attune-io/attune v0.1.25

on GitHub

What's New in v0.1.25

v0.1.25 adds kubectl attune doctor so you can check Kubernetes 1.32+ and pods/resize before you wait on recommendations. Canary is per app: cooldown, CREATE sizing, startup boost, and HPA retune no longer lock every workload on the same policy. Helm chart 0.1.25 defaults to the bare SemVer image tag that this release publishes.

Highlights

• Run kubectl attune doctor after install to confirm the cluster can in-place resize.
• Multi-app canary policies promote, cool down, and CREATE-size each app on its own clock.

kubectl attune doctor

• There was no preflight command to confirm the cluster can in-place resize. kubectl attune doctor checks the server version, discovery of pods/resize, and GET /-/healthy on Prometheus addresses from policies and defaults. The ping runs on the machine that invoked kubectl, not inside the operator pod. In-cluster DNS (.svc / .cluster.local) is skipped. HTTP 401/403 on an address that sets bearerTokenSecret or headers is skipped, because doctor does not send the operator's credentials. Optional Prometheus failures print WARN and keep exit code 0 (#582, #583, #585, #589).

Canary isolation

• Canary observation started before any pod was resized. The clock now starts on the first successful in-place canary resize. A revert clears that app's clock (#568).

• One app's cooldown or canary slice locked every other app on the policy. Cooldown, CREATE initial sizing, startup boost, and HPA retune are per app. After a cooldown skip, the operator requeues when that stamp expires instead of waiting a full new cooldown. Selector-based policies fetch the owning Deployment, StatefulSet, or DaemonSet so CREATE isolation applies there too (#571, #572, #574, #575).

• A request increase only compared this pod to node allocatable. The operator now subtracts other pods on the node before growing requests (#571).

• ReplicaSet CREATE often has an empty pod name. Canary matching does not treat that empty name as a canary-slice identity. When CREATE has no assigned name, Info logs use generateName. Verbosity 1 records the skip, including an empty pod field (#587, #589).

GitOps pull requests

• After upgrading from 0.1.22, cooldown expiry opened another empty GitOps PR. When the last-attempt URL is set and the drift fingerprint annotation is empty, the operator adopts the current table instead of opening a duplicate. A failed API open (empty URL) still retries (#577, #580).

CLI CSV output

• kubectl attune recommendations -o csv and savings -o csv write a header plus one data row per policy or container. The recommendations last column is confidence_or_status: a confidence percent when recommendations exist, otherwise the policy Ready reason (#581, #586).

Helm image tag

• Chart 0.1.24 prefixed v onto appVersion when image.tag was empty. This release publishes both v0.1.25 and 0.1.25. An empty image.tag now uses bare appVersion (ghcr.io/attune-io/attune:0.1.25). An explicit image.tag is left unchanged (#554).

Upgrade notes

1. Upgrade the chart to 0.1.25, or set image.tag to 0.1.25 or v0.1.25.
2. Pull ghcr.io/attune-io/attune:v0.1.25 or ghcr.io/attune-io/attune:0.1.25. Both tags point at the same digest.
3. Chart 0.1.24 still defaults to :v0.1.24. Chart 0.1.23 still needs --set image.tag=v0.1.23.
4. Run kubectl attune doctor after install to confirm Kubernetes 1.32+ and pods/resize.
5. Multi-app canary policies now promote per app. Check status.canary.workloads[] for each app's phase and remaining cooldown.

Full changelog

v0.1.24...v0.1.25