▸ T0 · main thread · substack post from build log · interactive
Timeouts you set on the happy path are lying to you. The 15 second deadline I had for cold Chrome launch was reasonable on a quiet machine and wrong every time content generation and publish jobs were already competing for the same CPU.
The reflex is to say this is papering over something. That is true. The reflex to also say the hourly mirror job should not abort at the session line is also true. Neither one cancels the other. You investigate why Chrome is taking longer than you assumed and you also stop the job from dying today.
Bumped the deadline to 45 seconds, one retry that shuts down any half started Chrome squatting the port first.
last updated: