AWS annoyances

AWS is the very best Cloud platform out there, but that mean I can't whine, right?

1. Unpredictable region roll out. When will Singapore get AWS Batch for e.g.?

2. Wrapping your head around regions first... how some things like ACM only seems to work in us-east-1. $SERVICE/$REGION might have been a better structure.

3. Autoscaling and load balancer seem poorly connected, no CloudWatch Per Autoscaling Group

4. No way to track memory exhaustion .. I have found AWS mon-scripts but I wish it was easier than this...

5. Launch time != Uptime

   Launch = day 0 Uptime = continuous availability since last OS restart

6. You can't seem to query CloudWatch metrics by tag

7. Amazon Linux doesn't support SystemD

8. Amazon Linux only updates on boot, so you must plan to rotate out your service.

9. Cloud Formation lag, making it difficult to orchestrate stuff. By the time it does come available, you've long forgotten about your manual process.

10. DynamoDB throughput exceeded... this happens even on autoscaled managed instances in my experience!

My DynamoDB #awswishlist
1. filter THEN limit during query/scan
2. unique secondary indexes
3. no provisioned throughput

— Rich Buggy (@richbuggy) June 29, 2017

11. DynamoDB JSON Why... WHY!?

12. IAM policies are a total PITA. An ongoing service to tighten IAM policies based on usage is sorely needed.

13. AMIs should be more like Docker images, i.e. infra as code, treated like cattle. AMI snapshots and systems manager services patching seems like treating your infra like pets to me.

14. Redirects in S3 can be quite unreasonably painful. Same goes for setting up CORS. Or cloudfront for that matter. Should be far more streamlined. AWS@Edge will hopefully paste over these cracks.

15. Debugging Lambda is such an incredible pain right now. Better Cloudwatch integration (error in Lambda? show me the exact log for the corresponding request id!!!) and at least storing events for replaying would be nice.

16. AWS Marketplace sellers only available to US incorporated business and it is (currently) not suited for a range of solutions on the AWS platform..

17. AWS has a lot of services (code build stuff that rivals github/travis that comes to mind) which are probably best left to third parties (at least properly encourage an ecosystem!) and they should focus on core commodified services (like a better DynamoDB/S3).

18. I've had a terrible experience with AWS ECS. Could I be coaxed to give it another try?

    • Clusters are limited by initial machine type choice
    • Difficult to gauge what the limits are, esp on high load systems
    • The task placement and service abstraction is over engineered. I want to simply map docker images onto EC2 resources!!
    • Need to change Dockerfile's port? Prepare to completely redo your service
    • Slow to deploy versus almost instant Lambda update