Actually, I think this may be the way Oracle Hot Backups work. It was my impression that feature temporarily suspends writes into a specific tablespace so you can take a snapshot of it. It has been a few years since I've had to do Oracle work though and I could be mis-remembering. People may be confusing Oracle and PostgreSQL.
>> Just to make sure anyone reading the mailing list archives isn't >> confused, running pg_start_backup does *not* make PG stop writing to >> BASEDIR (or DATADIR, or anything, really). PG *will* continue to write >> data into BASEDIR after pg_start_backup has been called. > > > > Correct. Unfortunately it is a very common myth that it does cause > PostgreSQL to stop writing to the base dir.
Never heard that one before. Wow. Who's been saying that?
It's taken me years to hunt down all invalid backup memes and terminate them.
Never fails to surprise me how many people don't read the docs.
I've seen it on stackexchange, and a few times on the pgsql mailing lists, and talking to people in person. I've never traced it back some "authoritative" source who is making the claim, I think many people just independently think up "How would I implement pg_start_backup if I were doing it" and then come up with the same false conclusion, and then all reinforce each other.
I don't think the docs are particularly clear on this. There is the comment "Some file system backup tools emit warnings or errors if the files they are trying to copy change while the copy proceeds. When taking a base backup of an active database, this situation is normal and not an error" but the reader could think that comment could apply to any of the files in the datadirectory (in particular, pg_xlog), and could think that it doesn't apply to the files in datadirectory/base in particular. In other words, once they form the wrong understanding, the docs (if read) don't force them to change it, as they could interpret it in ways that are consistent.
Of course the docs aren't a textbook and aren't trying to fully describe the theory of operation; just give the people a recipe they can follow. But people will make inferences from that recipe anyway. I don't know if it is worth trying preemptively dispel these mistakes in the docs.
