Skip to content

Commit 4805aa9

Browse files
kannon92kannon92
committed
generate api changes
1 parent a04d83b commit 4805aa9

16 files changed

+681
-273
lines changed

‎config/schemapatch/kueue-operator.crd.yaml

+63-47
Original file line numberDiff line numberDiff line change
@@ -2,7 +2,7 @@ apiVersion: apiextensions.k8s.io/v1
22
kind: CustomResourceDefinition
33
metadata:
44
annotations:
5-
controller-gen.kubebuilder.io/version: v0.17.2
5+
controller-gen.kubebuilder.io/version: v0.4.1
66
name: kueues.operator.openshift.io
77
spec:
88
group: operator.openshift.io
@@ -11,7 +11,7 @@ spec:
1111
listKind: KueueList
1212
plural: kueues
1313
singular: kueue
14-
scope: Namespaced
14+
scope: Cluster
1515
versions:
1616
- name: v1alpha1
1717
schema:
@@ -54,6 +54,46 @@ spec:
5454
required:
5555
- integrations
5656
properties:
57+
gangSchedulingPolicy:
58+
description: |-
59+
gangSchedulingPolicy controls how Kueue admits workloads.
60+
Gang Scheduling is the act of all or nothing scheduling.
61+
Kueue provides this ability.
62+
This field is optional.
63+
type: object
64+
required:
65+
- policy
66+
properties:
67+
byWorkload:
68+
description: |-
69+
byWorkload controls how admission is done.
70+
When admission is set to Sequential, only pods from the currently processing workload will be admitted.
71+
Once all pods from the current workload are admitted, and ready, Kueue will process the next workload.
72+
Sequential processing may slow down admission when the cluster has sufficient capacity for multiple workloads,
73+
but provides a higher guarantee of workloads scheduling all pods together successfully.
74+
When set to Parallel, pods from any workload will be admitted at any time.
75+
This may lead to a deadlock where workloads are in contention for cluster capacity and
76+
pods from another workload having successfully scheduled prevent pods from the current workload scheduling.
77+
type: string
78+
enum:
79+
- Parallel
80+
- Sequential
81+
policy:
82+
description: |-
83+
policy allows you to enable and configure gang scheduling.
84+
This is an optional field.
85+
The allowed values are ByWorkload and Disabled.
86+
The default value will be Disabled.
87+
When set to ByWorkload, this means each workload is processed and considered
88+
for admission as a single unit.
89+
Where workloads do not become ready over time, the entire workload may then be evicted and retried at a later time.
90+
type: string
91+
enum:
92+
- ByWorkload
93+
- Disabled
94+
x-kubernetes-validations:
95+
- rule: 'has(self.policy) && self.policy == ''ByWorkload'' ? has(self.byWorkload) : !has(self.byWorkload)'
96+
message: byWorkload is required when policy is ByWorkload, and forbidden otherwise
5797
integrations:
5898
description: |-
5999
integrations is a required field that configures the Kueue's workload integrations.
@@ -158,6 +198,8 @@ spec:
158198
maxItems: 64
159199
items:
160200
type: object
201+
required:
202+
- key
161203
properties:
162204
key:
163205
description: |-
@@ -175,45 +217,6 @@ spec:
175217
x-kubernetes-validations:
176218
- rule: self.matches(r'^([a-z0-9]([-a-z0-9]*[a-z0-9])?(\.[a-z0-9]([-a-z0-9]*[a-z0-9])?)*/)?([a-z0-9]([-a-z0-9]*[a-z0-9])?)$') && self.size() <= 317
177219
x-kubernetes-list-type: atomic
178-
<<<<<<< HEAD
179-
=======
180-
kueueGangSchedulingPolicy:
181-
description: |-
182-
kueueGangSchedulingPolicy controls how Kueue admits workloads.
183-
Gang Scheduling is the act of all or nothing scheduling.
184-
Kueue provides this ability.
185-
This field is optional.
186-
type: object
187-
properties:
188-
byWorkload:
189-
description: |-
190-
byWorkload controls how admission is done.
191-
When admission is set to Sequential, only pods from the currently processing workload will be admitted.
