Railway Postgres stores its data on a volume, and volume backups are

copy-on-write snapshots. If a scheduled snapshot is taken while the

Postgres service is running and accepting writes, is the resulting

backup transactionally consistent -- i.e. does restoring it give a

database in a clean, committed state -- or is it a crash-consistent

snapshot that Postgres recovers from via WAL replay on startup, as it

would after a power loss? Is there any risk of a restored volume being

unrecoverable or silently corrupt? We're storing financial records and

need to know which guarantee we have.