Having Chef upgrade itself is problematic and at this point we need to hit the reset button and try something different. Instead of having Chef determine if an upgrade is necessary and then run the upgrade we should have Chef determine if the upgrade is necessary and then kick off a script via a system scheduler to perform the upgrade.
Determine if an upgrade is necessary
Download the new version package
Lay down an upgrade script in /opt/chef_infra_updater or C:\opscode\chef_infra_updater
This will be templated to include values such as the lockfile and the location of the new version package
Schedule a task to run the upgrade script 2 minutes in the future
launchd on macos (1 time daemon using RunAtLoad config)
cron on solaris/aix/freebsd (remove post upgrade)
scheduled task: Windows
systemd timer on Linux
Exit Chef Run with upgrade exit code (yet to be determined)
Script executes:
Kill any running chef processes
Add a run lock file
Delete Chef dir (to avoid chef_gem mess)
Install the new version
Nuke the run lock
Run Chef Infra Client so legacy service based chef-clients can be started back up
Having Chef upgrade itself is problematic and at this point we need to hit the reset button and try something different. Instead of having Chef determine if an upgrade is necessary and then run the upgrade we should have Chef determine if the upgrade is necessary and then kick off a script via a system scheduler to perform the upgrade.