add new abstraction for simple Lambda that does not need explicit initialization

[tomerd]

motivation: make simple things simple

changes:

• add new abstraction names SimpleLambdaHandler that has a default initializer
• add test

[tomerd]

we could / should consider doing this on LambdaHander instead of the new abstraction, also not in love with the "Simple" prefix.

[stevapple]

We can drop context: from LambdaHandler’s initializer, then the SimpleLambdaHandler model will just fit (we can keep async and throws so users can still initialize some resources if they want).

[tomerd]

@stevapple context is needed when users need to initialize resources, it provides thinks like NIO's EventLoop and Terminator, so we cannot just drop it.

[stevapple]

Is there any reason they cannot use async/await for asynchronous startup, and use throw in place of manual termination? I think that's the "simple" way and should work for most people.

[tomerd]

termination is for resource cleanup when shutting down, there needs to be an entry point to that. also all NIO based clients (eg db client based on NIO) need access the the EventLoop for initialization to avoid thread hops

[fabianfett]

While I like the idea of making Lambda easier to use, I wonder if we actually achieve this goal through this change. Until now, IMO, it was clear which Handler protocol we recommend: LambdaHandler. Now the story is becoming harder.

Let's look at the pros and cons of this change:

Pro:

• We can remove 1-3 lines for very simple LambdaHandlers.
• Users don't need to specify an explicit init

Con:

• We need to explain to users the difference between SimpleLambdaHandler and LambdaHandler even in the Simple case. We need explain to users, that they shall use SimpleLambdaHandler if they don't have any resources that they want to share between invocations.
• The way from a very simple lambda to an actual lambda (I think every real lambda, will use some shared network resources) is becoming harder. We need to remind developers, that they need to change the Handler protocol and now need to implement a new init method. There is no compiler warning to point them into the right direction.

In my opinion the cons outweigh the pros. Is there something that I have overseen?

[tomerd]

superseded by #273