Skip to main content

Deployment Reference

ICE is designed for containerized orchestration in high-concurrency environments.

1. Docker Compose​

For single-node or development deployments, use the following docker-compose.yaml pattern.

version: '3.8'
services:
  ice-kernel:
    image: registry.dopove.com/ice/kernel:v2.7.755
    ports:
      - "8000:8000"
    environment:
      - DATABASE_URL=postgresql://user:pass@db:5432/ice
      - REDIS_URL=redis://cache:6379
      - ICE_LICENSE_JWT=${ICE_LICENSE_JWT}
    deploy:
      resources:
        limits:
          cpus: '4'
          memory: 16G
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:8000/v1/health"]
      interval: 30s
      timeout: 10s
      retries: 3

  db:
    image: ankane/pgvector:latest
    environment:
      - POSTGRES_DB=ice
      - POSTGRES_PASSWORD=pass

  cache:
    image: redis:7-alpine

2. Kubernetes​

For enterprise-scale deployments, use the following manifest patterns.

Deployment Manifest​

apiVersion: apps/v1
kind: Deployment
metadata:
  name: ice-kernel
spec:
  replicas: 3
  selector:
    matchLabels:
      app: ice
  template:
    metadata:
      labels:
        app: ice
    spec:
      containers:
      - name: ice
        image: registry.dopove.com/ice/kernel:v2.7.755
        resources:
          limits:
            cpu: "4"
            memory: "16Gi"
          requests:
            cpu: "2"
            memory: "8Gi"
        livenessProbe:
          httpGet:
            path: /v1/health
            port: 8000

ConfigMap (Resource Tuning)​

apiVersion: v1
kind: ConfigMap
metadata:
  name: ice-config
data:
  ICE_MEMORY_CAP_GB: "14"
  ICE_MAX_STITCH_CONCURRENCY: "32"

3. Resource Management​

  • Liveness Probes: Mandatory for auto-recovery in Kubernetes.
  • Horizontal Pod Autoscaling (HPA): Scale nodes based on CPU utilization or transaction velocity.
  • Volume Mounts: Use high-performance SSD-backed PVs for the ICE_UPLOAD_DIR to minimize ingestion latency.