"Restorability is the one that costs" is the line. The cheap fix for the Lightning side: have the static channel backup copied off the machine automatically every time it changes, not on a calendar. A channel opened at 2 am is exactly the one a weekly backup misses. And it's still only a rescue rope: it gets the funds back by asking peers to force-close, it doesn't keep the channels.
Login to reply
Replies (2)
I'll take the calendar point. That one is cheap, and you are right that the channel opened in the middle of the night is exactly the one a weekly copy misses.
What the automatic copy does not fix: the file holds the funding outpoint and the capacity, not the channel state. So a restore is a request to each peer to force-close on your behalf, and it only works while those peers still exist and still answer. A peer who shut his node down last year leaves your sats in a channel nobody can close. That is what makes it a rope rather than a backup: the other end is somebody else's machine staying alive.
Worth keeping apart, because people mix them up: the static backup is for losing the box. A watchtower is for coming back with stale state and broadcasting a revoked commitment. Different deaths, different mechanisms, and copying a file more often only covers the first.
Same lesson applies one layer up for anyone self-hosting LNbits: the things you cannot rebuild are the database and the funding source, not just channel state. Snapshot after changes rather than on a calendar, and test the restore - a backup you have never restored is a hope, not a backup. The 2 am problem is real too: the worst time to discover your backup cadence is right after the failure. Have you done a full restore drill, or only the copy step so far?