[davidhyde]

Here are some more points to consider.

1. Allowing user initiated early retry exit. Nothing is stopping some other developer from wrapping your function-with-retries in their own retry-able function. And then another developer even further removed from the details to wrap that one again. Now you have n pow 3 retries and you have services timing out after an hour instead of milliseconds or seconds. Remember that your function-with-retries can just look like a slower flakey function to them. Perhaps exposing the retry variables in the api is the answer, perhaps not.

2. Code readability and debug-ability. Wrapping a piece of work in a function pointer and giving it a generic name like “action” increases cognitive load and makes the codebase slower to read. This is especially true if the contents of the action is so far up the boilerplate chain that the developer has already forgotten what it does. This is a part of the codebase that will be read to troubleshoot the failure so it helps if it is self describing.

3. In my experience, retries are often added when the failure mode is not well understood or as a quick hack to get things over the line. However, we sometimes cannot control what we call so they are an also valid mechanism to use under certain circumstances.