AWS Fargate abstracts away the EC2 instances behind your ECS tasks. And while the management model has changed, your risk exposure hasn't. Your tasks still call databases, caches, and internal APIs that can slow down or stop answering, and now the only place left to test is the application itself.

Gremlin reaches that layer with Failure Flags. Deploy the sidecar alongside your container and you can inject faults into a running task without any host access whatsoever. Teams usually start with three checks:

• Does one slow dependency take down the whole task? Inject latency into a single call and discover whether other endpoints stay responsive or share the same thread pool.
• Do unhealty tasks impact the user experience? Fail a health check dependency and validate that the scheduler replaces it without interrupting availability.
• Do your circuit breakers open? Trigger repeated failures and confirm the breaker trips, then confirm that it closes again once the dependency recovers.

Fargate took away the host, not the responsibility for what runs on it. Testing the application is how you keep meeting that responsibility.