> On Jun 15, 2021, at 17:30, Peter Geoghegan <pg@xxxxxxx> wrote: > It pretty much works by making the WAL sender process on the primary > look like it holds a snapshot that's as old as the oldest snapshot on > the replica. > > A replica can block VACUUM on the primary *directly* by holding a > table-level lock, though in practice only when somebody on the replica > acquires an AccessExclusiveLock -- which is presumably rare in > practice. Thank you! Attempting to track down a sudden burst of every session waiting on lock_manager LW locks, and part of that is finding the source of locks.
