Load balancer types

Elastic Load Balancing supports the following types of load balancers: Application Load Balancers, and Network Load Balancers. Amazon ECS services can use these types of load balancer. Application Load Balancers are used to route HTTP/HTTPS (or Layer 7) traffic. Network Load Balancers and Classic Load Balancers are used to route TCP (or Layer 4) traffic.

Application Load Balancer

An Application Load Balancer makes routing decisions at the application layer (HTTP/HTTPS), supports path-based routing, and can route requests to one or more ports on each container instance in your cluster. Application Load Balancers support dynamic host port mapping. For example, if your task's container definition specifies port 80 for an NGINX container port, and port 0 for the host port, then the host port is dynamically chosen from the ephemeral port range of the container instance (such as 32768 to 61000 on the latest Amazon ECS-optimized AMI). When the task is launched, the NGINX container is registered with the Application Load Balancer as an instance ID and port combination, and traffic is distributed to the instance ID and port corresponding to that container. This dynamic mapping allows you to have multiple tasks from a single service on the same container instance. For more information, see the User Guide for Application Load Balancers.

Network Load Balancer

A Network Load Balancer makes routing decisions at the transport layer (TCP/SSL). It can handle millions of requests per second. After the load balancer receives a connection, it selects a target from the target group for the default rule using a flow hash routing algorithm. It attempts to open a TCP connection to the selected target on the port specified in the listener configuration. It forwards the request without modifying the headers. Network Load Balancers support dynamic host port mapping. For example, if your task's container definition specifies port 80 for an NGINX container port, and port 0 for the host port, then the host port is dynamically chosen from the ephemeral port range of the container instance (such as 32768 to 61000 on the latest Amazon ECS-optimized AMI). When the task is launched, the NGINX container is registered with the Network Load Balancer as an instance ID and port combination, and traffic is distributed to the instance ID and port corresponding to that container. This dynamic mapping allows you to have multiple tasks from a single service on the same container instance. For more information, see the User Guide for Network Load Balancers.

Application Load Balancer and Network Load Balancer considerations

The following considerations are specific to Amazon ECS services using Application Load Balancers or Network Load Balancers:

• Amazon ECS requires the service-linked IAM role which provides the permissions needed to register and deregister targets with your load balancer when tasks are created and stopped. For more information, see Using service-linked roles for Amazon ECS.

• For services that use an Application Load Balancer or Network Load Balancer, you cannot attach more than five target groups to a service.

• For services with tasks using the awsvpc network mode, when you create a target group for your service, you must choose ip as the target type, not instance.

• If your service uses an Application Load Balancer and requires access to multiple load balanced ports, such as port 80 and port 443 for an HTTP/HTTPS service, you can configure two listeners. One listener is responsible for HTTPS that forwards the request to the service, and another listener that is responsible for redirecting HTTP requests to the appropriate HTTPS port. For more information, see Create a listener to your Application Load Balancer in the User Guide for Application Load Balancers.

• After you create a service, the load balancer configuration can't be changed from the AWS Management Console. You can use the AWS Copilot, AWS CloudFormation, AWS CLI or SDK to modify the load balancer configuration for the ECS rolling deployment controller only, not AWS CodeDeploy blue/green or external. When you add, update, or remove a load balancer configuration, Amazon ECS starts a new deployment with the updated Elastic Load Balancing configuration. This causes tasks to register to and deregister from load balancers. We recommend that you verify this on a test environment before you update the Elastic Load Balancing configuration. For information about how to modify the configuration, see UpdateService in the Amazon Elastic Container Service API Reference.

• If a service's task fails the load balancer health check criteria, the task is stopped and restarted. This process continues until your service reaches the number of desired running tasks.

• When you use a Network Load Balancer configured with IP addresses as targets and Client IP Preservation disabled, requests are seen as coming from the Network Load Balancers private IP address. This means that services behind an Network Load Balancer are effectively open to the world as soon as you allow incoming requests and health checks in the target's security group.

• Using a Network Load Balancer to route UDP traffic to your Amazon ECS tasks on Fargate require the task to use platform version 1.4.0 (Linux) or 1.0.0 (Windows).

• Minimize errors in your client applications by setting the StopTimeout in the task definition longer than the target group deregistration delay, which should be longer than your client connection timeout. See the Builders Library for more information on recommended client configuration here .

  Also, the Network Load Balancer target group attribute for connection termination closes all remaining connections after the deregistration time. This can cause clients to display undesired error messages, if the client does not handle them.

• If you are experiencing problems with your load balancer-enabled services, see Troubleshooting service load balancers.

• Your tasks and load balancer (Application Load Balancer or Network Load Balancer) must be in the same VPC.

• The Network Load Balancer client IP address preservation is also compatible with Fargate targets.