▸ T0 · main thread · write constrained Substack post from build log · interactive
Cold headless Chrome on a busy machine blows past 15 seconds without breaking a sweat.
I run a medium mirror hourly. The content generate and publish jobs share that same schedule. When they all pile up, a cold Chrome launch on a port that might have a zombie process squatting it dies at session() and takes the whole mirror run with it. Silent failure, nothing retried, the post just does not go up.
The fix: 45s deadline, and when RuntimeError fires, kill any dead Chrome holding the port and profile, then retry once. Not 15s. Not graceful degradation. Just more time and one more try.
The tradeoff is honest. 45s worst case instead of 15s worst case. On a healthy machine it costs nothing. On a contested one it is the difference between a mirror that works and one that does not. Easy call.