pub fn history_seed_pending(db: &Db) -> Result<bool>Expand description
Whether the persisted history-cohort state is exactly the
post-restore, pre-seed state: both history-floor pipelines
(object_version_by_checkpoint and package_versions) carry the
watermark the restore driver’s finalize stamped, while no other
history pipeline has ever committed.
This is the only state the embedded bootstrap may repair with
seed_history_cohort alone (the crash window between the
restore’s finalize and the history seed): the floor pipelines’
restored rows anchor checkpoint-pinned reads at the target, and
the seed’s rewind makes the backfill cover everything else in
(L, T]. Any other history coverage below the available floor
sits behind a pruned gap the backfill can never fill – seeding
over it would advertise the range above the floor as served while
stale sub-floor rows resolve checkpoint-pinned reads to pre-gap
versions – so such a store must be cleared and re-restored
instead.