The pricing structure is misleading to the point of alomst being dishonest:

1. 99.5% availability is advertised. It seems that this requires your deployment being to at least 2 instances, ie double the advertised cost to achieve the advertised reliability.

2. "compute time" is "elapsed time" not "CPU time".

Given these two "gotchas", I wonder what else I will be cought out by? Am I the only one to feel uncomfortable?

This would really make Azure the platform of choice.
If a worker consumed processor resources only when it was needed and was "swapped out" when idle for a (configurable) time, it would make lightly used or periodically apps cheaper to host.

I agree entirely with royleban. The issue is not so much the size of the instance as paying for the instance when it isn't in use.

It would make a big difference if lightly used applications were charged by "CPU time" rather than "uptime".

It is critical that we have a way of protecting against user or programmer error in a production database, or for historic archive. Database Copy is profibitively expensive option if you want to keep (say) 7 days + 4 weeks + n months. Ideal solution: a way of periodically backing up to a blob, and a way of restoring from a selected blob to the same or a different database.