Starting from
https://www.postgresql.org/message-id/CAEepm%3D2vORBhWQZ1DJmKXmCVi%2B15Tgrv%2B9brHLanWU7XE_FWxQ%40mail.gmail.com
Here is a patch trying to implement what was proposed by Tom Lane:
"What we could/should do instead, I think, is have pgss_planner_hook
make its own pgss_store() call to log the planning time results
(which would mean we don't need the added PlannedStmt field at all).
That would increase the overhead of this feature, which might mean
that it'd be worth having a pg_stat_statements GUC to enable it.
But it'd be a whole lot cleaner."
Now:
pgss_post_parse_analyze, initialize pgss entry with sql text,
pgss_planner_hook, adds planning_time and counts,
pgss_ExecutorEnd, works unchanged.
but doesn't include any pg_stat_statements GUC to enable it yet.
note: I didn't catch the sentence "which would mean we don't need the added PlannedStmt field at all".
Regarding initial remark from Thomas Munro:
"I agree with the sentiment on the old thread that
{total,min,max,mean,stddev}_time now seem badly named, but adding
execution makes them so long... Thoughts?"
What would you think about:
- userid
- dbid
- queryid
- query
- plans
- plan_time
- {min,max,mean,stddev}_plan_time
- calls
- exec_time
- {min,max,mean,stddev}_exec_time
- total_time (being the sum of plan_time and exec_time)
- rows
- ...
Regards
PAscal
