Jelte Fennema-Nio
Jelte Fennema-Nio
Could you share the definition of `public.table2`? So that we can reproduce more easily.
@gokhangulbiz Could you have quick look at this to understand what is going on here? It seems related to the GENERATED AS IDENTITY work that you did.
I think testing only on Linux is fine, but this definitely needs a heavy rebase an the tests need to be changed to use the new Python based testing infrastructure....
To me the main one that is really bad is `query_wait_timeout`. I don't think anyone ever wants the behaviour that pgbouncer waits 2 minutes to even start running the query....
Re-opening this as I really think the current meaning of `avg_wait_time` does not make sense given its name. It's measuring total wait time per second, not average wait time. I...
In a sense yes, you could add it `statement_timeout` to `track_extra_parameters` but you'll still have to install a postgres extension that sets the GUC_REPORT for the `statement_timeout` to actually track...
Yes, setting statement_timeout at the role level would work. You could also have pgbouncer set `statement_timeout` for you using `connect_query`: ```ini [databases] mydb = host=somehost connect_query='set statement_timeout=5s;' ```
1800 is the default Keepalive Interval on Linux, so it seems like a likely culprit.
Some slowdown is expected, but not 40x slower. Up to 2x makes sense, since total traffic is multiplied by 2. PgBouncer and Odyssey don't have this huge slowdown.
I'm 99% sure that this test uses protocol level Parse packets