All posts

ECS Gateway vs a shared ALB vs Express Mode vs Service Connect: which one do you need?

How Amazon ECS Gateway compares with a load balancer you manage yourself, ECS Express Mode and Service Connect, what each one is for, and how to pick.

With ECS Gateway, there are now four AWS-native ways to route HTTP traffic to ECS services, and their names don’t make the differences obvious. One of them, Express Mode, even calls its API CreateExpressGatewayService, although it has nothing to do with ECS Gateway.

The short version: three of them get traffic into your services, and one handles traffic between them.

The four options

A shared ALB you manage yourself. The classic setup. One Application Load Balancer with a listener rule and a target group per service, usually defined in Terraform or CloudFormation. You own every piece, so you can do everything an ALB can do.

ECS Gateway. A platform team creates a gateway once, and services connect with a single setting. ECS creates the per-service rule, target group, certificate, hostname and security group, adds load balancers as they fill up and cleans up after deleted services.

ECS Express Mode. Launched in late 2025 for the “I have a container image, give me a URL” case. One call creates a Fargate service with a generated task definition, auto scaling, logs and an HTTPS URL on a load balancer it shares with other Express Mode services.

Service Connect. ECS’s service mesh for calls between services. A proxy next to each task routes http://api to a healthy task of the api service. It doesn’t accept traffic from the internet.

Side by side

Your shared ALBECS GatewayExpress ModeService Connect
TrafficInto servicesInto servicesInto servicesBetween services
Who sets it upYou, per servicePlatform team once, app teams add a connectorEach app team, per servicePer service, in a namespace
Existing servicesYesYes, with UpdateServiceNo, it creates its ownYes
ComputeAnyEC2, Fargate, Managed Instances (awsvpc)Fargate onlyAny
Custom domainsYesNot at launch, on.aws hostnames onlyOnly by editing the load balancer by handNot applicable
RoutingHost, path, headers, weights, redirectsHostname onlyHostname onlyService name and port
Load balancersAs many as you createUp to 25 services each, added for youUp to 25 services each, added for youNone, a proxy per task
CleanupYoursAutomaticAutomaticAutomatic

All four have no charge of their own. You pay for the load balancers, logs, metrics and data transfer they cause, and Service Connect costs the CPU and memory of its proxy containers.

When your own shared ALB is still the right call

Choose it when you need anything beyond “this hostname goes to this service”:

  • Your own domains, like api.example.com, without a CDN in front.
  • Path-based routing, like /api to one service and / to another under the same domain.
  • ALB features such as OIDC or Cognito authentication, redirects, fixed responses or weighted target groups.
  • One place for everything, when the load balancer already lives in infrastructure code that also manages DNS and certificates.

The cost is the per-service setup you write and maintain, and the cleanup nobody remembers to do.

When ECS Gateway fits

Choose it when a platform team runs the network and many application teams ship services:

  • Teams get HTTPS endpoints without writing load balancer code or asking for a new rule.
  • The platform team sets WAF, access logs, subnets and public or internal access once, and every connected service follows.
  • Services on EC2 and Managed Instances work too, not just Fargate.
  • Blue/green and canary deployments can send test traffic to the new revision by a request header the platform team defines.

The limits from the first part apply: no custom domains at launch, routing by hostname only, and some settings that can’t change after creation.

When Express Mode fits

Choose it for a single web app or API where speed matters more than control: prototypes, internal tools, small teams without a platform team. You hand over an image and two IAM roles and get a running service with scaling and a URL.

It’s also where App Runner users should go. AWS sunset App Runner on April 30, 2026, with no new customers or features, and points existing customers to Express Mode.

It’s the wrong choice for existing services, for anything that must run on EC2, and for setups where a platform team wants to enforce shared network settings. Like a gateway, it gives you an on.aws hostname, and a custom domain means editing the load balancer it manages, which it won’t keep in sync.

Where Service Connect fits

Service Connect is not an alternative to the other three. It answers a different question: how does checkout call payments inside the cluster? A typical setup uses a gateway or a shared ALB for traffic from users and Service Connect for traffic between services. The service networking post compares it with Cloud Map and internal load balancers.

How to pick

  1. Is the traffic between your own services? Use Service Connect, or one of the options from the service networking post.
  2. Do you need your own domain, path routing or ALB authentication? Use a shared ALB you manage yourself, or put CloudFront in front of a gateway.
  3. Do several teams ship services that a platform team should standardize? Use ECS Gateway.
  4. Is it one Fargate app that needs a URL today? Use Express Mode.

If you already run a shared ALB and it works, nothing forces you to move. The previous part of this series shows how to move a service to a gateway without downtime if you decide to.