On Tue, 2021-04-13 at 06:36 -0700, MaXinjian wrote: > > Why do you write the WAL to /tmp/pg_wal, only to later mount that at the > > default location? > > pg_wal dir has size limitation, if wal files are too large, they will be > overwrited, right? No, they won't. You could run out of space on the file system though. > > I see nothing wrong with what you are doing, but I may have got lost in > > your complicated procedure. > > You don't happen to remove "backup_label", do you? > > em, I do remove backup_label... Then that's your problem. That will corrupt your data, because recovery starts from the wrong checkpoint. > 1. It means recovery.conf is not necessary, backup_label is necessary? Yes, exactly. > 2. Which key in backup_label is necessary? The whole file needs to be preserved unchanged, just as it is. Don't mess with that file. > 3. I searched the log, it do has recoveried. > Then, if there is no backup_label, what's the default START WAL LOCATION and > CHECKPOINT LOCATION？ That's the catch. "backup_label" is the *only way* to tell a backup from a crashed PostgreSQL cluster. If there is no "backup_label", PostgreSQL will get the latest checkpoint from the control file (global/pg_control), which may well be later than the checkpoint that started the backup, so you will miss to recover some transactions. Yours, Laurenz Albe -- Cybertec | https://www.cybertec-postgresql.com
