Async: Avoiding Disaster and Unbounded Growth

Information gathering

Your project is dynamic:

• Make it traceable
• Constantly monitor

Backpressure

Backpressure is the process of "pushing back" on producers that are too fast.

This throttles the system, but allows it to not fail.

Bounded vs. unbounded growth

• Bounded channels are predictable
• Unbounded are more dynamic
• Bounded provide backpressure
• Unbounded are useful if you know they are never beyond a certain size

Holding state

• Tasks can hold state
• Otherwise, Mutexes and RWlocks allow sharing

Dropping futures

• Dropping a future means cancelling it
• Be aware of what happens if it is cancelled

Shutting down

• Make sure your signal handling is centralized
• Every component should subscribe to a cancel notification

Implementing a custom Future: Pinning

• Futures are not allowed to move in Memory
• The type that describes that is called Pin
• Pinning is hard, but there's support libraries

The poll protocol

Futures are poll based - that means they get asked if they are complete.

• This happens an infinite number of times, until they mark themselves complete
• The process is optimised through the Waker type
• Implementing poll yourself is rather easy