attune-io/attune v0.1.21

on GitHub

What's New in v0.1.21

v0.1.21 is a large product release. Attune now helps with GitOps handoff (versioned export schemas and optional pull request automation), multi-cluster fleet reports, safer memory work on Kubernetes 1.35+, language runtime profiles, and clearer operator status when resizes are Deferred or Infeasible. If you run GitOps, multi-cluster fleets, JVM-heavy workloads, or mixed 1.33–1.35 clusters, read the upgrade notes below before rolling out.

Highlights

• Opt-in GitOps pull request automation (with dry-run) plus versioned recommendation ConfigMaps for durable export.
• Fleet report ConfigMap for multi-cluster rollups, with metrics and dashboard panels.
• Kubernetes 1.35+ live memory limit decreases, with a usage floor so limits stay above recent usage.
• Runtime profiles (java, python, nodejs, golang) apply safe in-memory memory defaults.
• ResizeBlocked status when pods are Deferred or Infeasible, plus capacity and node-pressure skip metrics.

GitOps and multi-cluster

• Versioned recommendation export. ConfigMaps written for GitOps include schema-version and an attune.io/export-schema label so pipelines can evolve safely (#438).
• Opt-in GitOps pull request automation. When export pull requests are enabled, Attune can open or update PRs against your forge (dry-run path for CI without credentials). Missing head branches can be bootstrapped from the base branch (#449, #456, #455).
• Multi-cluster fleet report. Optional fleet report ConfigMap export for federation-friendly rollups, with Grafana and PrometheusRule coverage (#452).

See GitOps integration and Multi-cluster.

Memory safety and Kubernetes 1.35+

• Version-aware memory limit decreases. On Kubernetes 1.35+, Attune no longer clamps live memory limit decreases the way it does on 1.33–1.34 when the platform allows them and the policy uses controlledValues: RequestsAndLimits with decrease allowed. On 1.33–1.34, limit decreases remain clamped as before (#439).
• Usage floor on limit decreases. Target memory limits stay above recent usage times (1 + decreaseUsageMarginPercent/100) (default 10 percent), reducing OOM races when shrinking limits (#447).
• Language runtime profiles. Set runtimeProfile to java, python, nodejs, or golang for safe in-memory defaults (for example java uses higher memory overhead and does not enable memory decrease unless you set it). Defaults are not written back onto the CR (#440).

See Runtime profiles and Upgrading.

Capacity, pressure, and stuck resizes

• Node pressure-aware skips. Request increases are skipped when the node has MemoryPressure (memory), DiskPressure, or PIDPressure, so Attune does not pile more demand onto a stressed node. Decreases still proceed (#441).
• Capacity skip metrics and reclaimed capacity. Prometheus metrics and Grafana panels for capacity and pressure skips, plus reclaimed request capacity signals (#448).
• Deferred and Infeasible operator UX. Policy status surfaces workloads.deferred and workloads.infeasible, a ResizeBlocked condition with clear reasons, and related metrics and alerts (#436).

See Troubleshooting: Deferred or Infeasible and Bin packing.

Docs and diagnostics

• Recommendation explainability guide and skip-reason tables for kubectl attune explain and status explanations (#437).
• Standalone SLO guardrails guide (PromQL post-resize checks) linked from quickstart and nav (#470).
• GA-era messaging refresh across docs and ADOPTERS (#435).
• Pre-release E2E Nightly matrix guidance for maintainers (1.32–1.35) (#473).

Hardening and tests

• GitOps SSRF checks for pull request apiUrl, incomplete-config paths, and related unit and Chainsaw coverage (#455, #464).
• Cluster E2E for version-aware memory limit decrease, Deferred/Infeasible status injection, MemoryPressure skip with events, and java profile overhead in explanations (#471, #472).
• Full K8s 1.32–1.35 E2E Nightly matrix and fuzz coverage exercised on the release candidate tip before this tag.

Upgrade notes

• Most new capabilities are opt-in (GitOps PR automation, fleet report, runtime profiles). Existing policies keep working without YAML edits.
• On Kubernetes 1.35+, memory limits can decrease in place when the policy allows decrease and uses requests-and-limits control. If you rely on limits never going down, use memory.allowDecrease: false, a restrictive runtime profile, controlledValues: RequestsOnly (default), or stay on 1.33–1.34.
• Refresh CRDs with the release install path (helm upgrade or dist/crds.yaml / dist/install.yaml from the tag).
• Re-apply Grafana dashboard and PrometheusRule assets if you manage them out of band; new panels and alerts cover GitOps PR, fleet export, memory limit decrease safety, capacity skips, and Deferred/Infeasible.
• Upgrade kubectl attune with the release so explain shows GitOps PR and runtime profile effective fields.

kubectl attune status -A
kubectl attune explain -n <namespace> <policy>

Full upgrade detail: Upgrading.

Full changelog

v0.1.20...v0.1.21