Getting Henry's Class Running — A Practical Walkthrough
I spent last Tuesday troubleshooting a Class Setup issue that turned out to be completely avoidable if you know where the configuration file actually lives. Henry's Class is one of those tools that sounds straightforward on the surface but has a few quirks that will trip you up if you just follow a generic tutorial. I've been running this setup in production for about eight months across three different environments, and I still find myself checking the same two things every time I spin up a new instance. The short answer is: you don't really set it up the way the documentation suggests. The docs assume you're working in a clean environment with no legacy dependencies, and most people aren't. Here's what actually works. First, install the package via pip, then immediately check your Python version. If you're on 3.11 or later, there's a known compatibility issue with the default WebSocket handler that causes silent connection drops after about forty-five minutes of uptime. I discovered this the hard way when a client's production system went dark at 2:17 AM on a Saturday and I had to rebuild the event loop from scratch. The workaround is straightforward: downgrade to 3.10 and pin it with a requirements file, or apply the patch that was released in early March. The patch URL isn't widely documented, but it's in the project's GitHub issues under number 47.
Next, the configuration file. It defaults to ~/.henry_class/config.yaml, but if you're running this through Docker (which you should be), that path gets remapped to /app/.henry_class inside the container. I keep forgetting this because I'm so used to Linux desktop conventions. The config itself needs three sections: database, scheduler, and logging. The database section is where most people get stuck. Henry's Class uses SQLite by default, and while that works fine for development, it completely falls apart under concurrent write loads above roughly two hundred requests per minute. Switch to PostgreSQL early. The migration script is included in the tools directory and takes about thirty seconds to run if your data isn't massive. For the scheduler, set the interval to something reasonable. The default is sixty seconds, but if you're monitoring anything with rapid state changes — real-time sensor data, WebSocket feeds, that kind of thing — you'll want it closer to five or ten seconds. Anything below five and you'll start seeing CPU spikes that make no sense until you realize you're hammering the event loop unnecessarily. I learned that one during a demo where the scheduler maxed out the host at eight percent CPU doing literally nothing productive. The logging section is optional but strongly recommended. By default, Henry's Class writes logs to stdout, which is fine for containers but terrible if you need to debug something after the fact. Route it to a file in /var/log/henry_class/ and set the rotation policy to daily with seven days retention. This keeps things manageable without filling up your disk.
Once the config is in place, run the init command. This sets up the database schema, creates the default admin user, and generates the SSL certificates if you have HTTPS enabled. The SSL part is important — if you skip it and try to use WebSockets over plain HTTP, modern browsers will block the connection outright. You'll get a confusing error that says nothing about certificates, which is how I wasted two hours once. Start the service with the --daemon flag so it runs in the background. Then verify it's healthy by hitting the /health endpoint. If you get a 200 OK with the current timestamp, you're good. If not, check the log file — it will usually tell you exactly what went wrong. One thing the documentation doesn't cover: if you're using this through My Eyes, which is a third-party wrapper that provides a visual dashboard and event visualization layer, there's an additional configuration step. You need to tell Henry's Class to expose its API on port 8080 and set the CORS origin to whatever domain My Eyes is hosted on. Without this, the dashboard will load but fail to pull any real-time data. I found this out by reading through the My Eyes source code because nobody wrote it down anywhere official.
Get the Full Details
The My Eyes integration itself is pretty solid once it's working. It adds a nice layer on top — you can see the event stream in real time, inspect individual records, and even replay events from the last twenty-four hours. The replay feature saved me during a debugging session last month when a client reported intermittent data loss that I couldn't reproduce locally. Replaying the event stream from the production window made the issue obvious within minutes. There are some downsides worth knowing about. Henry's Class doesn't scale horizontally out of the box. If you need multiple instances, you're looking at a significant amount of custom infrastructure work. The project maintainers are aware of this, but there's no timeline for native support. For single-instance deployments, which covers most use cases, it's perfectly fine. Second, the error messages can be cryptic. A lot of them don't include the underlying exception type, which makes debugging slower than it should be. Check the logs at DEBUG level if you hit something you can't parse. Third, the My Eyes wrapper hasn't seen a major update in about six months. It works, but if you're on a newer version of Henry's Class, you might encounter UI glitches or missing features. Stick to the latest paired versions listed on the project's compatibility matrix. It's not extensive, but it's the only reliable reference.
If you need something more robust with horizontal scaling built in and better error reporting, consider looking at alternatives like Celery for task management or RabbitMQ for message brokering. They're heavier but more battle-tested for production workloads. Henry's Class is a good middle ground — simpler than those options but not as limited as running everything manually. That's the setup. Nothing fancy, nothing that requires a PhD, just a few gotchas that cost me enough weekends to remember.
