Hi Julien, thank you for your contribution. These changes look fine for me, it compiled well and existing tests passed. So I am going to merge it

BTW, I have found existing tests to be very limited, so I added some trivial checks just to be sure that all pg_wait_sampling user functions work well and return something —a1a340f.

However, it fails for 9.6 —https://travis-ci.org/github/postgrespro/pg_wait_sampling/jobs/701312916#L641

I checked it manually with REL9_6_STABLE and found that pg_wait_sampling_get_profile / pg_wait_sampling_get_history always return an empty set for me, pg_wait_sampling tables are always empty as well. Shared pg_wait_sampling library is loaded and I do everything in the same way as for the newer versions.

@rjuju, have you ever used pg_wait_sampling with 9.6? I mean the current master and some of the latests releases.

I just tried on 9.6 (commit 22dcb6c206) and pg_wait_sampling commit4c7549e:

I did a lock table t1 and select * from t1 on another session, then

=# select version();                                                            version-------------------------------------------------------------------------------------------------------------------------------- PostgreSQL 9.6.17@22dcb6c206 on x86_64-apple-darwin18.7.0, compiled by Apple clang version 11.0.0 (clang-1100.0.33.17), 64-bit(1 row)=# select * from pg_wait_sampling_get_history() limit 1;  pid  |              ts               | event_type |  event   | queryid-------+-------------------------------+------------+----------+--------- 28566 | 2020-06-23 17:35:51.063678+02 | Lock       | relation |       0(1 row)=# select * from pg_wait_sampling_get_profile();  pid  | event_type |  event   | queryid | count-------+------------+----------+---------+------- 28566 | Lock       | relation |       0 | 12799(1 row)

I'm not sure why you got failure on 9.6 and 9.6 only.

Thanks! OK, I got it. In the newer versions (>9.6) there are idle state wait events like Activity, Client (PG_WAIT_ACTIVITY, PG_WAIT_CLIENT), so pg_wait_sampling reports are always non-empty. However, in the 9.6 there are none of them, so you have to create an explicit lock waiting event like you described to get some info in the reports. So the problem was with my new test

Oh indeed, that makes sense!

Thanks a lot for investigating and merging!