This is the downstream issue of the upstream problem described at https://github.com/operator-framework/operator-lifecycle-manager/issues/2725 . Basically OLM first creates the operator ReplicaSet without the environment variables defined in the subscription, and then later patches it with those variables. This is inherently racy. In our case we noticed it because RHOAI uses an env variable called DISABLE_DSC_CONFIG which drives the creation of a DSCI object. And sometimes even though we set this variable to true, this DSCI would get created because the default value of DISABLE_DSC_CONFIG is false and the first replicaset will get created with this false value. Only later it gets patched to true, but then it is too late and the DSCI object has already been created.

    Any OCP version, triggered on both 4.16 and 4.17

    It's a race condition, in my test case about 50% of the time

    1.Install any operator subscription and add some config environment variables
    2.Observe two replicasets were generated (the first one without the env variables, the second one with)
    3.
    

  Two replicasets with differing environment variable for the pod definition

   A single replicaset with the environment variables defined in the subscription.

     OLM already has all the needed info to instantiate the deployment with the config env variables from the subscriptions, so it should create a single replicaset from the start.