v0.3.0
Changes since v0.3.0-alpha.1
-
🎉 update to latest kubernetes dependencies, allowing automatic use of delegated auth on clusters without client certificate auth enabled
-
🐛 ensure earliest timestamps are always reported for metric points. This means it's safe to assume that if a metric point came after pod initialization (factoring in the window), all metrics contained therein came from after pod initialization.
-
⚠️ / 🎉 change reported window size to 30s, to be more accurate (cAdvsior's calculation window can vary from 10s-30s, but it's not possible to tell the exact window at this point).
Changes since v0.2.1
⚠️ flag names have changed significantly. See #113 for more information⚠️ Prometheus metrics about metrics-server health have changed names and types⚠️ / 🎉 use the secure Kubelet port with auth by default. Use of the insecure port is deprecated and may be removed as soon as next release.⚠️ / 🎉 source usage rate from Kubelet directly This should improve metrics reliability, but means that the scrape interval no longer has any relation to the metrics calculation window (which is now effectively between 10-30s).⚠️ / 🎉 node addresses are now chosen with the same mechanics as kube-apiserver, and can be customized with an identically named flag. This should not affect most installations, but should make it easier to work with clusters with many different node addresses, and makes metric-server more consistent with the rest of kubernetes- 🎉 support IPv6 kubelet addresses
- 🐛 avoid passing bearer token to insecure port