Curious Designs

[carpnick]

We have looked into this issue extensively. We are not using a polling mechanism, but using knife winrm to execute chef-client runs. We are using a special exit code to determine if we are rebooting. Then the automated system on the other side will get that exit code and know what to do with it.

Basically we are using exit codes to notify the runner to say, this is expected, versus using straight raise, because we would have to parse the output to figure out if it was expected.

I would be willing to share some of what we have started doing if interested. I haven't gotten to the point of open sourcing both sides yet, but I could easily enough include the LWRPs. FYI - we currently depend on the windows reboot handler.

[vinyar]

@carpnick this is great. Thank you. I feel we may have chatted at ChefConf, because I heard someone doing the exact same thing.

[carpnick]

@vinyar We might of. Talked to alot of people at ChefConf. Basically this code, is in an LWRP we have that exits. We schedule a delayed reboot right before hand so we can guarantee we are the latest reboot schedule (Delayed notifications bump into eachother when setting common node run state)

action :Exit do
      rubyblock=Chef::Resource::RubyBlock.new("#{@new_resource.name}_chefcore_rubyblock", @run_context)
      rubyblock.block do
           puts "Stopping chef run for an immediate reboot";       
           exit 250;     
          end           
      rubyblock.run_action(:run)
end