Optimize Kubernetes

The Problem

Everybody knows their Kubernetes clusters are overprovisioned, so why aren’t they sized and scaling correctly?

• Kubernetes is complex Requiring engineers to understand multiple (shared) layers of infrastructure
• Optimization workflows are slow and frustrating. Engineers want to ship code and stay out of configs - not play guess and check with configs and dashboards.
• Autoscaling takes time to master. Picking the right autoscaler, deploying and tuning for every workload & cluster is a full-time job.
• Vendors require invasive permissions to automate sizing recommendations and realize cost savings.

Using Flightcrew to Optimize Kubernetes

Flightcrew is a config copilot that optimizes Kubernetes for availability and cost without invasive permissions.

• Flightcrew monitors, forecasts and evaluates workloads and infrastructure
• Engineers are alerted when
  • Workloads are failing, at-risk or wasteful
  • A configuration change will have a negative impact on their workload, or a neighboring workload
• Flightcrew generates GitHub PRs with the correct configuration.
  • Pod and Node Sizing
  • Resource Lifecycle
  • Pod and Node Autoscaling (including KEDA and Karpenter)
• Engineers review and deploy Flightcrew recommendations through standard CI/CD processes

Why Flightcrew

Flightcrew uses a holistic understanding of Kubernetes workloads to generate personalized sizing and scaling recommendations for every workload:

• Flightcrew recommendations understand architecture, framework and dependencies (ex: customScaler: QueueSize for QueueConsumer).
• Flightcrew understands resource lifecycles, pod and node interactions and autoscaling behavior
• Flightcrew uses sophisticated AI/ML to understand workload utilization across time periods, replicas and clusters.
• Recommendations are surfaced and approved through native engineering workflows: no invasive permissions and no unreviewed changes
• Flightcrew makes Kubernetes accessible and self-serve through guardrails, AI explanations and conversation

Results

• Save >40% in cloud costs from personalized workload, node and autoscaling recommendations, deployed through safe, GitOps workflows
• Save >10% of engineering time by simplifying Kubernetes resource management
• Prevent incidents from resource starvation and node bullying