Source: containerjournal.com. Original announcement made at OSCON 2019.
Carbon Relay Launches Red Sky Ops for Kubernetes Optimization
At the Open Source and Software Development (OSCON) conference in July 2019, Carbon Relay announced Red Sky Ops, a platform that applies machine learning algorithms to optimize resource allocation in Kubernetes environments. The platform targets a core operational pain point: the sheer combinatorial complexity of tuning Kubernetes clusters at scale.
Why Kubernetes Needs AIOps
Carbon Relay CEO Matt Provo argues that the configuration surface area of a Kubernetes environment is simply too large for human operators to manage manually. With potentially billions of combinations of settings across namespaces, resource requests, limits, scheduling policies, and workload requirements, even experienced platform engineering teams are forced to make trade-offs.
One common consequence is over-provisioning: many organizations run a single application per cluster to reduce the risk of resource contention. While this simplifies configuration, it accelerates cluster sprawl, increasing both infrastructure costs and operational overhead across the enterprise.
How Red Sky Ops Works
Red Sky Ops follows a study-then-optimize workflow:
- Study and replicate: ML algorithms first observe and replicate the application environment.
- Stress-test: The platform simulates load conditions to understand performance boundaries.
- Optimize proactively: Once trained, the algorithms determine optimal configurations, schedules, and resource allocations before problems occur.
- Continuous recalculation: Red Sky Ops works alongside the Kubernetes scheduler, continuously recalculating parameters as cluster conditions change — reducing alert noise in the process.
The platform also factors in service requirements, hardware and software policy constraints, workload-specific deadlines, and scheduling rules when generating its recommendations.
Open Source vs. Enterprise Editions
Red Sky Ops was made available in two tiers:
| Feature | Open Source | Enterprise |
|---|---|---|
| Kubernetes load balancer | ✓ | ✓ |
| Controller & API services | ✓ | ✓ |
| Authentication services | ✓ | ✓ |
| Deep reinforcement learning | ✗ | ✓ |
| Automatic application configuration | ✗ | ✓ |
| Data sharing | ✗ | ✓ |
| Advanced automation & scheduling | ✗ | ✓ |
The enterprise edition’s deep reinforcement learning capability is particularly significant: it allows the AI agent to continue training on live production data, improving its recommendations over time rather than relying solely on an initial training run.
Claimed Performance and Cost Benefits
Carbon Relay stated that Red Sky Ops can deliver:
- Up to 50% improvement in operational Kubernetes performance
- Up to 30% reduction in application management costs
These figures are vendor-claimed and were not independently verified at the time of the announcement.
The Broader AIOps Trend in Platform Engineering
The launch of Red Sky Ops reflects a wider industry reality: Kubernetes is an exceptionally capable platform, but it was designed by and for engineers with deep infrastructure expertise. Most enterprise IT organizations still rely on administrators whose primary skills are operational rather than software-development oriented. Bridging that gap at scale requires one of two approaches:
- Hiring dedicated platform engineers to manage Kubernetes infrastructure, or
- Adopting AIOps tooling that abstracts optimization complexity away from day-to-day operations.
Neither path eliminates the need for human judgment. AI models require training data, validation, and ongoing oversight. As application environments grow more dynamic — with workloads scaling up and down constantly, new services being deployed continuously, and policy requirements shifting — the feedback loop between human operators and AI systems becomes the critical operational primitive.
AIOps as an Augmentation Tool, Not a Replacement
A common concern when AI tooling is introduced into operations workflows is that it poses a threat to existing team roles. In practice, however, the scale at which modern containerized applications operate means that human operators are already overwhelmed by the volume of signals and decisions required. AIOps platforms like Red Sky Ops are better understood as force multipliers: they handle the high-frequency, data-intensive optimization decisions, freeing human engineers to focus on architecture, policy, and the edge cases that AI models handle poorly.
Building organizational trust in AI-driven recommendations does take time. Ops teams that start with Red Sky Ops in advisory mode — reviewing suggested configurations before they are applied — can gradually increase automation as confidence in the model improves.
Summary
Red Sky Ops represents an early but directionally important example of applying reinforcement learning to Kubernetes resource management. As Kubernetes adoption has matured since 2019, the problems Carbon Relay identified — cluster sprawl, over-provisioning, and configuration complexity — have only become more acute in large-scale deployments. AIOps tooling that can close the loop between observation, learning, and automated remediation will remain a significant area of investment for platform engineering teams.