Inconsitancies in pg_stat_bgwriter and pg_stat_database returned values

Hello,

On one particular PG12.3 instance (same behaviour with PG12.4) we are experimenting occasional strange values in the above tables.

For instance looping (10s. period) over this request:

SELECT current_timestamp, checkpoints_timed,
checkpoints_req,
checkpoint_write_time,
checkpoint_sync_time,
buffers_checkpoint,
buffers_clean,
maxwritten_clean,
buffers_backend,
buffers_backend_fsync,
buffers_alloc
FROM pg_stat_bgwriter

Gives:

020-09-16 13:41:39.781609+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850
2020-09-16 13:41:49.815612+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850
2020-09-16 13:41:59.851812+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850
2020-09-16 13:42:09.89053+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850
2020-09-16 13:42:19.923291+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850
2020-09-16 13:42:29.959282+00,3291,227,1278516493,26327,66708986,1743379,16800,253585756,0,59080296
2020-09-16 13:42:39.98198+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850
2020-09-16 13:42:50.028727+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850
2020-09-16 13:43:00.067928+00,3291,227,1278516493,26327,66709285,1743379,16800,253585756,0,59080296
2020-09-16 13:43:10.090266+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850
2020-09-16 13:43:20.140272+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850
2020-09-16 13:43:30.170372+00,3291,227,1278516493,26327,66709585,1743379,16800,253585756,0,59080296
2020-09-16 13:43:40.204747+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850
2020-09-16 13:43:50.239374+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850
2020-09-16 13:44:00.26905+00,3291,227,1278516493,26327,66709884,1743379,16800,253585756,0,59080296
2020-09-16 13:44:10.314927+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850
2020-09-16 13:44:20.351091+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850
2020-09-16 13:44:30.408797+00,3291,227,1278516493,26327,66710184,1743379,16800,253585756,0,59080296
2020-09-16 13:44:40.44833+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850
2020-09-16 13:44:50.484298+00,438,2,10095976,0,611207,870625,8695,612255,0,39028850

One can see that some columns that are supposed to increase only are not always. For instance checkpoints_timed suddenly jumps from 438 to 3291, then back to 438.

We experiment the same issue in pg_stat_database.

Thanks