This complex diagram outlines the various processes and machines that get involved in a single voice request, and any piece of it can give you trouble implementing something. I got this picture directly from a nice description I ran across on Amazon, and let's start off going there to understand what is going on <link>.
Robert McCauley, the author, is unusual in that this article can be understood; at least the second or third time you read it. Ignore his comment about referring to the Amazon quick start documentation for IoT, it will only confuse you and it really doesn't talk about anything Robert doesn't. However, this article is a bit terse (meaning he passes the buck entirely) on how to use the Alexa voice service. The thing you want to learn from this is that Amazon uses mqtt to control your devices. You can actually interact with them using the AWS mqtt tool that Amazon provides and the instructions that Robert provides.
I actually created his water pump and messed with it, but I deleted it to keep from confusing myself later when I created my 'house' device. What I did was create a single device 'house' and it carries stuff like outside temperature, state of various lights, garage door position, etc. It would have been a real pain to create each of those devices and deal with them separately. The huge thing to understand here is the use of the JSON shadow device. You deal exclusively with the shadow that represents the last reported state of your device (my house). The shadow contains a 'reported' section, a 'desired' section, and a 'delta' section. For example, here is my current (as of the date above) shadow; remember it's a JSON document.
I don't have all the items I'm interested in yet, and I only voice control a few things so far, but this could be representative of anyone's implementation. This is the reported section. If I wanted to control something, I'd put in a desired section that would contain what I wanted to change. So, if I want to turn on the outside lights:
That would get combined with the reported section and then inspected for differences. Any differences would show up in a new delta section. That would make the entire shadow document look like:
So you have a JSON document that holds what was last reported, what you want to change, and the difference. See why they call the sections 'reported', 'desired' and 'delta'? It actually makes sense. When you get a device hooked up to the mqtt server at Amazon, the delta portion is the only part you see coming in. That makes it easier to understand what you need to do with the actual device. So, on your Pi, or whatever, you subscribe to the Amazon mqtt service, and wait for a delta message to come in. When it does, you do what it says and report back the new status. There are a few complications in there that you'll have to deal with, but I hope to visit each of them as we go through this process.
In Robert's article, he interacts with the shadow document using Amazon's mqtt client to subscribe and publish to the shadow. That's exactly the kind of thing you will need to do in code to support your device. This stuff is handled by what's called a Lambda function. The Lambda funtion is simply code that you run on their machine that can bridge the gap between the Alexa voice service and the mqtt server to get something down to your machine to change something. In the other direction, you'll publish the latest state of your devices back so everything is kept up to date.
Now, do you understand why a little hands on with the Amazon IoT is necessary? Follow the lead from Robert's article and work with mqtt and the shadow a bit to see what is actually going on. Totally ignore his comments about Alexa, "Once you are familiar with the Alexa Skills Kit and understand how to create an Alexa skill;" That phrase is just about totally useless.
I have a suspicion that very few of Amazon's people that work on these projects actually do anything with them since their explanations leave so much out.
That's enough for you to get a feel for the interaction if you actually do it and then look at the messages that are passed in mqtt. You WILL need this when you start troubleshooting your own code.
Next post we're going to look at getting data from your device all the way back to Alexa. We won't actually try to change the state of a device because you have to have a reported section to the shadow before you can change it.
And, you probably won't need the water pump device that Robert had you create.
Continued here <link>