Distributed GPU Cluster Infrastructure

EQ Cloud & Cluster

Scale inference across multi-region GPU compute nodes with automated 15-second heartbeats, circuit breaker edge gateways (:9000), and Kubernetes HPA auto-scaling.

k8s/eq-cluster.yamlyaml

apiVersion: apps/v1
kind: Deployment
metadata:
  name: eq-worker-pool
spec:
  replicas: 4
  selector:
    matchLabels:
      app: eq-worker
  template:
    metadata:
      labels:
        app: eq-worker
    spec:
      containers:
      - name: worker
        image: everestq/engine:v1.0.0
        args: ["eq-worker", "--cloud-api", "http://eq-cloud:8080"]
        resources:
          limits:
            nvidia.com/gpu: 1

Edge Gateway (:9000)

Circuit breaker request routing with tracing headers, rate-limiting, and load shedding.

Cloud Control Plane

Multi-tenant workspace organization management, API key rotation, and GPU-second metering.

Cluster Heartbeats

Continuous 15s node health pinging with node draining and failover for zero downtime.