Uploaded image for project: 'Puppet'
  1. Puppet
  2. PUP-6885

Remove warning about server_list and server conflicting



    • Improvement
    • Status: Closed
    • Normal
    • Resolution: Fixed
    • None
    • PUP 5.5.12, PUP 6.0.7
    • None
    • Known Issue
    • Conflict warning between `server` and `server_list` has been removed. There are scenarios where it is appropriate to have both `server` and `server_list` defined in a config.


      When we added server_list, it was under the idea that it was a path to deprecating the server setting. Currently, a warning is printed at Puppet startup when both settings are set. In addition, for non-agent commands, we will use one of an edpoint-specific setting, the first entry of server_list, or the server setting, depending on the particular command and how it interacts with the netcode.

      All of this is super complicated, and makes things bad for users for a number of reasons. The big ones are:

      • Our netcode is incredibly complex, and in many ways smeared across several layers of abstraction. This makes it very hard to understand all the ways that the server setting could sneak in
      • In the PE HA world, the server_list setting will be managed by puppet, but because we have not previously managed the server setting, we can't safely remove that old setting (since that might conflict with a user's resource that manages server).

      My proposal is to make server_list apply only to agent runs. The big user-visible changes are:
      *remove the warning when both settings are set, and instead print a notice in the agent run that it is preferring server_list

      • Change the low-level netcode to use server instead of the first entry of server_list if failover has not yet happened.

      This would require only minor code changes to a few places:

      • The settings code that handles the warning, to remove that
      • The agent application, to add the notice
      • The low-level netcode stuff, to change default server selection to use server instead of the first entry of server_list when failover has not happened.

      Moving forward longer-term, we should absolutely look in to deprecating server again and cleaning up the network layers to do the right thing more easily. I just don't feel like we built the right thing to be able to do that right now.


        Issue Links



              melissa Melissa Stone
              branan Branan Riley
              0 Vote for this issue
              10 Start watching this issue



                Zendesk Support