基于AWS CodeDeploy与负载均衡的多应用多端口部署问题咨询

Alright, let's break down how to make this work for your multi-port EC2 setup with AWS CodeDeploy and a load balancer—no dropped requests guaranteed. Here's a step-by-step approach tailored to your 4 instances running 3 apps each on ports 81/82/83:

First, you need to map each app to its own target group (since each runs on a unique port):

• Create 3 target groups: app1-tg (port 81), app2-tg (port 82), app3-tg (port 83)
• Add all 4 EC2 instances to each target group (each instance will be registered with its respective port for each group)
• For each target group:
  • Set a drain timeout (under Target Group Attributes) to 60-120 seconds—this gives the load balancer time to finish processing in-flight requests before stopping traffic to the instance
  • Configure health checks to match each app's health endpoint (e.g., /health on port 81 for app1)

When setting up your CodeDeploy deployment group:

• Select EC2/On-premises as the platform
• Under "Load balancing", check "Enable load balancing" and select your load balancer type (ALB/NLB)
• Add all 3 target groups you created earlier—this tells CodeDeploy to manage instance registration/deregistration across all groups for your multi-port apps
• Choose a rolling deployment strategy (e.g., CodeDeployDefault.HalfAtATime or a custom rule) to ensure only a subset of instances are taken down at once, keeping the rest online to handle traffic
• Ensure your EC2 instances have the CodeDeploy agent installed and an IAM role with permissions for elasticloadbalancing:RegisterTargets, elasticloadbalancing:DeregisterTargets, and CodeDeploy operations

Your appspec.yml will orchestrate the instance drain/register process across all three ports. Here's a working template:

version: 0.0
os: linux
hooks:
  # Deregister instance from all target groups before installing new code
  BeforeInstall:
    - location: scripts/deregister_all_targets.sh
      timeout: 300
      runas: root
  # Start your updated app processes
  ApplicationStart:
    - location: scripts/start_all_apps.sh
      timeout: 300
      runas: root
  # Re-register instance once apps are healthy
  AfterApplicationStart:
    - location: scripts/register_all_targets.sh
      timeout: 300
      runas: root

Example Scripts #

deregister_all_targets.sh

This script pulls the current instance ID and removes it from all three target groups, then waits for connections to drain:

#!/bin/bash
set -e

# Get current EC2 instance ID
INSTANCE_ID=$(curl -s http://169.254.169.254/latest/meta-data/instance-id)

# Replace these ARNs with your actual target group ARNs
TG_ARN_APP1="arn:aws:elasticloadbalancing:us-east-1:YOUR_ACCOUNT_ID:targetgroup/app1-tg/YOUR_TG_ID"
TG_ARN_APP2="arn:aws:elasticloadbalancing:us-east-1:YOUR_ACCOUNT_ID:targetgroup/app2-tg/YOUR_TG_ID"
TG_ARN_APP3="arn:aws:elasticloadbalancing:us-east-1:YOUR_ACCOUNT_ID:targetgroup/app3-tg/YOUR_TG_ID"

# Deregister from each target group with the correct port
aws elbv2 deregister-targets --target-group-arn $TG_ARN_APP1 --targets Id=$INSTANCE_ID,Port=81
aws elbv2 deregister-targets --target-group-arn $TG_ARN_APP2 --targets Id=$INSTANCE_ID,Port=82
aws elbv2 deregister-targets --target-group-arn $TG_ARN_APP3 --targets Id=$INSTANCE_ID,Port=83

# Wait for drain timeout (match your target group's drain setting)
sleep 90

register_all_targets.sh

This script re-registers the instance to all target groups once apps are running:

#!/bin/bash
set -e

INSTANCE_ID=$(curl -s http://169.254.169.254/latest/meta-data/instance-id)

TG_ARN_APP1="arn:aws:elasticloadbalancing:us-east-1:YOUR_ACCOUNT_ID:targetgroup/app1-tg/YOUR_TG_ID"
TG_ARN_APP2="arn:aws:elasticloadbalancing:us-east-1:YOUR_ACCOUNT_ID:targetgroup/app2-tg/YOUR_TG_ID"
TG_ARN_APP3="arn:aws:elasticloadbalancing:us-east-1:YOUR_ACCOUNT_ID:targetgroup/app3-tg/YOUR_TG_ID"

# Register instance with each app's port
aws elbv2 register-targets --target-group-arn $TG_ARN_APP1 --targets Id=$INSTANCE_ID,Port=81
aws elbv2 register-targets --target-group-arn $TG_ARN_APP2 --targets Id=$INSTANCE_ID,Port=82
aws elbv2 register-targets --target-group-arn $TG_ARN_APP3 --targets Id=$INSTANCE_ID,Port=83

start_all_apps.sh

Customize this to start your three app processes (example for systemd-managed apps):

#!/bin/bash
set -e

# Stop existing app processes first
systemctl stop app1.service app2.service app3.service

# Deploy new code (e.g., unzip your artifact)
unzip /tmp/deployment-package.zip -d /opt/apps/

# Start updated apps
systemctl start app1.service app2.service app3.service

# Wait for apps to become healthy (adjust based on your app's startup time)
sleep 10

• Monitor the CodeDeploy console to confirm instances are being deregistered, updated, and re-registered correctly
• Check your load balancer's target group metrics (in CloudWatch) to ensure instances transition from draining to healthy without gaps
• Test traffic during deployment using your load balancer's DNS name to verify no requests are dropped

内容的提问来源于stack exchange，提问作者Mahantesh M Ambi