"Evgeny Gridasov" <eugrid(at)fpm(dot)kubsu(dot)ru> writes:
> autovacuum process (when active) did not respond to kill (TERM). Only kill
> -9 helped to stop autovacuum process.
autovacuum does respond to shutdown requests, but in poking at this I
found that btree index vacuuming may fail to notice a pending interrupt
for long periods, if you've got large indexes. I've committed a fix
for that.
regards, tom lane