Portfolio project of Elixir and Phoenix LiveView, to sharpen skills, practice picking up new technologies quickly, and enjoy programming on the Erlang runtime.
The objective is to instrument HVAC refrigerant coils:
30-second window streaming data from (primitive) simulation of nine sensors.
- Clean Code
- Data Model
- Chart.js Integration
- LiveView Integration
- API Auth Middleware (Plug)
- Sensor Ingest Endpoint
- Sensor Simulator
- PubSub Streaming to Chart.js
See web/ or just run make:
make web # In one terminal.
make sim # In another.
View the running app at http://localhost:4000.
We have minisplit HVAC, and have had something of a nightmare scenario: leaking twenty seven pounds of refrigerant over time; a grinding compressor that would soon fail; its replacement louder than the originally promised "quiet" model; the inability to run any room on the low fan setting; intermittent, daily system shutdowns (with a generic error) without any cause; and without any diagnosis no matter how many times we call the installer (or a second opinion) or have the manufacturer hook into the system over screenshare.
Have you tried turning it off and back on again? Only thing keeping us cool.
Once the walls went up, installing central air was no longer an alternative. Minisplits are still the right answer, anyway, to set temperatures per room. Instrument the HVAC system. Data logging from HVAC had no anomalies, and systematically stepping through the installation components did not reveal additional errors.
- Clip temperature sensors onto each HVAC branch refrigerant coil.
- Stream telemetry from microcontrollers into a central database.
- Corroborate system shutdowns with ground-truth data of each subsystem.
Exploring Erlang over the years.
2008. I reviewed mnesia to better understand the architecture of a project where I was lead on creating a low-cost SMS gateway out of an ordinary Android phone. I also wanted to better understand emerging NoSQL technologies.
2011. I had selected XMPP to prototype projects exploring social, local, mobile capabilities of an in-house platform. This involved deploying OTP and writing ejabberd modules. It would later be replaced by redis.
2018. I wanted a general-purpose language in my "personal SDK" that wasn't blub. I included Elixir as part of the evaluation, and would ultimately choose OCaml. In particular, I need to ship binaries as a systems programmer.
Personal language choice for systems programming.
I would later choose V, in which blub is secondary to the design principle of zero dependencies. A project needs only POSIX and a C compiler, vendoring everything including the language runtime. It is otherwise remarkably difficult to escape the vast incidental complexity introduced by the toolchain.
2020-2022. I needed a UI toolkit for rapid application development in a clinical setting for iterative (IRB approved) workflows in managing drug-resistant chronic disease.
My notes on Elixir Desktop and Scenic from 2022.
(Ed. note: Unedited notes to self from a highly critical evaluation of UI toolkits viewed from a top-down software design process. When looking across many technologies, it's quite likely that you miss something, either because it wasn't well documented at the time or you were moving too quickly. There's no doubt that the projects in question have their own merit.)
Elixir Desktop is a container for Phoenix LiveView. Scenic is GL-based (and only Linux and macOS).
I was interested in Elixir Desktop because I have a use case that's appropriate to LiveView: integrate telemetry from multiple sensors, in a specific environmental/individual context (i.e. a given patient in a specific room setting), synchronize to wall-clock time, radiate information to one or many heterogeneous dashboard views. What is more, one-off prototyping would be more common than a deployed web form—enable a physician to try different questions without updating native apps. Any production use is still beyond the horizon, so my exercise is primarily in making sure I have the right tools and runtimes available to me for when the time comes.
I know that nothing really comes "for free" but if someone's sorted out the build workflow across big3 wx and mobile, I would leverage this if I could. It's not obvious from the project description or the 2021 conference video, but Elixir Desktop is apparently first and foremost about self-contained applications that runs locally. Meaning, it talks to a web server on localhost.
Now, this is great for sport, but what's the point except for adding a distribution channel for a single-user web application. At which point, the use of OTP is risible.
So, obviously I don't want Elixir Desktop. Some of the ideas might still be relevant, if I really do think that LiveView simplifies my programming model, and it might. But let's be honest, Lisp macros on the server and native clients are not all that verbose, especially if you can sort out some interface definition language.
Meanwhile, Scenic is really about creating UIs for the Nerves project. If your use case is to package "firmware" in the form of OTP with just enough OS (which impressively is marketed at 30MB), and to have a UI for IoT, then this has merit.
I would later choose Flutter.
2025. Stay sharp on technologies for high-volume in-house telemetry.
