On Thu, Mar 12, 2020 at 10:31 AM, Julien Rouhaud wrote: > > * bufusage still only counts the buffer usage during executor. > > > > Now we have the ability to count the buffer usage during planner but we > keep > > the bufusage count the buffer usage during executor for now. > > The bufusage should reflect the sum of planning and execution usage if > track_planning is enabled. Did I miss something there?
Ah, you're right. I somehow misunderstood it. Sorry for the annoying. > > * We add Assert in pg_plan_query so that query_text will not be NULL > > when executing planner. > > > > There's no case query_text will be NULL with current sources. It is not > > ensured there will be any case query_text will be possibly NULL in the > > future though. Some considerations are needed by other people about > this. > > There's at least the current version of IVM patchset that lacks the querytext. I saw IVM patchset but I thought it is difficult to impose them to give appropriate querytext. > Looking at various extensions, I see that pg_background and pglogical call > pg_plan_query internally but shouldn't have any issue providing the query > text. > But there's also citus extension, which don't keep around the query string at > least when distributing plans, which makes sense since it's of no use and > they're heavily modifying the original Query. I think that citus folks > opinion on > the subject would be useful, so I'm Cc-ing Marco. Thank you for looking those codes. I will comment about this in another mail. -- Yoshikazu Imai
