Hi,
On Fri, Apr 07, 2023 at 01:32:22PM -0400, Tom Lane wrote:
> "wangw.fnst@fujitsu.com" <wangw.fnst@fujitsu.com> writes:
> > On Tues, Apr 4, 2023 at 23:48 PM Tom Lane <tgl@sss.pgh.pa.us> wrote:
> >> I like the "per eligible process" wording, at least for guc_tables.c;
> >> or maybe it could be "per server process"? That would be more
> >> accurate and not much longer than what we have now.
>
> > Thanks both for sharing your opinions.
> > I agree that verbose descriptions make maintenance difficult.
> > For consistency, I unified the formulas in guc_tables.c and pg-doc into the same
> > suggested short formula. Attach the new patch.
>
> After studying this for awhile, I decided "server process" is probably
> the better term --- people will have some idea what that means, while
> "eligible process" is not a term we use anywhere else. Pushed with
> that change and some minor other wordsmithing.
I stumbled upon this change while looking at the documentation searching
for guidance and what max_locks_per_transactions should be set to (or
rather, a pointer about max_locks_per_transactions not actually being
"per transaction", but a shared pool of roughly
max_locks_per_transactions * max_connections).
While I agree that the exact formula is too verbose, I find the current
wording ("per server process or prepared transaction") to be misleading;
I can see how somebody sees that as a dynamic limit based on the current
number of running server processes or prepared transactions, not
something that is allocated at server start based on some hardcoded
GUCs.
I don't have a good alternative wording for now, but I wanted to point
out that currently the wording does not seem to imply
max_{connection,prepared_transactions} being at play at all. Probably
the GUC description cannot be made much clearer without making it too
verbose, but I think the description in config.sgml has more leeway to
get a mention of max_connections back.
Michael