What are the benefits of having older revisions of your InferenceService?
- We can roll back to a previous version of our serving object if a mistake was made;
- We can do a gradual rollout of traffic to the newer revision, for example, 90% / 10% split, and test it before rolling it out for everyone.
Knative has revisions enabled by default. This behaviour should be familiar to k8s ops, and we can find it in Kubernetes Deployments. In k8s, Deployments keep a revision history of 10 (by default) so that you can roll back in case of a bug. For more information related to deployments have a look here.
Should I disable this option?
I don't think so. Of course, your specific example might be different, but this is another safety net/layer that allows you to recover quickly from an unexpected bug or mistake.
If you don't need any of the functionality that revisions offer, it's safe to disable it, and we can provide instructions for this.
References:
Comments
0 comments
Please sign in to leave a comment.