September 2026 - V1.21.0

✨✨✨ This release makes workload and node optimization more predictable across AWS, GKE, and AKS. It adds storage-aware node provisioning, clearer and safer rebalance execution, better controls for batch workloads, richer audit histories, and more precise Workload Autoscaler policies.

🚀 Highlights

More Predictable Workload Sizing

Workload Autoscaler policies can now set an optional minimum JVM Heap Xms. When the calculated initial Heap is below that floor, CloudPilot AI raises the Heap and Memory recommendation together while continuing to enforce the configured policy limits and readiness checks. See Minimum Heap Xms.

ResourceStartupBoost can now use a dedicated startup RecommendationPolicy instead of deriving startup resources from the steady-state policy. Startup CPU and Memory targets are evaluated independently, while ordinary optimization waits until both the steady-state and startup recommendations are ready. See ResourceStartupBoost.

Storage-Aware Node Provisioning

AWS root EBS volumes, GCP boot disks, and Azure OS disks can now use dynamic sizing policies based on vCPU count and NodeClaim ephemeral-storage demand. Base, per-vCPU, and maximum size controls let each new node receive enough system-disk capacity without assigning the largest size to every machine type.

GKE Local SSD configuration adds an Automatic mode. CloudPilot AI selects the smallest valid attached-disk count that satisfies each NodeClaim’s ephemeral-storage demand while preserving the existing bundled-only and fixed-count modes.

Safer, More Observable Rebalancing

Rebalance progress now follows target readiness, drainable Pod progress, and remaining source nodes. Proactive replacement paths coordinate Agent-observed drain progress before source-node termination, making it easier to distinguish an active drain from a stalled or incomplete operation.

NodePools dedicated to Jobs and other eligible run-to-completion Pods can now configure a minimum active node count and a provisioning delay. These controls steer new work toward a smaller active set and allow short bursts to reuse capacity released by completing work before another node is created. The feature is available for AWS, GKE, and AKS NodePools.

Audit Histories for Automation and Operations

The Workload Autoscaler Audit Log records a bounded, best-effort timeline of recommendations, planned and attempted changes, Kubernetes acceptance, observed results, convergence, and OOM-related facts. Collection is isolated from optimization, so an unavailable timeline does not stop workload updates. See Workload Autoscaler Audit Log.

Organization owners can use SaaS Operation Audit Logs to review selected state-changing user operations and organization-scoped MCP tool calls. The newest-first history includes bounded filters and allowlisted details while keeping sensitive request and response data out of stored events. See SaaS Operation Audit Logs.

⚙️ Enhancements

Add optional minimum JVM Heap Xms enforcement and a dedicated startup RecommendationPolicy for ResourceStartupBoost.
Add dynamic AWS root EBS, GCP boot disk, and Azure OS disk sizing policies.
Add demand-aware automatic GCP Local SSD count selection while retaining bundled-only and fixed-count modes.
Add three-stage rebalance progress reporting and Agent-observed proactive drain coordination.
Add batch workload NodePool controls for minimum active nodes and delayed provisioning on AWS, GKE, and AKS.
Add bounded Workload Autoscaler activity collection and workload-level Audit Log views.
Add owner-only SaaS Operation Audit Logs for selected user and MCP operations.
Extend PodMutation with native node affinity matching, key-based or complete affinity removal, and safe affinity composition. See Use PodMutation During a NodePool Merge.

These updates improve resource sizing, storage efficiency, rebalance safety, workload placement, and operational visibility across CloudPilot AI. For questions or support, join our Slack community

Stay tuned for more updates! 🚀