An error can occur anytime during the execution of a function. A function might not be executed only once if it failed. The model of execution depends on the type of function.

For example, a simple synchronous HTTP request will be invoked once, at most. This means that the function invocation will be failed and never retried. The caller side is responsible for error handling and the retry strategy on its own.

While asynchronous functions will be invoked at least once, as is the nature of these asynchronous calls, so we need to prepare for a situation that this kind of function will be invoked multiple times. Also, the state to be modified by these functions should be idempotent and robust. For example, we may need to implement a state machine to control the states of the system.