[OpenAFS-devel] Re: request for code review

[Adam Megacz]

Jeffrey Hutzelman <jhutz@cmu.edu> writes:
> I think I'm in favor of handling TERM by shutting down cleanly.
> However, it's worth noting that the bosserver _is_ a process
> supervisor, and may legitimately take considerably longer to shut down
> cleanly than some tools are willing to wait.

Agreed.  I have runit in mind, which (correctly) waits as long as it
takes.  I think daemontools and that thing solaris uses share this
property, but I'd have to check.


Derrick Brashear <shadow@gmail.com> writes:
> For instance, up to 30 minutes for the fileserver to shut down.
> You're impatient? Well, how should we handle it?

Upon reciept of SIGTERM, bosserver should initiate an orderly shutdown.
This will take as long as it takes.  Bossever makes no promise about how
quickly it will finish, although it should not unnecessarily delay the
process or allow itself to become stuck indefinitely.

But what really matters at the end of the day is that we don't wind up
in a situation where bosserver has died but its children are still
alive.  That can lead to really bad stuff.

  - a