Hi,
On 2019-08-02 10:54:35 +0200, Julien Rouhaud wrote:
> On Thu, Aug 1, 2019 at 11:05 PM Andres Freund <andres@anarazel.de> wrote:
> >
> > I'm actually quite unconvinced that it's sensible to update the global
> > value for nested queries. That'll mean e.g. the log_line_prefix and
> > pg_stat_activity values are most of the time going to be bogus while
> > nested, because the querystring that's associated with those will *not*
> > be the value that the queryid corresponds to. elog.c uses
> > debug_query_string to log the statement, which is only updated for
> > top-level queries (outside of some exceptions like parallel workers for
> > parallel queries in a function or stuff like that). And pg_stat_activity
> > is also only updated for top level queries.
>
> Having the nested queryid seems indeed quite broken for
> log_line_prefix. However having the nested queryid in
> pg_stat_activity would be convenient to track what is a long stored
> functions currently doing. Maybe we could expose something like
> top_level_queryid and current_queryid instead?
Given that the query string is the toplevel one, I think that'd just be
confusing. And given the fact that it adds *substantial* additional
complexity, I'd just rip the subcommand bits out.
Greetings,
Andres Freund