Upstart: feature questions and test code
cjdahlin at ncsu.edu
Wed Jun 11 01:56:37 BST 2008
Garrett Cooper wrote:
> One point of concern... has anyone considered executing upstart in a
> failover fashion at all, i.e. one process runs as pid = 1 and another
> pid = 2, so pid = 1 watches the processes it spawns + the failover
> daemon, while pid = 2 just monitors the pid = 1 copy (seems circular,
> but that way all of the bases would be covered to a certain degree)?
> Or if pid = 1 dies is it considered "game over, system completely
> I know that would make process ownership difficult if nigh /
> completely impossible, but at one point there was some discussion of
> having failover process management daemons.
> Not a likely case, but just something I remembered hearing about from before...
The death of pid 1 causes an immediate kernel panic. This is a kernel
decision and applies to any init daemon, not just upstart.
Like the kernel itself, this is a component that simply can't fail.
More information about the upstart-devel