On Sun, Aug 9, 2015 at 10:23 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> If there's actually a use case for that sort of thing, I would vote
> for moving the jumble-calculation code into core
I think that there'd be a good case for doing that for several other
reasons. It would be great to have a per-queryId
log_min_duration_statement, for example.
Most likely, this would work based on a system of grouping queries.
One grouping might be "SLA queries", that the user hopes to prevent
ever hobbling loading the application's home page. Outliers in this
group of queries are far more interesting to the user than any other,
or expectations are in some other way quite different from the
average. At the same time, maybe for reporting queries 60 seconds is
considered an unreasonably long time to wait.
--
Peter Geoghegan