On Fri, Apr 29, 2016 at 1:38 PM, Bruce Momjian <bruce@momjian.us> wrote:
On Fri, Apr 29, 2016 at 01:36:35PM +0200, Magnus Hagander wrote: > > > http://www.postgresql.org/docs/devel/static/functions-admin.html > > > > > > Note that it's technically backwards compatible, however, the existing > > > exclusive backup approach is bad and may mean that the database doesn't > > > automatically recover from a crash which happens while the backup is > > > running. All backup tools should be updated to the new APIs to avoid > > > that risk. > > > > Is this risk new? Is it documented? > > No, it's not new. I'm not sure off-hand if it's documented. > > > It is not new. > > And no it is at least not well documented. But there are other parts which > aren't as well - such as the requirement to actually run pg_stop_backup() even > in the case you want to abort a backup, or your next one will fail. Basically > the old method is quite fragile and no, we haven't documented that properly > previously.
Well, if the API was in wide use and will be supported for several more years, we had better document it, no?
Absolutely. That was part of that "bigger backup docs rewrite" that hasn't happened yet, but hopefully will before the final release.