2 ms·
Yes, this is exactly it. It's also an important part of the "crash only" concept - the moment that you detect a bug or inconsistency, it's important to crash so
by pjc50 4y ago
Yes, this is exactly it. It's also an important part of the "crash only" concept - the moment that you detect a bug or inconsistency, it's important to crash so that you discard potentially corrupt state in memory and re-load from the last known good point. This makes it harder to write out wrong state, because that's much harder to recover from.
The absolute nightmare scenario is state getting corrupted causing a crash-restart loop.
(a couple of days ago we had the "everything is about state" discussion on here: this is why separating "persistent state" into a separate box and keeping a very careful eye on it is important)