Solar Inverter Monitoring
2024-07-18
I installed a hybrid solar inverter and battery bank in my home a little while ago. I have panels on the way right now, but in the meantime I've had the inverter up and running as a whole-house UPS for a little while.
But, what does this have to do with computer hackery and comp-sci theory you ask? It's a fair question and I'm getting there. The SolArk, like most modern pieces of technical hardware, comes complete with a bunch of web integration. You plug in a little WiFi dongle and your data gets whisked off to a datacenter somewhere, where you can have the privilege of accessing it in some limited ways through a mediocre web interface and maybe some annoying REST APIS. I found this unsatisfactory for a number of reasons
- This is a hybrid solar inverter. Part of it's purpose is to power my home when grid power is down, which means, network communication might also be down. When I *most* want to see my data, I'm likely to not have communication.
- I want to access the data MY way, with my own scripts
- No-one but me needs to ever see it.
- It offends my sensibilities. Why is my data traveling all over the world when I only need or want it *here*.
- I don't trust IOT devices to have properly implemented security, because, they usually don't, so it's having this thing on my network is a security hole in my network.
And my solution was this
![Solark Dashboard](https://blog.computersarehard.net/images/solark_dashboard.png)
First Steps
Pulling the data
But, the Solark fit my needs really well otherwise. So, rather than whine and whinge about it I decided to see what would be required to get the data off the device myself. I thought this would be a HUGE project, but as it turns out it wasn't at all. With a little searching online I found that the solark speaks modbus, and they actually published a spec for it . You can talk to the solark over a couple of different interfaces, but I already had a USB->RS232 adapter lying around, so I plugged it in and after playing around in python for a little while I started getting data.
Logging the data
I already had an IoTaWatt for monitoring power usage in my home. It's both open-source hardware and software, and it supports pushing the resulting data into InfluxDB. So, I'm also running InfluxDB on my home server. I actually used this set of tooling to help understand my power usage so I could properly size my solar build. Anyway, the point is I already have my power usage data in InfluxDB, and it'd be great to have my solar data in the same place. So I dug up a python library for influxDB and pretty soon I had my data streaming into InfluxDB... pretty cool.
Alerting the data
Well, that's all well and good, but I have to keep looking at my dashboards. If my power goes out my WHOLE HOUSE is backed up, so there's no obvious way to even know that my power is out. That's... not great. So, I want my wife and I to get notifications somehow. I asked my wife what she'd prefer and she it should just come to our cellphones. That made sense to me. The two of us currently use the matrix chat protocol to talk to each other wherever we are. Though, we currently use an external server for this (so it won't work our communications are out), it's fully open source and I can set up and run my own server, and probably will.
So, I dug up a matrix library. This turned out to be the hardest part, especially as the first library I tried wasn't flexible enough for what I needed. I ended up moving the entire program from asynchronous, to synchronous, and then back to asynchronous again, but I eventually got it working. And, here's the result
Getting Serious
I ran that program for maybe a month and eventually got it largely working, but still felt... sketchy. In python it's hard to have any confidence in codepaths that aren't actively tested. This program is really just communication at it's heart, which means a lot of possible points of failure. But, then I hurt my foot and had a lot of time to kill, and I'd been meaning to learn rust, so I rewrote the whole thing
This was my first real project in rust and it didn't disappoint. As I expected the type safety gave me far more trust in corner cases that I'm unlikely to bother testing. The "?" operator is a really convenient cross of exceptions and error return codes. I found that it helped me think about all the error cases without adding much complexity to the code. That, combined with the single mutable reference restrictions in rust did make some functional idioms harder to use, and I definitely tried a few cute "Build up all the futures and then run them" sorts of tricks that didn't work. But, all in all it's better than any other language I've tried for this type of code.
Final notes
The final result is one of the few personal projects I could see someone else actually wanting to run, with proper configuration options and files, and all of that jazz. The real takeaway here though is that I had a problem, and I solved it, and while not *easy*, it wasn't that hard either. I run my new monitor on system startup, so I don't have to think about it and hopefully it will just hum away logging what's going on and letting me know when something interesting happens.
If you're in the space you may wonder why I didn't integrate with Home Assistant. The answer is that Home Assistant is cool and open source and stuff but... not really up my ally. It feels large and bloated, it's nearly impossible to run without virtualisation, it's intended for people who like clicky-button interfaces rather than configuration-as-code, and I just don't need it. Having Home Assistant play middleman would, if anything, obscure the simplicity of what's going on here, adding layers of abstraction, conversion, protocols, DSLs that would be in my way.
FOLLOWUP
I've been running this software and indeed it is more stable than my python version. It took me about 3 days to write. By the time I had all the features and it felt like it was working, it was working. After that I had one bug in room joining that was a pure logic error on my part, and that was it. I added a few more design improvements and adjustable values, but it was a drastically different experience from the weeks of bugs still popping up occasionally in the python version.
I noticed that the development process in rust is very different. Before retiring recently my job was coding in C++. I'm not an "outline" kind of person. I'll start in one of two places. If I'm worried that something is going to be too hard or impossible, I'll start there to gain clarity on that piece of the system and "de-risk" the project. If I'm not worried and I'm fuzzy on the overarching design I'll start with the part I do know so I can build outwards from there. Both of these approaches work well in the languages I've used the most professionally: C++, python, javascript, and a number of DSLs, but rust has it's own ideas.
In rust most of your time is spent trying to get the program to compile. I'm a newbie so this is extra true for me, but I have enough experience in languages like SML and Ocaml to have a good idea that this remains true even for fairly skilled developers. In fact, arguing with the compiler constantly is the point of rust. You spend your time trying to get it to compile, possibly re-writing it a couple of times as your realize your way of thinking of the problem isn't compatible with rust's, instead of spending time debugging. The compiler is telling you "no" and forcing you to work through things you wouldn't realize until testing in other languages. That's a wonderful property. There's a side-effect though where you quickly find you're constantly asking "will the compiler let me do that". To answer that you need to be able to ask, and for to ask it the rest of your program needs to be valid. So, there's this strong incentive to keep the whole thing compilable most of the time as you develop. For this reason I quickly found myself switching to an outline style of programming. I'd write a "dummy" for a function or a struct so I could write outer code first, then go write the inner code later. This isn't a bad thing, but it's the first time I've run into a language who's design so strongly encourages one approach over another.