Previously, my understanding was that a ray service instantiates one RayCluster and all compute is carried out through head and worker nodes on that ray cluster.
In practice, however, I’ve seen that a RayService may kill a ray cluster (for whatever reason) and then create a new ray cluster.
How often is it expected that a rayservice kills a ray cluster? I’m having issues deploying and hence a new ray cluster gets spun up, but are there any other cases where this happens?