On Tue, Jun 09, 2020 at 01:27:20AM -0400, Tom Lane wrote: ! Adrian Klaver <adrian.klaver@xxxxxxxxxxx> writes: ! > On 6/8/20 7:33 PM, Peter wrote: ! >> That "cp" is usually not synchronous. So there is the possibility ! >> that this command terminates successfully, and reports exitcode zero ! >> back to the Postgres, and then the Postgres will consider that log ! >> being safely away. ! ! > Which is why just following the above command in the docs is: ! > "(This is an example, not a recommendation, and might not work on all ! > platforms.) " ! ! Yeah. There have been discussions about changing that disclaimer to be ! more forceful, because in point of fact a plain "cp" is generally not safe ! enough. You need to fsync the written file, and on many filesystems you ! also have to fsync the directory it's in. It certainly does not need to be "more forceful" - because this is not about behavioural education, like training dogs, horses, or monkeys, and neither do we entertain a BDSM studio. What it needs instead is mention of the magic word "fsync". Because, we already know that - we just need a reminder at the proper place. Systems integrators are professional people. They are not in need of more beating (spell: forceful education), only of substantial technical hints and informations. ! > Generally for peace of mind folks use third party tools like: ! ! +1. Rolling your own archive script is seldom advisable. Well then, using commercial solutions brings it's own problems. E.g., the people I happened to work for often had problems with torsion, which happens when the solution gets longer than, say, twenty meters, and these are walked at high speeds. They didn't have a problem with scripting - rather the opposite, they were happy with it and paid good money for. cheerio, PMc
