DaemonSet can succeed before rollout finishes

[KnVerey]

Bug report

We've seen DaemonSet deploy succeed when the DS was actually only partially rolled out.

Our current success check looks like this:

    def deploy_succeeded?
      return false unless exists?
      rollout_data["desiredNumberScheduled"].to_i == rollout_data["currentNumberScheduled"].to_i &&
      rollout_data["desiredNumberScheduled"].to_i == rollout_data["numberReady"].to_i &&
      current_generation == observed_generation
    end

A you can see, it isn't looking at updatedNumberScheduled, so it will succeed the first time during the rollout that full availability is (temporarily) reached. Since DS can be configured to roll out in batches, this state can happen multiple times in the middle of the rollout.

@dturn @klautcomputing @mkobetic

[karanthukral]

Merged

[mkobetic]

Ironically, fixing this bug will probably make the short term cluster-services pain worse, but that needs to be fixed elsewhere.