How to Run Deployments That Take Hours

Long deploys need checkpoints, not anxiety spirals.

[ essay ]

Break into phases with verifiable checkpoints: migrate expand, backfill percent, flip read path, drain old queue.

Communicate ETA ranges. Automate progress metrics — rows migrated, lag, error budget.

Assign a commander for the window; others watch dashboards, not Slack threads.

Patience is procedure. Hurry without checkpoints causes the incidents patience was meant to avoid.

— JV · Dark Heart Labs.