Re: "Alter system" command documentation

On Mon, Jun 15, 2015 at 2:10 PM, David G. Johnston <
david(dot)g(dot)johnston(at)gmail(dot)com> wrote:

> On Mon, Jun 15, 2015 at 4:58 PM, Euler Taveira <euler(at)timbira(dot)com(dot)br>
> wrote:
>
>> On 15-06-2015 13:32, Jeff Janes wrote:
>> > From http://www.postgresql.org/docs/current/static/sql-altersystem.html
>> >
>> > "Values set with ALTER SYSTEM will be effective after the next server
>> > configuration reload (SIGHUP or pg_ctl reload), or after the next server
>> > restart in the case of parameters that can only be changed at server
>> start."
>> >
>> > Isn't it unfortunate that the command used for changing the
>> > configuration file from within SQL mentions SIGHUP and pg_ctl reload,
>> > but not "select pg_reload_conf();", as a way to activate that change?
>> >
>> SIGHUP is an implementation detail. If I were to change that sentence,
>> my preference is replace SIGHUP with pg_reload_conf.
>> ​
>>
>
> ​I ​
> disagree on SIGHUP being an implementation detail.
> ​
> ​ Using signals in IPC (inter-process communication) is a public API that
> should be documented when applicable. Yes, various command-line and SQL
> commands and functions have been provided so that one does not have to use
> raw signals but they are still part of the public API for the application.
>
> Adding mention of pg_reload_conf may be warranted (probably is but I
> haven't made an independent evaluation) but removing SIGHUP is a separate
> concern and one that doesn't seem necessary.
>

Certainly SIGHUP should be (and is) documented elsewhere in the docs. But
there is no reason for it to be documented on this particular page, which
is about a command we added specifically so you can make config changes
without having to drop into an OS shell. If we are going to mention (on
this particularly page) only some of the ways to trigger a reload,
certainly the way that also doesn't require dropping into an OS shell
should be among the ones included. I don't know that pg_reload_conf()
should be the only one we mention here. But surely it should not be the
only one we don't mention.

The cleanest way would probably be to mention this in addition to the other
two, as to mention only this one might imply the others don't work for this
purpose. We could not mention any of them and let the person go read the
other parts of the docs on their own to find the methods used to trigger a
reload, but that seems legalistic and unhelpful.

In any event, I was mildly annoyed to go read the docs about a feature I
was aware of but never needed to use it before, and have it tell me about
two things which I already knew about but which aren't very useful in this
context (that is, not having convenient shell access at the moment), while
not telling me about the third option which actually is useful in this
context but which I didn't know how to spell (because I almost always use
the other two methods instead--except for right now).

Cheers,

Jeff