ݸਓతʹ໘ന͔ͬͨηογϣϯ
AutoScalerɺObservabilityɺPerformanceʹภͬͯ·͢
‣ Autoscaling at Scale: How We Manage Capacity
‣ Kubernetes On Cgroup v2
‣ Scaling Kubernetes Networking Beyond 100k Endpoints
‣ Do The Math: Auto-Scaling Applications with Kubernetes
‣ Hubble - eBPF Based Observability for Kubernetes
‣ eBPF and Kubernetes: Little Helper Minions for Scaling Microservices
‣ Standardizing Applications For the Cloud at a Global Scale
‣ Virtual Cluster - A Practical Kubernetes Hard Multi-tenancy Solution
‣ Deep Dive into Autoscaling
‣ Better Histograms for Prometheus
‣ Dynamic Configuration with ComponentConfig and the Control Loop
Slide 14
Slide 14 text
Autoscaling at Scale: How We Manage Capacity
Zalandoࣾͷӡ༻Λͱʹͨ͠HPA/CA/VPAͷҙͱ৽ػೳͷհ
‣ HPA/CA/VPAͷ֓ཁΛͬ͘͟Γհ
‣ HPAͰRPSΧελϜϝτϦΫεΛ
ར༻͢ΔͨΊAdapterΛࣗ࡞
https://github.com/zalando-incubator/kube-metrics-adapter
‣ metrics.typeʹContainerResource͕
Ճ͞ΕΔ (v1.20ʹೖΓͦ͏ͳงғؾ)
‣ VPAv0.7.1Λforkͯ͠OOMKillϋϯυ
ϦϯάΛվળͨ͠ͷΛ։ൃ͍ͯ͠Δ
https://github.com/zalando-incubator/autoscaler