Hi, On 2023-02-08 15:47:51 +0900, Michael Paquier wrote: > This one was intentional to let extensions play with jumbling of such > nodes, but perhaps you are right that it makes little sense at this > stage. If there is an ask for it later, though.. Using > shared_preload_libraries = pg_stat_statements and compute_query_id = > regress shows that numbers go up to 60% for funcs.c and 30% for > switch.c. Removing nodes like as of the attached brings these numbers > respectively up to 94.5% and 93.5% for a check. With a check-world, I > measure respectively 96.7% and 96.1% because there is more coverage > for extensions, ALTER SYSTEM and database commands, roughly.
Given that we already pay the price of multiple regress runs, and that jumbling is now really a core feature, perhaps we should enable pg_stat_statements in pg_upgrade or 027_stream_regress.pl? I'd hope it wouldn't add a meaningful amount of time? A tiny bit of verification at the end should also be ok. Both pg_upgrade and 027_stream_regress.pl have some advantages. The former would test pg_upgrade interactions with shared_preload_libraries, the latter could do some basic checks of pg_stat_statements on a standby. Greetings, Andres Freund
