Add SchedulerHints to server controller #632

eshulman2 · 2026-01-08T16:44:24Z

Add schedulerHints field with support for:

serverGroupRef - schedule server in a server group
differentHostServerRefs / sameHostServerRefs - affinity/anti-affinity with other servers
query, targetCell, differentCell, buildNearHostIP - advanced placement options
additionalProperties - arbitrary scheduler hints

NOTE! this change MOVED the ServerGroupRef inside the ServerSchedulerHints

winiciusallan

Looks good overall! Great job! Just a few comments. Let me know what you think.

api/v1alpha1/server_types.go

internal/controllers/server/tests/server-create-full/00-create-resource.yaml

mandre · 2026-01-15T16:27:38Z

api/v1alpha1/server_types.go

-	// will be created in the server group.
-	// +optional
-	// +kubebuilder:validation:XValidation:rule="self == oldSelf",message="serverGroupRef is immutable"
-	ServerGroupRef *KubernetesNameRef `json:"serverGroupRef,omitempty"`


I agree this should move in the ServerSchedulerHints struct, however, have you considered making this change backward compatible? This would require leaving the field at this level, that we mark as deprecated, and add some glue code to copy its value if the field in ServerSchedulerHints is unset.

Perhaps it's also a good time to introduce more backward incompatible changes in preparation for the next major release 🤷

I was wondering that too. I think it makes sense. Also, we don't need to wait until the next major for get closer to CAPO parity.

We can create a issue so we don't lose the track.

TBH I wasn't sure what is the best approach myself, I mainly wanted to avoid having to consolidate situations where the field was set in both making it confusing to use those fields. I believe that if we are on the verge of a new major we should keep it non-backward compatible and note that on the release and docs. In case the next major release is further away maybe we should have backward compatibility but I suspect this might be confusing to use and might result in unexpected results for costumers. I tend to lean towards keeping it non-backward compatible but I don't have a strong objections for making it BC either.

Well, I can't provide enough feedback here, because I don't know the planned ORC release schedule, but since there are few changes for a new major release, I believe it will take a little longer.

I'll leave it up to both of you.

Let's see what other potentially breaking changes we need to introduce for the CAPO feature parity work before making a decision.

api/v1alpha1/server_types.go

internal/controllers/server/actuator.go

mandre · 2026-01-16T11:14:48Z

api/v1alpha1/server_types.go

+	// buildNearHostIP specifies a subnet of compute nodes to host the server.
+	// +kubebuilder:validation:MaxLength:=255
+	// +optional
+	BuildNearHostIP string `json:"buildNearHostIP,omitempty"`


According to the docs, this must go together with os:scheduler_hints.cidr:

Schedule the server on a host in the network specified with this parameter and a cidr (os:scheduler_hints.cidr)

However we don't provide this option. Should we instead have a reference to a subnet?

Actually in Gopher cloud you provide it as 10.10.10.10/24 in one string and gophercloud split it on its own here , so I think in this case we don't require a separate cider field. the user should just provide it as ip/cider I will expend on it in the doc string above to make it clearer.

As this is a host subnet I believe we shouldn't use subnet ref.

Expended doc string let me know if thats ok now

What do you think of:

buildNearHost: ip: 10.10.10.10 prefixLength: 24

This allows easier validation, and PrefixLength could be marked as optional (according to the docs, it defauls to 24 if not specified).

by validation you are referring to the prefix length min-max 1-32? or something additional?

To me the cider format feels more intuitive so I personally prefer the cider format. It also saves us from some additional string manipulation as this is how Gophercloud expects to receive it. I would prefer keeping it as it is now.

api/v1alpha1/server_types.go

winiciusallan

Thanks for your effort, @eshulman2!

To me, it is very close to a merge state, I just left a comment on the dependency that you added, let me know what you think.

internal/controllers/server/actuator.go

internal/controllers/server/controller.go

- Add SchedulerHints to server controller - enable required nova filters for testing hints NOTE! this change MOVED the ServerGroupRef inside the ServerSchedulerHints

mandre

I believe this is almost ready. I would still like to keep this for a little longer in the queue to make up my mind about the breaking/non-breaking change.

mandre · 2026-01-19T15:22:24Z

api/v1alpha1/server_types.go

+	// buildNearHostIP specifies a subnet of compute nodes to host the server.
+	// +kubebuilder:validation:MaxLength:=255
+	// +optional
+	BuildNearHostIP string `json:"buildNearHostIP,omitempty"`


What do you think of:

buildNearHost: ip: 10.10.10.10 prefixLength: 24

This allows easier validation, and PrefixLength could be marked as optional (according to the docs, it defauls to 24 if not specified).

mandre · 2026-01-19T15:29:13Z

api/v1alpha1/server_types.go

-	// will be created in the server group.
-	// +optional
-	// +kubebuilder:validation:XValidation:rule="self == oldSelf",message="serverGroupRef is immutable"
-	ServerGroupRef *KubernetesNameRef `json:"serverGroupRef,omitempty"`


Let's see what other potentially breaking changes we need to introduce for the CAPO feature parity work before making a decision.

internal/controllers/server/actuator.go

eshulman2 force-pushed the shints branch from 465ee4e to 4f6d197 Compare January 8, 2026 16:45

github-actions bot added the semver:major Breaking change label Jan 8, 2026

eshulman2 force-pushed the shints branch from 4f6d197 to 3274e0d Compare January 11, 2026 08:16

winiciusallan reviewed Jan 13, 2026

View reviewed changes

eshulman2 force-pushed the shints branch 10 times, most recently from 7b0df63 to bdf5e4f Compare January 15, 2026 15:10

mandre reviewed Jan 15, 2026

View reviewed changes

mandre reviewed Jan 16, 2026

View reviewed changes

eshulman2 force-pushed the shints branch 2 times, most recently from f6e8392 to 8894f8f Compare January 18, 2026 08:00

eshulman2 requested a review from mandre January 18, 2026 09:12

eshulman2 force-pushed the shints branch 6 times, most recently from 4bc925c to 54a3ad6 Compare January 19, 2026 10:11

winiciusallan reviewed Jan 19, 2026

View reviewed changes

api/v1alpha1/server_types.go Outdated Show resolved Hide resolved

eshulman2 force-pushed the shints branch 2 times, most recently from ce7d860 to d979876 Compare January 19, 2026 10:55

eshulman2 requested a review from winiciusallan January 19, 2026 12:01

winiciusallan reviewed Jan 19, 2026

View reviewed changes

internal/controllers/server/actuator.go Show resolved Hide resolved

internal/controllers/server/controller.go Outdated Show resolved Hide resolved

Add SchedulerHints to server controller

eac92cb

- Add SchedulerHints to server controller - enable required nova filters for testing hints NOTE! this change MOVED the ServerGroupRef inside the ServerSchedulerHints

eshulman2 force-pushed the shints branch from d979876 to eac92cb Compare January 19, 2026 14:59

eshulman2 requested a review from winiciusallan January 19, 2026 15:12

mandre reviewed Jan 19, 2026

View reviewed changes

Add SchedulerHints to server controller #632

Are you sure you want to change the base?

Add SchedulerHints to server controller #632

Conversation

eshulman2 commented Jan 8, 2026

Uh oh!

winiciusallan left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

eshulman2 Jan 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

winiciusallan left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

mandre left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

eshulman2 Jan 19, 2026 •

edited

Loading

winiciusallan left a comment •

edited

Loading