5-Spot Scheduler: Cut Bare Metal Kubernetes Costs by 70%

The problem nobody talks about: 24/7 bare metal

Cloud has autoscaling. Kubernetes has Horizontal Pod Autoscaler. But what about physical servers? Bare-metal machines? Edge infrastructure?

They run 24/7. Even when nobody is using them. Even on weekends. Even at 3 AM when the only traffic is monitoring pinging itself.

For financial services firms, this is a massive cost sink. Trading platforms need full capacity during market hours — but those same machines sit idle 16 hours a day and all weekend. Development clusters burn electricity around the clock for workloads that only run during business hours.

Organizations typically see 40-70% cost reduction potential for non-critical workloads, yet lack tools designed for time-based, proactive scheduling of physical infrastructure.

Enter 5-Spot Machine Scheduler.

What is 5-Spot Machine Scheduler?

5-Spot is a Kubernetes operator being contributed to FINOS — the Fintech Open Source Foundation. Born from real-world needs at RBC (Royal Bank of Canada), it provides intelligent, time-based scheduling for physical machine deployments.

Think of it as a cron job for your entire bare-metal fleet — but smarter, Kubernetes-native, and with enterprise-grade lifecycle management.

Core capabilities

• Proactive time-based scheduling — automatically provisions and de-provisions physical machines based on configurable day-of-week and hour-of-day schedules with timezone support
• Physical infrastructure focus — built specifically for bare-metal servers, edge devices, and hybrid cloud deployments, not just cloud VMs
• Kubernetes-native — leverages operator patterns with Custom Resource Definitions (CRDs), enabling Infrastructure-as-Code and GitOps workflows
• Graceful lifecycle management — configurable graceful shutdown periods for workload migration and emergency kill-switch functionality
• Enterprise features — multi-instance HA, priority-based resource allocation, comprehensive observability
• Cluster API and k0smotron integration — works with the Kubernetes cluster lifecycle ecosystem

Why this matters for financial services

Financial services has unique infrastructure challenges that 5-Spot directly addresses:

Market-hours workloads

Trading systems, analytics pipelines, and risk calculations follow strict business-hour patterns. Markets open at 9:30, close at 16:00. The infrastructure to support these workloads does not need to run from midnight to 6 AM.

With 5-Spot, you define a schedule:

# Example: Trading infrastructure active during market hours
schedule:
  timezone: America/New_York
  windows:
    - days: [monday, tuesday, wednesday, thursday, friday]
      start: "08:00"
      end: "18:00"
  gracefulShutdownMinutes: 30

The operator handles provisioning before markets open and graceful de-provisioning after they close — including a 30-minute buffer for end-of-day processing.

Compliance and audit trails

Financial regulators require auditable infrastructure management. Because 5-Spot uses Kubernetes CRDs, every scheduling decision is:

• Declarative and version-controlled (GitOps)
• Auditable through Kubernetes event logs
• Policy-driven with RBAC controls
• Reproducible across environments

Development and testing environments

Banks run massive development clusters that mirror production. These environments rarely need to be online outside business hours, yet they often run 24/7 because nobody built the automation to turn them off safely.

5-Spot solves this with zero custom scripting.

The FINOS ecosystem fit

5-Spot fits into FINOS’s growing grid computing ecosystem. It complements projects like OpenGRIS (task scheduling) by handling the compute scheduling layer — ensuring physical machines are available when tasks need to run and powered down when they do not.

The FINOS maintainers of OpenGRIS have already provided positive feedback on the integration potential. Together, these projects enable:

1. OpenGRIS schedules tasks to optimal time windows
2. 5-Spot ensures the physical infrastructure is available during those windows
3. Cost savings compound because both task scheduling and compute scheduling are optimized

Beyond financial services

While born in financial services, 5-Spot’s value extends to any organization with:

• Predictable workload patterns — CI/CD clusters, batch processing, analytics pipelines
• Edge infrastructure — retail stores, manufacturing floors, remote offices with compute that follows business hours
• Hybrid cloud — bare-metal clusters that complement cloud resources for cost-sensitive workloads
• Sustainability goals — reducing energy consumption by 40-70% for non-critical infrastructure

This aligns with the carbon-aware computing movement and broader sustainability efforts in the cloud-native ecosystem.

Technical architecture

5-Spot follows standard Kubernetes operator patterns:

1. Custom Resource Definition (CRD) defines the scheduling policy — machines, time windows, grace periods, priorities
2. Controller watches CRDs and reconciles state — provisioning machines before scheduled windows, initiating graceful shutdown after
3. Cluster API integration manages the actual machine lifecycle through the Kubernetes-native infrastructure abstraction
4. k0smotron support enables lightweight cluster management for edge and hybrid scenarios

The operator supports:

• Multiple schedules per machine group
• Priority-based resource allocation when machines are shared
• Emergency kill-switch for immediate de-provisioning
• Multi-instance deployment for high availability
• Prometheus metrics for observability

How this connects to multi-tenant GPU infrastructure

At KubeCon Europe 2026, I presented on multi-tenant GPUs on bare metal. The challenge is the same: physical infrastructure is expensive, and sharing it safely across teams requires intelligent scheduling.

5-Spot addresses the time dimension — when machines should be active. Combined with GPU sharing technologies like MIG, MPS, and time-slicing, you get both temporal and spatial optimization of bare-metal resources.

Imagine a GPU cluster that:

• Runs full capacity during business hours for model training
• Scales to 30% capacity after hours for batch inference
• Powers down completely on weekends except for critical production serving

The Safe, Fair, Efficient framework applies directly: Safe scheduling with graceful shutdowns, Fair allocation with priority policies, Efficient utilization through time-based provisioning.

Getting involved

5-Spot is currently going through the FINOS onboarding process. The contribution follows five stages:

1. ✅ Describing the contribution (complete)
2. 🔄 Approval by FINOS TOC (in progress)
3. ⬜ Preparing for onboarding
4. ⬜ FINOS infrastructure onboarding
5. ⬜ Public announcement

If your organization runs bare-metal infrastructure with predictable usage patterns, this project is worth watching. The potential for 40-70% cost reduction on non-critical workloads is significant — and having it open-sourced under FINOS governance means financial services firms can adopt it with confidence.

Related: Multi-tenant GPUs on bare metal, Carbon-aware Kubernetes scheduling, KubeCon 2026 numbers. Need help with bare-metal Kubernetes infrastructure? Book a consultation.