Backoff and retry status polling

[jacobtomlinson]

Fixes #56.

We currently poll for the pending container status every second until it is running. If ECS takes a long time to start we can eventually run into a ThrottlingException.

This PR catches the exception and continues retrying but with an exponential backoff to a maximum of 20 seconds. Once a request is successful it will resume polling every second until the exception is hit again.

This will continue indefinitely until the container moves to a running or failed state.

[eganjs]

Looks pretty great, just a few comments:

• Can the maximum backoff can be made configurable?
• Should a note be added in the documentation to cover this?

[jacobtomlinson]

Thanks for the review @eganjs.

The 20 second maximum is described in the AWS documentation. Do you have a reason why you would like this to be configurable?

I'm not sure this needs to be documented, this is more of a bug fix to respect the AWS API. It shouldn't affect the usage of the library.

[eganjs]

That's a good challenge.

I've re-read through #56 and the cases I was thinking of would work fine under the implementation.

I think I was being a bit overzealous with my concerns 😅

[eganjs]

Today the default limits for Fargate On-Demand resource count and Fargate Spot resource count were increased from 100 to 500 and 250 to 500 respectively (see https://aws.amazon.com/about-aws/whats-new/2020/09/aws-fargate-increases-default-resource-count-service-quotas/). Without this PR I suspect users of the library will be unable to leverage these increases, it would be good to get this merged ASAP unless there is more work to be done here?

[jacobtomlinson]

Thanks for the nudge @eganjs