Yep, interesting. Checked with PostgreSQL 10.1.
=> select *,random() from generate_series(1,10);
generate_series | random
-----------------+-------------------
1 | 0.308531506918371
2 | 0.126279713585973
3 | 0.984668150078505
4 | 0.884970095474273
5 | 0.692738385871053
6 | 0.290897831786424
7 | 0.914066118188202
8 | 0.031909613404423
9 | 0.574441066011786
10 | 0.631192437838763
(10 rows)
=> select *,(select * from random()) from generate_series(1,10);
generate_series | random
-----------------+--------------------
1 | 0.0718352268449962
2 | 0.0718352268449962
3 | 0.0718352268449962
4 | 0.0718352268449962
5 | 0.0718352268449962
6 | 0.0718352268449962
7 | 0.0718352268449962
8 | 0.0718352268449962
9 | 0.0718352268449962
10 | 0.0718352268449962
(10 rows)
=> select *,(select random()) from generate_series(1,10);
generate_series | random
-----------------+-------------------
1 | 0.848611807450652
2 | 0.848611807450652
3 | 0.848611807450652
4 | 0.848611807450652
5 | 0.848611807450652
6 | 0.848611807450652
7 | 0.848611807450652
8 | 0.848611807450652
9 | 0.848611807450652
10 | 0.848611807450652
(10 rows)
Looked like random() is "volatile", but in subselect it works like "stable".
