We at Wix do things at scale: literally thousands of services written in different programing languages must be deployed to production in the most cautious and generic way. Different gradual rollout methods (dc by dc, ab, abc,..) become a necessity. Since these rollouts might take some time, its execution must be durable, and if it fails a system state must be reset to the initial one. There are many off-the-shelf solutions for the problem, like Temporal or ArgoCD, but pretty much all of them only partially solve challenges we face. Thus we had to implement our own Saga-like durable execution engine with unique features. In this talk we will discuss what are these features and why you might need them in your rollout pipelines.