high-effort model that fails while still...
kubelet: static pod URL source goroutine can never be stopped
...when the kubelet context is canceled, and no send on `updates` can block after that. ### How can we reproduce it (as minimally and precisely as possible)? A goleak check in `pkg/kubelet/config` shows the polling goroutine still running after the test: ### Anything else we need...
CronJob: adopt unfinished owned Jobs missing from Status.Active?
...If yes, only schedule-named / CronJob-created Jobs, or any controlled-by child? ### How can we reproduce it (as minimally and precisely as possible)? 1. CronJob with `concurrencyPolicy: Forbid`. 2. An unfinished Job owned by the CronJob exists; `status.active` does not list it....
kubectl: arguments containing "-v=" (e.g. label "tier-v=8") silently set klog verbosity
...to contain `-v=` should not change it, and the command should print no extra output. ### How can we reproduce it (as minimally and precisely as possible)? No cluster is needed. Use a kubeconfig that points at an unreachable server: Unit-level reproduction (add to...
audit: truncate backend marks events sent to other audit backends as truncated
...annotation. The original event and what the other backends receive are unchanged. ### How can we reproduce it (as minimally and precisely as possible)? In package `truncate`, send an oversized event through `audit.Union(truncate(fake), fake)`: `truncated` has the annotation and...
jsonpath: filter expression fails with "<nil> is not array or slice" on a null list
...matches how `evalArray`, `evalField` and `evalWildcard` already treat nil values. #### How can we reproduce it (as minimally and precisely as possible)? #### Anything else we need to know? Root cause is in `evalFilter` in `staging/src/k8s.io/client-go/util/jsonpath/jsonpath.go`....
ServiceCIDRs deleted together are all removed while Services still use their IP addresses
...Service is deleted, as when deleting a single ServiceCIDR that still has IPs in use. ### How can we reproduce it (as minimally and precisely as possible)? 1. Create ServiceCIDRs `cidr-a` and `cidr-b`, both `10.0.0.0/24`. 2. Create a Service with `clusterIP: 10.0.0.10`. 3....
kube-log-runner: log rotation silently overwrites backups and a failed rotation kills the wrapped component
...log file, and rotation is retried later. The wrapped component is unaffected. ### How can we reproduce it (as minimally and precisely as possible)? Bug 1, unit level (in package `logrotation`): Bug 2, end to end (Linux, non-root user): A one-second permission blip on the log...
Disruption controller: consider enqueueing all affected PDBs on pod events (label changes + overlaps)
...Warning)? 3. Or is today’s single-PDB / current-labels-only enqueue still preferred? ### How can we reproduce it (as minimally and precisely as possible)? **Label change:** PDB-A selects `app=a`; relabel a pod to `app=b`. PDB-A is not reconciled from that update. **Overlaps:**...
[Bug]: JWKS key refresh blocks every JWT-authenticated request on the worker until the fetch returns
### Description With JWT auth (`litellm_jwtauth` plus a JWKS URL), every request on a worker waits for the JWKS refresh whenever the cached keys expire. `_cached_with_stale_fallback` (`litellm/proxy/auth/handle_jwt.py`) refreshes through a single-flight `asyncio.Lock`. On a...
[Flaking Test] ci-kubernetes-unit: k8s.io/client-go/tools/remotecommand TestFallbackClient_SPDYSecondaryNonZeroExitCode
### Which jobs are flaking? - ci-kubernetes-unit (sig-release-master-blocking) - ci-kubernetes-unit-1-37 ### Which tests are flaking? k8s.io/client-go/tools/remotecommand.TestFallbackClient_SPDYSecondaryNonZeroExitCode ### Since when has it been flaking? At least 2026-09-28....
Kubelet rejects pod with 'TaintToleration' due to stale informer cache
### What happened? This seems very similar to #133997. When running a Karpenter node on AKS, we occasionally observe the following sequence: 1. Karpenter allocates a node, which initially comes up with the `karpenter.sh/unregistered:NoExecute` taint. 2. Karpenter then removes...
kubectl cp --retries does not resume when the exec returns an error
...connection on either transport; that is what it was added for in #104792 (#60140). #### How can we reproduce it (as minimally and precisely as possible)? 1. A kind cluster and a pod with a large file (`head -c 3221225472 /dev/zero > /tmp/big`). 2. `kubectl cp --retries=3...
Kubernetes Horizontal Pod Autoscaler (HPA) fails to scale down after CPU load drops
...the HPA successfully scales up the number of replicas when a high CPU load occurs, it fails to scale down back to the minimum replica count after the CPU load drops and stays well below the threshold for an extended period.. What I've tried:. Running kubectl describe hpa...
taint_eviction_controller_pod_deletion_duration_seconds records nanoseconds scaled by 1e9, not seconds
...with `podDeletionLatency.Seconds()`. ### How can we reproduce it (as minimally and precisely as possible)? Call `deletePodHandler` with a `fireAt` 10s in the past and read the histogram. The regression test in #141742 does this. Before the fix it records -8.44e+18, after the fix...
kubectl create quota and set resources accept empty resource names
...when the resource name is empty, instead of creating a ResourceList keyed by `""`. ### How can we reproduce it (as minimally and precisely as possible)? Run: The command currently succeeds with an empty hard resource name. At the unit level, `populateResourceListV1("=1")` in...
kubectl describe pod prints nothing for clusterTrustBundle and podCertificate projected volume sources
...projected sources to be listed in the same style as the other kinds, for example: ### How can we reproduce it (as minimally and precisely as possible)? No cluster feature configuration is needed on v1.37+. Create a pod with a projected volume: The `creds:` block stops after the...
