fix(aws): skip failing RDS and VPC collectors instead of aborting provider

[stephan-rayner]

Summary

• RDS and VPC collector init failures previously returned nil, err, killing the entire AWS provider and dropping all metrics from every other service
• Aligns RDS and VPC error handling with S3 and EC2: log the error at ERROR level and continue so remaining collectors stay intact

Fixes #716

Test plan

• make test passes
• Verify that a pricing config error for RDS/VPC no longer prevents S3/EC2 metrics from being exported

[leonorfmartins]

hm, I'm more unsure about this one. I agree that we should not block other collectors from progressing if one of the collectors fails to be initialised. But just logging an error and continuing will not be enough because then our error rate SLOs will never fire and we won't know a collector broke in the first place.
What if we, on top of this change that you are proposing, return an error on each collector's initialisation? That way, we will still know if a collector fails to be initialised but healthy collectors will keep doing its thing.

[stephan-rayner]

hm, I'm more unsure about this one. I agree that we should not block other collectors from progressing if one of the collectors fails to be initialised. But just logging an error and continuing will not be enough because then our error rate SLOs will never fire and we won't know a collector broke in the first place. What if we, on top of this change that you are proposing, return an error on each collector's initialisation? That way, we will still know if a collector fails to be initialised but healthy collectors will keep doing its thing.

I like that @leonorfmartins! It was a kind of a coin flip for this change. A little more than half of the resources we look at do it the way in the PR and a little less than half do it the way you are suggesting.

Upon further reflection, and after having slept on it, I agree with you!