It's somewhere around the 10th of March. I lost track of the exact day because I've been awake at weird hours staring at a black terminal that keeps telling me "permission denied" and I keep typing the word back at it like it owes me an explanation.
Last week I pushed my first batch of bots live. This week they have been dying. Quietly. No alarms, no error popup, nothing dramatic. I'd check the dashboard and a bot that was running yesterday is just... not. No log line that means anything to me. The Telegram bridge in particular has it out for me personally. Every time the box restarts, the bridge crashes. Every. Single. Time.
I had a laptop, an email account, and a social media handle nobody followed. That was the starting kit. I had never opened a terminal before January. I didn't know what an API key was until somebody on a Discord told me to stop pasting mine in screenshots. So this week, when the errors started piling up, my workflow looked like this:
- Bot dies.
- I SSH in. (I learned what SSH meant maybe three weeks ago.)
- I see a permission error or a path error or a service that's just dead.
- I open ChatGPT in the other tab and type something like "what does chmod 755 mean and why does it want that."
- I read the answer twice. I half-understand it. I try the command.
- Different error. Back to step 4.
I asked what sudo actually does. I asked what systemd is and why it has opinions about my Python script. I asked why a .service file needs a WorkingDirectory and what happens if you lie about it (it does not work, is what happens). I asked what a daemon is. I asked if my user account was the same as the root user and got a small lecture about why that question matters.
I'm still learning. Senior devs reading this will spot ten amateur moves in one paragraph. I know. I'm not pretending otherwise.
The Telegram bridge thing finally cracked open on attempt forty-seven. I'm not exaggerating the number, I have the bash history. The fix was stupid and small: the service was launching before the network was actually up, so the bot tried to phone home to nothing, choked, and systemd shrugged and gave up retrying. Adding After=network-online.target and Wants=network-online.target to the unit file fixed it. One line. Two lines, technically. After a week.
When the test ping came back from Telegram I made a noise that scared the dog.
Here's the part I keep trying not to write about, but it's the engine under all of this. Shoulder surgery on August 11 is on the calendar. That's not a soft date, that's a real one with a hospital attached to it. Every day I burn fighting a .service file is a day I don't get back. I'm racing the calendar. So when I tell you I sat there at 2am asking a chatbot what chmod means, I wasn't being cute about being a beginner. Started from zero is not a metaphor, it's a logistics problem.
27 bots is the goal. 60 days is the window. This week's count of new bots shipped: zero. This week's count of existing bots kept alive: all of them, eventually.
I'll take it.
Question for anyone who's been doing this longer than me: when a systemd service fails silently, what's the first place you look that isn't journalctl -u?