Remove callchain approach from MetricFilter resource type to reduce latency in stack operations

[miparnisari]

Issue #, if available: https://github.com/aws-cloudformation/aws-cloudformation-resource-providers-logs/issues/39

Description of changes:

• cfn test passes: ========= 12 passed, 7 warnings in 214.08s (0:03:34) =========
• Integration tests pass

Before this change:

With this change:

By submitting this pull request, I confirm that you can use, modify, copy, and redistribute this contribution, under the terms of your choice.

[PatMyron]

Is this applicable to all resource types?

[miparnisari]

the latency could be reduced by removing read handler call at the end, call chain does not add additional latency. we should keep call chain

I tried that, I didn't get a major improvement. Tests still took 20 minutes to run

Is this applicable to all resource types?

No idea. This is a simple type, with no stabilization.

[ammokhov]

the latency could be reduced by removing read handler call at the end, call chain does not add additional latency. we should keep call chain

I tried that, I didn't get a major improvement. Tests still took 20 minutes to run

Is this applicable to all resource types?

No idea. This is a simple type, with no stabilization.

we can rework this together. KMS has lower latency than than usual by using the callchain

[wbingli]

I don't think callchain framework is suitable for resource without stabilization. It abstracts the re-invoke but because of the abstraction, it lacks the flexibility to control it. If we need to workaround and really careful to avoid pitfalls to make what we need, it takes more effort than this implementation, with simple request-response handling.

Same as for issue in (https://github.com/aws-cloudformation/aws-cloudformation-resource-providers-logs/pull/42#issuecomment-653652724), it brings difficulty for development and debugging. To solve such issue, the developer eventually needs to understand the complex framework, it's not easy and over kill for a resource simple as this.

[dchakrav-github]

I don't think callchain framework is suitable for resource without stabilization. It abstracts the re-invoke but because of the abstraction, it lacks the flexibility to control it. If we need to workaround and really careful to avoid pitfalls to make what we need, it takes more effort than this implementation, with simple request-response handling.

Same as for issue in (#42 (comment)), it brings difficulty for development and debugging. To solve such issue, the developer eventually needs to understand the complex framework, it's not easy and over kill for a resource simple as this.

Did we ever remove the 60s wait time that was introduced to support handshake model for the original CWE, which had a minimum of 60s delay that was needed?