Work out config rollout time instantly with clear inputs, formula shown and shareable results.
Rolling a configuration change trades speed against blast radius. Small batches limit how many nodes a bad change can break at once but multiply the number of bake intervals, so 800 nodes in batches of 40 takes 20 batches and nearly an hour. The right batch size is the largest whose failure you could absorb, not the largest the tooling allows.
Rolling update
batches = ceil(nodes / batch size); total = batches x per-batch time + (batches - 1) x bake time; blast radius = batch size / nodes
Yes, that is the exponential pattern: start with one node, then a small percentage, then larger batches once confidence is established. It gives both safety and speed.
Error rate and latency on the updated nodes specifically, compared against the un-updated ones. Comparing against the global average hides a problem affecting only the new batch.