I found out in [1] that I am not correctly tracking the workers for
vacuum commands. I trap workers used by
parallel_vacuum_cleanup_all_indexes but not
parallel_vacuum_bulkdel_all_indexes.
Back to the drawing board.
[1]
https://www.postgresql.org/message-id/flat/783bc7f7-659a-42fa-99dd-ee0565644e25(at)dalibo(dot)com
--
Benoit Lobréau
Consultant
http://dalibo.com