ECS实例0-1伸缩咨询：单每日任务实例自动启停实现疑问

Your scenario is absolutely feasible with ECS, but it relies on setting up capacity providers and auto-scaling groups correctly to link task demand with EC2 instance lifecycle. Here's a step-by-step breakdown of the right implementation, plus common issues to troubleshoot:

Step 1: Configure an Auto Scaling Group (ASG) for ECS #

First, you need an ASG that can scale from 0 to 1 instances (since you only run one task at a time):

• Use an ECS-optimized AMI (Amazon Linux 2 or 2023) — these come pre-configured with the ECS agent.
• Add user data to ensure instances join your ECS cluster:

  echo ECS_CLUSTER=your-cluster-name >> /etc/ecs/ecs.config
  

• Set ASG parameters:
  • Minimum capacity: 0
  • Desired capacity: 0
  • Maximum capacity: 1
• Disable termination protection (you want instances to shut down automatically when no tasks are running).

Step 2: Create an ECS Capacity Provider #

This is the critical link between your ECS service and the ASG:

• Go to your ECS cluster → Capacity Providers → Create capacity provider.
• Link it to the ASG you just created.
• Enable Managed Scaling (this lets ECS control the ASG's desired capacity based on task demand).
• Set Managed Termination Protection to ENABLED — this prevents ECS from terminating instances that are still running tasks.
• Set Target Capacity Percentage to 100% — ensures ECS scales the ASG exactly to meet the required task capacity.

Step 3: Update Your ECS Cluster & Service #

• Associate the new capacity provider with your ECS cluster (set it as the default if this is your primary workload).
• For your ECS service:
  • Set initial desired task count to 0.
  • Ensure the service uses the capacity provider you created (check under the service's Capacity provider strategy settings).

Step 4: Automate the Task Schedule (Optional but Recommended) #

Instead of manually adjusting the desired task count, use EventBridge to automate this:

• Create two EventBridge rules:
  • Rule 1: Trigger daily at your start time, use the UpdateService API action to set desired task count to 1.
  • Rule 2: Trigger daily 2 hours later, use UpdateService to set desired task count back to 0.
• Make sure the EventBridge execution role has permissions to call ecs:UpdateService on your service.

Common Issues & Troubleshooting #

ASG Doesn’t Scale Up When Setting Desired Tasks to 1 #

• Check capacity provider links: Verify the capacity provider is associated with both the cluster and service. Go to ECS Cluster → Capacity Providers to confirm.
• Inspect ASG activity history: In EC2 → Auto Scaling Groups → Activity tab, look for errors like insufficient instance capacity, launch template issues, or IAM role problems.
• Validate ECS agent health: When an instance launches, check /var/log/ecs/ecs-agent.log on the instance for errors joining the cluster.
• Match task resources to instance type: Ensure your task’s CPU/memory requirements don’t exceed the ASG instance type’s capacity (e.g., a t2.micro can’t run a task needing 2 vCPUs).

ASG Doesn’t Scale Down After Setting Desired Tasks to 0 #

• Wait for task termination: Managed termination protection means ECS won’t shut down instances until all tasks are in the STOPPED state. Confirm your task completed successfully and isn’t stuck.
• Check ASG cooldown: The default cooldown is 300 seconds (5 minutes). Give it time after setting desired count to 0 before expecting scaling down.
• Verify target capacity percentage: If it’s set below 100%, ECS might leave extra instances running. Ensure it’s set to 100%.

Task Fails to Launch on the EC2 Instance #

• Check CloudWatch logs: Look for errors like image pull failures, missing IAM permissions, or port conflicts in the task’s CloudWatch log group.
• Validate instance IAM role: The EC2 instance role needs permissions to pull images from ECR, send logs to CloudWatch, and communicate with ECS.
• Review security groups: Ensure the instance’s security group allows outbound traffic to ECS endpoints, ECR, and any services your task depends on.

Manual Desired Count Adjustments Don’t Work #

• Confirm service/cluster context: Double-check you’re modifying the correct service in the right cluster.
• Check deployment status: Go to ECS Service → Deployments tab to see if there are failed deployments or configuration errors blocking the change.

内容的提问来源于stack exchange，提问作者rix