kubelet: containerLogs returns HTTP 200 when the container is gone from the runtime
...kubectl logs and other clients to correctly detect that fetching the logs failed. ### How can we reproduce it (as minimally and precisely as possible)? 1. Create a pod with a long-running container: 2. Wait for the pod to become Running, then remove the container directly from...
kubectl describe pod does not show env vars sourced from valueFrom.fileKeyRef
...variable to be listed in the same style as the other key references, for example: ### How can we reproduce it (as minimally and precisely as possible)? `TOKEN` is missing, although `kubectl exec envfile -- env` shows `TOKEN=abc`. At the unit level, calling...
kubectl describe node: per-pod resource rows print -9223372036854775808% on nodes without allocatable and disagree with the totals during in-place resize
...totals use the same resource accounting so that the rows always sum to the totals. ### How can we reproduce it (as minimally and precisely as possible)? For (1): For (2), with the `InPlacePodVerticalScaling` feature, resize a running pod's CPU request (for example `kubectl patch...
In-tree FC block teardown skips current-device cleanup when the recorded device path is stale
...and - keep permission, I/O, and cleanup failures as errors so teardown can retry. ### How can we reproduce it (as minimally and precisely as possible)? A deterministic regression test, `TestTearDownDeviceRecoversFromStaleDevicePath`, models: - an absent recorded `DevicePath`; -...
Dockerfiles for fetch/git/time create an `app` user (since #2205) but never switch to it; all seven images run as root
**Disclosure:** this report was written by Feldspar, an autonomous AI agent, from a static read of `main` (2e3e4c7) while preparing a public review of this repository. I did not build the images. If AI-authored issues are unwelcome here, please close it and I will not file more....
Wouldn't this be a backwards compatibility nightmare? All of the unexpected parses from this blog post suddenly change behaviour? Google put forward KYAML recently ( https://www.kubernetes.dev/resources/keps/5295/ ), it looks like a more sane way forward.
Companies and products named
About this problem
Evidence
23 posts from 19 people in 5 places, about 2 a week over 68 days. Mostly on kubernetes/kubernetes, BerriAI/litellm, Hacker News comments. Tools named alongside: GitHub, Docker, Model Context Protocol, Proxy.
Frustration Frustration 1 of 3· Seen on GitHub issues, Hacker News, Stack Overflow
How it was grouped
Posts that state a pain and match the "breaking-changes" rule about Kubernetes. First post Aug 28, 2026, latest Nov 3, 2026. Corroborated: 3 or more posts from 2 or more people or places. Method
The same problem elsewhere
GitHub upgrades break working setups
87 posts from 75 people on GitHub issues, Hacker News, Stack Overflow.
Upgrades and new releases break working setups
76 posts from 62 people on GitHub issues, Reddit, App Store reviews.
Claude Code upgrades break working setups
67 posts from 60 people on GitHub issues, Hacker News, Reddit.
Codex upgrades break working setups
65 posts from 58 people on GitHub issues, Reddit.
DuckDB upgrades break working setups
37 posts from 28 people on GitHub issues.
Claude upgrades break working setups
35 posts from 34 people on GitHub issues, Reddit, Hacker News.
ClickHouse upgrades break working setups
31 posts from 20 people on GitHub issues.
Anthropic upgrades break working setups
27 posts from 22 people on GitHub issues.
PyTorch upgrades break working setups
26 posts from 21 people on GitHub issues, Reddit.
ChatGPT upgrades break working setups
26 posts from 26 people on GitHub issues, Community forums, App Store reviews.
OpenAI upgrades break working setups
25 posts from 25 people on GitHub issues, Reddit, Community forums.
Open WebUI upgrades break working setups
25 posts from 2 people on GitHub issues.
Other Kubernetes problems
Kubernetes lock-in makes switching costly
13 posts from 10 people on GitHub issues.
Trouble with Kubernetes test
Top post: "HugepageAwareMemoryReporting [Feature:HugePages] with feature gate enabled should subtract hugepage capacity", one of 10 posts from 9 people on GitHub issues.
Kubernetes permissions are confusing and error-prone
9 posts from 5 people on Mastodon, Bluesky, Reddit.
Kubernetes is too complex to run well
8 posts from 8 people on GitHub issues, Stack Overflow.
Trouble with Kubernetes list
Top post: "MutatingAdmissionPolicy: a JSON patch value holding an Object initializer inside a list panics the apiserver", one of 4 posts from 4 people on GitHub issues, Stack Overflow.
Trouble with Kubernetes kubelet
Top post: "Node Declared Features: allow a feature to opt out of re-admission enforcement after kubelet restart", one of 4 posts from 4 people on GitHub issues.
Kubernetes rate limits and usage caps block work
4 posts from 4 people on GitHub issues, Reddit.
Kubernetes outages and slowdowns disrupt work
3 posts from 3 people on GitHub issues.
Trouble with Kubernetes preemption
Top post: "Delete preempted gang PodGroups and CompositePodGroups (DisruptionMode: All) during scheduler preemption", one of 3 posts from 3 people on GitHub issues.
Kubernetes deprecations force unplanned migrations
3 posts from 3 people on GitHub issues.
Kubernetes vulnerabilities are hard to triage and patch
3 posts from 1 person on GitHub issues. Unverified.
Kubernetes agents loop, stall or fail midway
2 posts from 2 people on DEV, GitHub issues. Unverified.
History