GitHub Repository

SpotVortex is an open-source Kubernetes operator for one hard problem: getting Spot savings without gambling on reliability.

0 starsGo

SpotVortex helps SRE teams push more Kubernetes capacity onto Spot

Name: SpotVortex helps SRE teams push more Kubernetes capacity onto Spot
Availability: InStock
Author: softcane

by softcane·Mar 12, 2026·1 point·2 comments

Visit Project View on HN

AI Analysis

●●SolidNiche GemShip It

Transition-aware TFT model tracks market hazard before scaling Spot node pools.

Strengths

•Deterministic mode with 10-minute control cadence gives predictable, auditable behavior.
•Tracks PDB slack, zone spread, and stateful workload mix before adjusting Spot ratios.
•RL runs in shadow mode only, recording telemetry without actuating production changes.

Weaknesses

•AWS-only support limits adoption for multi-cloud Kubernetes deployments.
•Spot instance management already has Karpenter and cluster-autoscaler solutions.

Post Description

I built SpotVortex for a simple reason: most teams want more Spot savings, but they do not want to compromise service safety to get them.

SpotVortex runs inside the cluster and controls Spot at the node-pool level. It watches market risk and pool health, then decides when a pool should grow, spot, hold, freeze, or move back toward On-Demand.

A few concrete details about the shipped runtime:

* The deterministic mode is the active path * The control cadence is 10 minutes * Decisions stay at the node-pool level, not the pod level * Workload data stays in-cluster.

The controller also looks at pool blast radius, not just price. It tracks things like:

* PDB Slack under node loss * How much critical workload is sitting on the Spot restart and recovery time * Stateful workload mix * zone spread * evictability * available On-Demand headroom * Current AWS support is explicit. The shipped model scope covers 60 instance families across:

compute: c5, c6, c7 general purpose: m5, m6, m7 memory optimized: r5, r6, r7 burstable: t2, t3, t4g

Happy to answer hard questions.