Remove SIGTERM/SIGINT handling #72
Conversation
The previous code had bugs, see: 62fbff0 The version in RHEL doesn't have this patch, and we're hitting shutdown bugs in RHEL CoreOS. There's no actual good reason though to have a `SIGTERM` handler at all; doing things like `pthread_join()` is pointless becuase the kernel will take care of removing all resources. In fact it's really an anti-pattern to do this, because if every process has these types of things, it makes shutdowns much less efficient as the process needs to be fully scheduled in rather than just having the kernel take care of it.
|
I don't agree in principle with the notion of not having an application gracefully release its resources on shutdown. Instead of removing that code, why not just open a bug with RHEL to backport the needed change to fix the problem? You're going to have to do that anyway to get any change made here today into RHEL |
Can you elaborate more? I think this isn't just bikeshedding, it's gets to a really fundamental principle of systems design. See also e.g. https://en.wikipedia.org/wiki/Crash-only_software - that strongly influenced ostree. I spent a lot of time designing it so that it's totally safe to be interrupted at any point. It was only recently that I was forced to add a |
Well...we could also change the systemd unit easily in RHCOS to use |
|
Sure, we could do that too, but thats not my point. My point is that this is already fixed here in the upstream project, and regardless of what you want the fix to be, it will need to be backported into RHEL. So why not take the path of least resistance and open a RHEL bz requesting that the existing fix be backported? |
|
thank you, I'll take care of it ASAP |
The previous code had bugs, see:
62fbff0
The version in RHEL doesn't have this patch, and we're hitting
shutdown bugs in RHEL CoreOS.
There's no actual good reason though to have a
SIGTERMhandlerat all; doing things like
pthread_join()is pointless becuasethe kernel will take care of removing all resources.
In fact it's really an anti-pattern to do this, because if every
process has these types of things, it makes shutdowns much less
efficient as the process needs to be fully scheduled in rather
than just having the kernel take care of it.