On 2020/03/20 3:39, Andres Freund wrote:
> Hi,
>
> On 2020-03-19 17:21:38 +0900, Fujii Masao wrote:
>> Pushed! Thanks!
>
> FWIW, I'm a bit doubtful that incuring the overhead of this by default
> on everybody is a nice thing. On filesystems with high latency and with
> a lot of small relations the overhead of stating a lot of files can be
> almost as high as the actual base backup.
Yeah, so if we receive lots of complaints like that during beta and
RC phases, we should consider to change the default behavior.
Also maybe I should measure how long the estimation takes on the env
where, for example, ten thousand tables (i.e., files) exist, in order to
whether the default behavior is really time-consuming or not?
Regards,
--
Fujii Masao
NTT DATA CORPORATION
Advanced Platform Technology Group
Research and Development Headquarters