On 03/31/22 15:11, Richard W.M. Jones wrote:
So this patch unfortunately isn't correct for a few reasons:
(1) Non-Linux/BSD OSes don't support a concept like --exit-with-parent. nbdkit will exit with an error if you try to use this feature on such OSes. You can test if --exit-with-parent is supported before trying to use it by:
nbdkit --exit-with-parent --version
which will exit with an error if the feature is not supported.
However this isn't really a major issue for us. The bigger one is:
(2) --exit-with-parent is not reliable. It's best thought of as a best-effort, fallback attempt to kill nbdkit. It requires that nbdkit issues a prctl(2), so any time that the parent (virt-p2v) dies before nbdkit issues the prctl, nbdkit will not get cleaned up.
So I would say to implement this you'll have to:
- Check if “system ("nbdkit --exit-with-parent --version")” returns 0.
- If so, add the --exit-with-parent flag (otherwise, don't).
- Keep the PID tracking / kill code that already exists.
I've been completely prepared to just drop this patch. In fact, I started writing the question "what does --exit-with-parent gain us?" in response to your <https://listman.redhat.com/archives/libguestfs/2022-March/028475.html>. But then as I progressed with my counter-arguments, I just figured I'd pose my question as a patch. And yes, that's the crux: if we need to keep the PID tracking / killing / reaping (and I agree that we do!), then what does "--exit-with-parent" buy us? I don't think it improves anything. So if I can't remove the PID tracking (and I agree I shouldn't), I'd just forget about "--exit-with-parent" and drop this patch altogether. Thanks! Laszlo