* Make propellor resistent to changes to shared libraries, such as libffi,
which might render the propellor binary unable to run. This is dealt with
by checking the binary both when running propellor on a remote host,
and by Cron.runPropellor. If the binary doesn't work, it will be rebuilt.
* Note that since a new switch had to be added to allow testing the binary,
upgrading to this version will cause a rebuild from scratch of propellor.
Note that if it fails to spin a host, it will stop. I think this is better
than continuing to the next, because there might be a reason to spin hosts
in some specific order (ie, update dns first and then use it).
When propellor is deployed by uploading the binary, there's no git repo, so
each spin needs to re-upload it to get any config changes. This should be
rare since this is only intended to be used when taking over a host and
getting it properly set up from source, but it still needs to be supported.
Since the containers are no longer on the host list, they were not found
while provisioning, oops.
To fix, had to add to a host's info a map of the containers docked to it.
Unfortunately, that required Propellor.Types.Info be glommed into
Propellor.Types, since it needed to refer to Host.
It was fetching from the central repo, then building that, and then running
the client-to-client git update, and the building after that.
Remove the first build, as all that linking does take time.
Lock a lock file while provisioning inside, otherwise propellor could be
running to init the container when the system has just booted, or the
container was just started from being stopped, and at the same time,
propellor run outside the container chains into it to provision.
Previously, simplesh prevented this in a different way.
Joey Hess
Joey Hess