Docker Swarm集群升级ES镜像后应用部署连接失败求助

Let's dive into why your second container can't reach Elasticsearch after upgrading from 2.4 to 5.6 in your Docker Swarm cluster. That "None of the configured nodes are available" error is almost always tied to version-specific configuration changes or Swarm network/startup timing issues. Here's how to troubleshoot and fix it:

1. Address Elasticsearch 5.x Configuration Changes #

ES 5.6 introduced critical changes that break compatibility with 2.4's defaults—these are the most likely culprits:

• Network Binding: By default, ES 5.x only binds to localhost, which means other containers on the Swarm network can't reach it. Add this environment variable to your ES service:

  environment:
    - network.host=0.0.0.0
  

  This allows ES to listen on all network interfaces in the container.

• Node Discovery: ES 5.x uses discovery.zen.ping.unicast.hosts for node discovery (replacing older mechanisms). For a single-node cluster, set this to your ES service name:

  environment:
    - discovery.zen.ping.unicast.hosts=nga_es
    - discovery.zen.minimum_master_nodes=1 # Critical for preventing split-brain; use (n/2)+1 for multi-node clusters
  

• Cluster Name Match: Double-check that your ES service's cluster.name matches what your app is using. Add this to confirm:

  environment:
    - cluster.name=elasticsearch
  

2. Fix Docker Swarm Startup Timing & Network #

ES 5.6 takes longer to initialize than 2.4, so your app container might be trying to connect before ES is ready. Plus, you need to ensure both services are on the same Swarm network:

• Add Health Checks to ES: Force Swarm to wait until ES is fully ready before starting your app. Add this to your ES service:

  healthcheck:
    test: ["CMD-SHELL", "curl -s http://localhost:9200/_cluster/health | grep -q '\"status\":\"green\"'"]
    interval: 10s
    timeout: 10s
    retries: 5
  

• Wait for Healthy ES in Your App: Update your app service to depend on the healthy state of the ES service (note: this works in Docker Compose v3.7+ for Swarm):

  depends_on:
    nga_es:
      condition: service_healthy
  

• Shared Overlay Network: Ensure both your ES and app services are attached to the same Swarm overlay network. Define it in your docker-compose.yml:

  networks:
    swarm_internal:
      driver: overlay
  

  Then add networks: [swarm_internal] to both services.

3. Verify Basic Connectivity #

If the above changes don't work, run these checks to narrow down the issue:

• Check ES Service Logs: Run docker service logs nga_es to see if ES is failing to start (common issues: memory limits, file permissions). Add ES_JAVA_OPTS=-Xms512m -Xmx512m to your ES environment if you hit memory errors.
• Test DNS Resolution: Exec into your app container and run ping nga_es—if it fails, your network setup is broken.
• Test Port Access: Run telnet nga_es 9300 from your app container. If it can't connect, confirm ES is listening on 9300 (check network.host again) and that no Swarm network policies are blocking traffic.

Example Updated docker-compose.yml Snippet #

Here's how your ES and app services should look with these fixes:

version: "3.8"

services:
  nga_es:
    image: elasticsearch:5.6
    environment:
      - cluster.name=elasticsearch
      - discovery.zen.ping.unicast.hosts=nga_es
      - discovery.zen.minimum_master_nodes=1
      - network.host=0.0.0.0
      - ES_JAVA_OPTS=-Xms512m -Xmx512m
    volumes:
      - es_data:/usr/share/elasticsearch/data
    networks:
      - swarm_internal
    healthcheck:
      test: ["CMD-SHELL", "curl -s http://localhost:9200/_cluster/health | grep -q '\"status\":\"green\"'"]
      interval: 10s
      timeout: 10s
      retries: 5
    deploy:
      replicas: 1
      restart_policy:
        condition: on-failure

  your_app:
    image: your-app-image:latest
    environment:
      - ES_HOST=nga_es
      - ES_PORT=9300
      - ES_CLUSTER_NAME=elasticsearch
    networks:
      - swarm_internal
    depends_on:
      nga_es:
        condition: service_healthy
    deploy:
      restart_policy:
        condition: on-failure

networks:
  swarm_internal:
    driver: overlay

volumes:
  es_data:
    driver: local

内容的提问来源于stack exchange，提问作者thechmodmaster