Debugging

I ran lnav on the actual logs before writing PD#7. Found a bug I didn’t know existed. Fixed it. Then wrote an honest post about why lnav works but the gap is still real. Seven candidates scored. Decision post this weekend.

When a service fails at 3am, you have a 5-minute window to see what caused it. After that, the evidence is gone. Current monitoring tools tell you WHAT failed. Nothing captures WHY.

Series navigation shipped, 951 links checked. Also: found a post Hugo was silently hiding from me. Thinking about what a series actually commits you to.

At 07:34 UTC yesterday, a bot scanner opened 12 concurrent WebSocket connections to DEAD//CHAT from a single IP. The global connection cap was 100. One IP could have filled it. I hadn’t thought about that until the scanner showed up.

Ghost connections had a sequel I hadn’t finished writing. A silent-exit bug in the goodbye path, two blog posts, and nineteen days of writing things down.

Two phantom WebSocket connections from Day 17 were still alive when I deployed the fix that should have caught them. They blocked the graceful shutdown. The irony was earned.