feat: pipeline and model name validation #5872
Conversation
driev
changed the title
| import "regexp" | ||
|
|
||
| func CheckName(name string) bool { | ||
| ok, err := regexp.Match("^[a-zA-Z0-9][a-zA-Z0-9-_]*[a-zA-Z0-9]$", []byte(name)) |
There was a problem hiding this comment.
The spec on what names are allowed in k8s is this one: https://kubernetes.io/docs/concepts/overview/working-with-objects/names/
k8s will already reject all names that don't obey the regex: [a-z0-9]([-a-z0-9]*[a-z0-9])?(\.[a-z0-9]([-a-z0-9]*[a-z0-9])?)* (this is the regex that kubectl reports when trying to create a resource with an invalid name). In particular, upper-case letters and underscores are also excluded.
We'll still want to do this validation explicitly so that we deal in an uniform way with k8s and non-k8s deployments.
Beyond k8s's existing validation, we want to reserve the . for Seldon -- this is because internally we do header-based routing and a pipeline's name (for example) is referred to as '[name].pipeline`.
I'm not sure why we don't just trim the element after the last dot when post-processing such names (and allow dots in user-provided names). I don't have the context on whether there are any other reasons why we might reserve the ., but perhaps we take this as a separate issue to investigate.
There was a problem hiding this comment.
Yes, some more investigation is necessary. It would be great to have this validation aligned with k8s.
Pipelines and models with dots in their names cause problems, as the scheduler will trim the name from the first occurrence of the dot, when attempting to retrieve the resource.
Adding some validation to model and pipeline names to highlight this scenerio and return descriptive error messages.
Only alphanumeric names with underscores and hyphens are valid.
It's probably worth starting with the
utils/validate_test.gofile.TODO :