Em 09/06/2025 11:22, Laurent Bercot wrote:
But really, crond is expected to be a long-running process, and
it is a process dispatcher. Not protecting its process table
from undue growth by an unexpected zombie pile-up is Not A Good
Idea.
The unexpected zombie pile-up is not on crond, which was not
designed to be pid 1 and should not have that expectation placed
upon it.
I thought I was pretty clear that IMHO one should not run a container
without a proper pid 1 in the first place. And I certainly agree that
pid 1 needs a lot more than just being a decent zombie reaper. I
mentioned it needs to properly broadcast signals to its children, for
example (although that's a bad example because you likely want some of
that on crond as well).
But I was not kidding about the "been there, done that" part of the
email, either: IME long-lived programs like crond that spawn other
programs and track children can have bugs in the child tracking just
like they can have bugs in any other code paths.
So, IMO any such programs should be designed to reap unknown dead
processes as a defense against unexpected zombie pile-ups due to the
possibility of *bugs* on the child tracking.
--
Henrique de Moraes Holschuh
www.nic.br
_______________________________________________
busybox mailing list
[email protected]
https://lists.busybox.net/mailman/listinfo/busybox
