Rolling Deployment dengan Docker Swarm

Strategi zero-downtime deployment menggunakan Docker Swarm rolling update dan auto-rollback

Rolling deployment dengan Docker Swarm memastikan rilis versi baru tanpa downtime dengan memulai container baru terlebih dahulu, menunggu health check lolos, kemudian menghentikan container lama secara bertahap.

Alur Kerja Rolling Deployment

State Machine Deployment

Pipeline CI/CD

Konfigurasi Rolling Update

Docker Swarm rolling update dikonfigurasi per service di docker-compose.swarm.yml:

services:
  frontend:
    image: ghcr.io/yomu/frontend:latest
    deploy:
      replicas: 1
      update_config:
        parallelism: 1        # Update 1 replica sekaligus
        delay: 15s            # Tunggu 15 detik antar update
        order: start-first    # Start baru sebelum stop lama
        failure_action: rollback  # Auto-rollback jika health check gagal
      restart_policy:
        condition: on-failure
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:3000/health"]
      interval: 10s
      timeout: 5s
      retries: 3
      start_period: 30s

Parameter Update Config

Parameter	Value	Deskripsi
`parallelism`	`1`	Jumlah replica yang diupdate bersamaan
`delay`	`15s`	Jeda waktu antar update replica
`order`	`start-first`	Start container baru sebelum stop yang lama
`failure_action`	`rollback`	Auto-rollback ke versi sebelumnya jika gagal

Health Check per Service

Service	Health Check Endpoint	Interval	Timeout
frontend	`GET /health`	10s	5s
java	`GET /actuator/health`	10s	5s
rust	`GET /health`	10s	5s

CD Trigger

Deployment otomatis dipicu oleh GitHub Actions saat push ke branch main atau staging:

# .github/workflows/deploy.yml
jobs:
  deploy:
    runs-on: ubuntu-latest
    steps:
      - uses: appleboy/ssh-action@master
        with:
          host: ${{ secrets.VPS_HOST }}
          username: ${{ secrets.VPS_USER }}
          key: ${{ secrets.VPS_SSH_KEY }}
          script: |
            cd /opt/yomu
            docker stack deploy -c docker-compose.swarm.yml yomu

Alur Deployment

Push code ke branch main (production) atau staging (staging)
GitHub Actions build Docker image dan push ke GHCR
SSH ke VPS via appleboy/ssh-action
docker stack deploy deploy stack dengan konfigurasi baru
Swarm rolling update start container baru → health check → delay 15s → stop container lama
Auto-rollback jika health check gagal

Auto-Rollback Monitor

Script auto-rollback-monitor.sh memantau metrik deployment via Prometheus dan trigger rollback manual jika error rate melebihi threshold:

#!/bin/bash
# auto-rollback-monitor.sh

ERROR_RATE=$(curl -s 'http://prometheus:9090/api/v1/query?query=rate(traefik_service_requests_total{code=~"5.."}[5m])' | jq '.data.result[0].value[1]')
P95_LATENCY=$(curl -s 'http://prometheus:9090/api/v1/query?query=histogram_quantile(0.95, rate(traefik_service_request_duration_seconds_bucket[5m]))' | jq '.data.result[0].value[1]')

if (( $(echo "$ERROR_RATE > 0.05" | bc -l) )); then
    echo "ERROR: Error rate > 5% ($ERROR_RATE)"
    docker service rollback yomu_frontend
    docker service rollback yomu_java
    docker service rollback yomu_rust
    exit 1
fi

if (( $(echo "$P95_LATENCY > 2.0" | bc -l) )); then
    echo "ERROR: P95 latency > 2s ($P95_LATENCY)"
    docker service rollback yomu_frontend
    docker service rollback yomu_java
    docker service rollback yomu_rust
    exit 1
fi

echo "OK: Deployment healthy"

Threshold Monitoring

Metrik	Threshold	Aksi
Error Rate	> 5% (5xx responses)	Rollback semua service
P95 Latency	> 2 detik	Rollback semua service

Skrip Deployment

Skrip	Fungsi
`deploy-staging.sh`	Deploy staging environment via SSH
`health-check.sh`	Verifikasi health check semua service
`auto-rollback-monitor.sh`	Monitor error rate dan latency, trigger rollback
`smoke-test-subdomains.sh`	Test routing semua subdomain setelah deploy

Contoh deploy-staging.sh

#!/bin/bash
# deploy-staging.sh

ssh ${VPS_USER}@${VPS_HOST} << 'EOF'
  cd /opt/yomu
  docker stack deploy -c docker-compose.swarm.yml yomu
  ./health-check.sh
  ./smoke-test-subdomains.sh
EOF

Manfaat Rolling Deployment dengan Swarm

Manfaat	Deskripsi
Zero Downtime	Container baru start dulu sebelum yang lama stop (`start-first`)
Auto-Rollback	Swarm built-in rollback jika health check gagal
Gradual Update	Update 1 replica dengan delay 15s untuk stabilitas
Health Check Native	Integrasi health check per service di Docker
Sederhana	Satu command `docker stack deploy`, tidak perlu switch manual
Resource Efisien	Tidak perlu duplikasi environment seperti blue-green

Praktik Terbaik

Pastikan health check terdefinisi untuk semua service dengan timeout yang sesuai
Gunakan delay 15s untuk memberi waktu stabilisasi container baru
Monitor error rate dan latency selama 5-10 menit setelah deploy
Simpan image tag di GHCR untuk rollback manual jika diperlukan
Jalankan smoke test setelah deployment untuk verifikasi routing

Rolling Deployment dengan Docker Swarm

On this page