Retry function invocation in case of failure

[mthenw]

This is a Feature Proposal

• What is the use case that should be solved?

Right now, in case of failed invocation error is returned to a user. We can improve that process by retrying invocation with exponential backoff, similarly, as AWS does that for Event type invocations (and also for messages from Kinesis etc.).

In the first pass, we can hardcode the number of retires. Eventually, this should be configurable either on event level or function level (providing that value while registering function).

Similar or dependent issues:

• Handling of Concurrent Execution Limit of AWS Lambda #315

[dzimine]

$0.02 - I would use caution here: blindly retrying like M$ EventGrid does makes no sense (what's the point of calling again if the response code is “401 unauthorized”? How about “414 URL too long" - is it getting any shorter with the retries?). Giving a user an option to set up retry policy (count, interval, back-off) is better; I would think about solving this as an error handling policy that comprises retry policy and error handler, a choice from "stock" responses or custom function call.

AWS StepFunction is IMO a good example of how to handle errors & how to syntactically expose it to the users.