I guess this cannot be pointed out too often ;) I'm not intending to do this but the behaviour of the system with
fsync=offled me to further understand that the bottleneck had to be CPU-related rather than I/O-related. Of course, I
neverswitched off fsync on the production system.
Best wishes,
________________________________________
From: Laurenz Albe [laurenz.albe@cybertec.at]
Sent: Wednesday, August 28, 2019 05:45
To: Holtgrewe, Manuel; Luca Ferrari
Cc: pgsql-general
Subject: Re: [ext] Re: Pointers towards identifying bulk import bottleneck (walwriter tuning?)
Holtgrewe, Manuel wrote:
> Switching off fsync leads to a drastic time improvement but still
> higher wall-clock time for four threads.
Don't do that unless you are ready to start from scratch with a new
"initdb" in the case of a crash.
You can do almost as good by setting "synchronous_commit = off",
and that is crash-safe.
Yours,
Laurenz Albe
--
Cybertec | https://www.cybertec-postgresql.com