On Jan 27, 2008, at 10:56 AM, Tommy Nordgren wrote:
I need to run a custom program that I expect to need several weeks
of (wall clock) time to complete.
How can I set up and run so that I can save state on any signal, and
having my program automatically resumed
on start up, with as little extra work as possibly when resuming.
My program is a pure batch job, by the way.
It's called checkpoint/restart, and you can do it in user space using
an appropriate third party library (such as "Libtckpt"). There are
bunches of other options, but that the first one that comes up in
google, and it's one I played with once upon a time.
This is not a service offered by the OS, since it has no idea what
passwords you used for all your network connections, network shares,
or how to reopen some types of decriptors (e.g. those to named pipes
in /var/run or /tmp, which no longer exist, socket connections to
servers that are now down or were over a VPN that you didn't start
back up, firewire disks that have since been unplugged from the
machine, etc.).
For example:
cat /tmp/foo | my_batch_program
How would you reopen descriptor 0 at the right place?
This is a case where you are up against the halting problem, and it's
better to use a library and constrain your applications behaviour to
only those things that the library knows how to restart, rather than
trying to solve the problem generally for all possible applications.
-- Terry
_______________________________________________
Do not post admin requests to the list. They will be ignored.
Darwin-dev mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
http://lists.apple.com/mailman/options/darwin-dev/email@hidden
