$ cat kubernetes/month/2026-10-01.log
the month in review · October 2026
node-problem-detector 1.37.0 patches three CVEs
By RepoJournal · composed from the cited sources · human-reviewed weekly · methodology
The same release swaps OpenCensus for OpenTelemetry; elsewhere kubelet drops two deprecated feature gates and CRD types find a new home.
v1.37.0 kubernetes/node-problem-detector
The release patches three CVEs and replaces OpenCensus with OpenTelemetry for metrics export; the release notes say Prometheus metric names are unchanged, so dashboards and recording rules keep working. It also adds a per-rule invoke_interval for custom plugin monitor rules and zswap metrics memory/zswap_bytes_used and memory/zswap_compression_efficiency.
HPA: compute resource utilization ratio without int64 overflow kubernetes/kubernetes
GetResourceUtilizationRatio summed per-pod metrics and requests as raw int64, multiplied by 100 and narrowed to int32, and all three steps wrapped silently near math.MaxInt64. The HPA now computes the ratio without that overflow, which matters if you push large requests or high-cardinality metric values through scale decisions.
cacher: stream watch cache snapshots without an item slice kubernetes/kubernetes
The cacher can now stream watch cache snapshots without materializing an item slice, the work that later carried Limit support and alignment fixes through the month. For big clusters it targets the memory spike on large List and watch initializations rather than request latency.
AEP-9970: per-VPA target percentiles kubernetes/autoscaler
AEP-9970 adds targetCPUPercentile and targetMemoryPercentile to ContainerResourcePolicy, extending AEP-8026's per-VPA configuration to percentile-based targets. If p95 CPU rather than the mean matches your workload, this is the knob you will be tuning.
Update the GCP provider to allow users to skip firewall actions kubernetes/cloud-provider-gcp
A new ManageFirewallRules boolean lets clusters skip creation, deletion, and update of firewall rules entirely, for operators who manage them outside the provider. The input notes users may not want the provider touching them; flip it off and the provider stops reconciling.
add watch-list mode to request benchmark kubernetes/perf-tests
An optional HTTP WatchList benchmark mode measures WatchList latency while ruling out client-side typed-object decoding as a source of overhead. Useful if you are deciding whether WatchList cost comes from the wire or your own decoders.
Migrate 5,000-node scale tests to c4d-standard-96 kubernetes/test-infra
All 5,000-node scale test control planes move from c4-standard-96 to c4d-standard-96, reverting an earlier experiment. Upstream notes severe run-to-run variance on c4-standard-96 from host processor differences in the cloud provider pool, so older scale results may not be comparable.
$ ls kubernetes/month/ # the briefings behind this review
Keep up with Kubernetes in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.
fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.
Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.