Therefore, rather than "improving" pg_usleep (and uglifying its API),
the right answer is to fix parallel vacuum leaders to not depend on
pg_usleep in the first place. A better idea might be to use
pg_sleep() or equivalent code.
Yes, that is a good idea to explore and it will not require introducing
an awkward new API. I will look into using something similar to
pg_sleep.
Looking through the history of the sleep in vacuum_delay_point, commit
720de00af49 replaced WaitLatch with pg_usleep to allow for microsecond
sleep precision [1].
Thomas has proposed a WaitLatchUs implementation in [2], but I have not
yet tried it.
So I see there are 2 possible options here to deal with the interrupt of a
parallel vacuum leader when a message is sent by a parallel vacuum worker.
Option 1/ something like my initial proposal which is
to create a function similar to pg_usleep that is able to deal with
interrupts in a sleep. This could be a function scoped only to vacuum.c,
so it can only be used for vacuum delay purposes.
——
Option 2/ to explore the WaitLatchUs implementation by
Thomas which will give both a latch implementation for a sleep with
the microsecond precision.
It is worth mentioning that if we do end up using WaitLatch(Us) inside
vacuum_delay_point, it will need to set only WL_TIMEOUT and
WL_EXIT_ON_PM_DEATH.
i.e.
(void) WaitLatch(MyLatch, WL_TIMEOUT| WL_EXIT_ON_PM_DEATH,
msec
WAIT_EVENT_VACUUM_DELAY);
This way it is not interrupted by a WL_LATCH_SET when a message
is set by a parallel worker.
——
Ultimately, I think option 2 may be worth a closer look as it is a cleaner
and safer approach, to detect a postmaster death.
Thoughts?
[1] https://postgr.es/m/CAAKRu_b-q0hXCBUCAATh0Z4Zi6UkiC0k2DFgoD3nC-r3SkR3tg%40mail.gmail.com
[2] https://www.postgresql.org/message-id/CA%2BhUKGKVbJE59JkwnUj5XMY%2B-rzcTFciV9vVC7i%3DLUfWPds8Xw%40mail.gmail.com