192-
Once all pods from the current workload are admitted, and ready, Kueue will process the next workload.
193-
Sequential processing may slow down admission when the cluster has sufficient capacity for multiple workloads,
194-
but provides a higher guarantee of workloads scheduling all pods together successfully.
195-
When set to Parallel, pods from any workload will be admitted at any time.
196-
This may lead to a deadlock where workloads are in contention for cluster capacity and
197-
pods from another workload having successfully scheduled prevent pods from the current workload scheduling.
198-
type: string
199-
enum:
200-
- Parallel
201-
- Sequential
202-
x-kubernetes-validations:
203-
- rule: self.policy==ByWorkload
204-
message: byWorkload is only valid if policy equals ByWorkload
205-
policy:
206-
description: |-
207-
policy allows for changing the kinds of gang scheduling Kueue does.
208-
This is an optional field.
209-
The allowed values are ByWorkload and Disabled.
210-
The default value will be Disabled.
211-
ByWorkload allows for configuration how admission is performed
212-
for Kueue.
213-
type: string
214-
enum:
215-
- ByWorkload
216-
- Disabled
217220
preemption:
218221
description: |-
219222
preemption is the process of evicting one or more admitted Workloads to accommodate another Workload.
@@ -224,6 +227,17 @@ spec:
224227
description: |-
225228
preemptionStrategy are the types of preemption kueue allows.
226229
Kueue has two types of preemption: classical and fair sharing.
230+
Classical means that an incoming workload, which does
231+
not fit within the unusued quota, is eligible to issue preemptions
232+
when the requests of the workload are below the
233+
resource flavor's nominal quota or borrowWithinCohort is enabled
234+
on the Cluster Queue.
235+
Fairsharing means that ClusterQueues with pending Workloads can preempt other Workloads
236+
in their cohort until the preempting ClusterQueue
237+
obtains an equal or weighted share of the borrowable resources.
238+
The borrowable resources are the unused nominal quota
239+
of all the ClusterQueues in the cohort.
240+
FairSharing is a more heavy weight algorithm.
227241
type: string
228242
enum:
229243
- Classical
@@ -235,21 +249,23 @@ spec:
235249
This field is optional.
236250
type: object
237251
properties:
238-
queueLabelPolicy:
252+
policy:
239253
description: |-
240-
queueLabelPolicy controls whether or not Kueue reconciles
254+
policy controls whether or not Kueue reconciles
241255
jobs that don't set the label kueue.x-k8s.io/queue-name.
242256
The allowed values are QueueNameRequired and QueueNameOptional.
243-
If set to QueueNameRequired, then those jobs will be suspended and never started unless
244-
they are assigned a queue and eventually admitted. This also applies to
245-
jobs created before starting the kueue controller.
257+
QueueNameOptional means that workloads will be suspended on
258+
creation and a label will be added via a mutating webhook.
259+
QueueNameRequired means that workloads that are managed
260+
by Kueue must have a the label kueue.x-k8s.io/queue-name.
261+
If this label is not present on the workload, then Kueue will
262+
ignore this workload.
246263
Defaults to QueueNameRequired; therefore, those jobs are not managed and if they are created
247264
unsuspended, they will start immediately.
248265
type: string
249266
enum:
250267
- QueueNameRequired
251268
- QueueNameOptional
252-
>>>>>>> 22bc930 (add FairSharing, QueueLabelPolicy and GangScheduling)
253269
logLevel:
254270
description: |-
255271
logLevel is an intent based logging for an overall component. It does not give fine grained control, but it is a

