--
You received this message because you are subscribed to the Google Groups "kubernetes-api-reviewers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kubernetes-api-rev...@googlegroups.com.
To post to this group, send email to kubernetes-a...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/kubernetes-api-reviewers/CAO_RewZS6f39PZvgRoyBAEmx%2BormjEMXXN-iYX-pyAhzVs1OVg%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.
Using a new field isn't possible. The same information can't be represented in 2 different fields, and also minReplicas defaults to 1.This technically can't be solved without a new resource type, as we did when introducing ReplicaSet to supersede ReplicationController, but that would cause significant user pain.As much as I dislike case-by-case decisions on issues like this, I think we should allow the change. Zero replicas is valid for workload controllers.I think this change is unlikely to break clients, but it's worth some outreach to ask. We're doing similarly with "kubectl convert" and "kubectl get --export".Some day, I still would like to expose validation information via OpenAPI, which would enable clients to not hardcode such constraints.
--On Tue, Apr 16, 2019 at 4:09 PM 'Tim Hockin' via kubernetes-api-reviewers <kubernetes-a...@googlegroups.com> wrote:Hi all,
Context: https://github.com/kubernetes/kubernetes/pull/74526/files
We certainly consider making validation more restrictive to be a
breaking change. Strictly speaking, we should also consider expanding
validation to be a breaking change -- there could be validators or
tools out there that embed our documented validation rules.
How do we want to handle these changes? In the above PR we have a
relatively simple and straight-forward expansion -- allowing 0 where
it used to not be allowed. Should we block such changes and require a
new field? Or do we accept that this sort of breaking change is
allowed? If so, do we have boundaries for such expansion?
Looking forward to a lively and pedantic discussion.
Tim
--
You received this message because you are subscribed to the Google Groups "kubernetes-api-reviewers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kubernetes-api-rev...@googlegroups.com.
To post to this group, send email to kubernetes-a...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/kubernetes-api-reviewers/CAO_RewZS6f39PZvgRoyBAEmx%2BormjEMXXN-iYX-pyAhzVs1OVg%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.
You received this message because you are subscribed to the Google Groups "kubernetes-api-reviewers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to kubernetes-api-rev...@googlegroups.com.
To post to this group, send email to kubernetes-a...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/kubernetes-api-reviewers/CAKCBhs4ApjE16k4uJM5EXKnM_JB1B8DFeFGPXD%2Bgpk9jTe9J1Q%40mail.gmail.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/kubernetes-api-reviewers/CAB_J3bbhc-g%2B4uF5J1iRWHSB%3DruuHuFic7kRFqbNz4AVZJBgPg%40mail.gmail.com.
Agree a second field doesn't really help (and actually makes things worse if it changes the semantic meaning of a field existing clients look at and think they understand).Expanding to allow 0 where it wasn't previously allowed isn't concerning to me. The types of value expansion that require more thought are ones where the new values could reasonably lead to unsafe use (e.g. envvar character expansion to include ':' or configmap key expansion to include '/')Practically, whenever we want to loosen validation, we need to do so over two releases so that the API server in various scenarios (like HA API clusters) doesn't choke on persisted data (described here).
To view this discussion on the web visit https://groups.google.com/d/msgid/kubernetes-api-reviewers/CAKCBhs5whCmdK9rfnPRMUKYRVpcHwnq%3DgN%3D%3DYgq5HUd2O%3DPMJg%40mail.gmail.com.
So the rules for widening validation are:1. you shouldn't do it UNLESS
By "1-X" did you mean minus or "to"?Probably the latter?
Authoring tools that validate minReplicas might not allow 0 until they get updated.