Stellar RPC - Resource limit alerting

[dmkozh]

Currently RPC only alerts users about the simulation budget being exceeded via simulation failure. Budget only enforces CPU and memory network limits. Besides the fact that this alerting is almost accidental (i.e. it's a runtime property, not an explicit alert), the error messaging is not very clear here either - in simulation the failure happens when a maximum transaction instruction limit has been exceeded, while during the real execution the same 'budget exceeded' failure happens when the instruction limit declared in the transaction has been exceeded.

This discussion is for figuring out a more consistent approach to all the limits, that makes it clear what exactly has been exceeded, and does so in a consistent fashion for all the network limits.

Here are some key inter-dependent discussion points that I think need to be resolved:

1. DOS protection vs network limits: should the simulation execution limits be exactly the same as the network limits, or can they be proportionally larger? (say, 2x larger)
2. Alerting mechanism - when/how we inform RPC users of limit being exceeded. This is related to the previous point. Some options include:
   • Fail transaction as soon as possible when a limit has been exceeded and notify users of that limit (note, that this is not applicable for write bytes limit)
   • Fail transaction after executing as much as possible while still maintaining DOS protection, and notify users of all the exceeded limits
   • Do not fail transaction if possible (while still maintaining DOS protection), and add information on the limits exceeded
3. Alerting/limit customization: the more failure conditions we introduce, the trickier it might become testing and debugging contracts on the local RPC (where DOS protection or even the network limits may not be of concern). How much customization should RPC provide and at which level? (e.g. RPC config vs request parameters)