‎config/schemapatch/operator.openshift.io_kueues.yaml

+61-45
Original file line numberDiff line numberDiff line change
@@ -54,6 +54,46 @@ spec:
5454
required:
5555
- integrations
5656
properties:
57+
gangSchedulingPolicy:
58+
description: |-
59+
gangSchedulingPolicy controls how Kueue admits workloads.
60+
Gang Scheduling is the act of all or nothing scheduling.
61+
Kueue provides this ability.
62+
This field is optional.
63+
type: object
64+
required:
65+
- policy
66+
properties:
67+
byWorkload:
68+
description: |-
69+
byWorkload controls how admission is done.
70+
When admission is set to Sequential, only pods from the currently processing workload will be admitted.
71+
Once all pods from the current workload are admitted, and ready, Kueue will process the next workload.
72+
Sequential processing may slow down admission when the cluster has sufficient capacity for multiple workloads,
73+
but provides a higher guarantee of workloads scheduling all pods together successfully.
74+
When set to Parallel, pods from any workload will be admitted at any time.
75+
This may lead to a deadlock where workloads are in contention for cluster capacity and
76+
pods from another workload having successfully scheduled prevent pods from the current workload scheduling.
77+
type: string
78+
enum:
79+
- Parallel
80+
- Sequential
81+
policy:
82+
description: |-
83+
policy allows you to enable and configure gang scheduling.
84+
This is an optional field.
85+
The allowed values are ByWorkload and Disabled.
86+
The default value will be Disabled.
87+
When set to ByWorkload, this means each workload is processed and considered
88+
for admission as a single unit.
89+
Where workloads do not become ready over time, the entire workload may then be evicted and retried at a later time.
90+
type: string
91+
enum:
92+
- ByWorkload
93+
- Disabled
94+
x-kubernetes-validations:
95+
- rule: 'has(self.policy) && self.policy == ''ByWorkload'' ? has(self.byWorkload) : !has(self.byWorkload)'
96+
message: byWorkload is required when policy is ByWorkload, and forbidden otherwise
5797
integrations:
5898
description: |-
5999
integrations is a required field that configures the Kueue's workload integrations.
@@ -158,6 +198,8 @@ spec:
158198
maxItems: 64
159199
items:
160200
type: object
201+
required:
202+
- key
161203
properties:
162204
key:
163205
description: |-
@@ -175,45 +217,6 @@ spec:
175217
x-kubernetes-validations:
176218
- rule: self.matches(r'^([a-z0-9]([-a-z0-9]*[a-z0-9])?(\.[a-z0-9]([-a-z0-9]*[a-z0-9])?)*/)?([a-z0-9]([-a-z0-9]*[a-z0-9])?)$') && self.size() <= 317
177219
x-kubernetes-list-type: atomic
178-
<<<<<<< HEAD
179-
=======
180-
kueueGangSchedulingPolicy:
181-
description: |-
182-
kueueGangSchedulingPolicy controls how Kueue admits workloads.
183-
Gang Scheduling is the act of all or nothing scheduling.
184-
Kueue provides this ability.
185-
This field is optional.
186-
type: object
187-
properties:
188-
byWorkload:
189-
description: |-
190-
byWorkload controls how admission is done.
191-
When admission is set to Sequential, only pods from the currently processing workload will be admitted.
192-
Once all pods from the current workload are admitted, and ready, Kueue will process the next workload.
193-
Sequential processing may slow down admission when the cluster has sufficient capacity for multiple workloads,
194-
but provides a higher guarantee of workloads scheduling all pods together successfully.
195-
When set to Parallel, pods from any workload will be admitted at any time.
196-
This may lead to a deadlock where workloads are in contention for cluster capacity and
197-
pods from another workload having successfully scheduled prevent pods from the current workload scheduling.
198-
type: string
199-
enum:
200-
- Parallel
201-
- Sequential
202-
x-kubernetes-validations:
203-
- rule: self.policy==ByWorkload
204-
message: byWorkload is only valid if policy equals ByWorkload
205-
policy:
206-
description: |-
207-
policy allows for changing the kinds of gang scheduling Kueue does.
208-
This is an optional field.
209-
The allowed values are ByWorkload and Disabled.
210-
The default value will be Disabled.
211-
ByWorkload allows for configuration how admission is performed
212-
for Kueue.
213-
type: string
214-
enum:
215-
- ByWorkload
216-
- Disabled
217220
preemption:
218221
description: |-
219222
preemption is the process of evicting one or more admitted Workloads to accommodate another Workload.
@@ -224,6 +227,17 @@ spec:
224227
description: |-
225228
preemptionStrategy are the types of preemption kueue allows.
226229
Kueue has two types of preemption: classical and fair sharing.
230+
Classical means that an incoming workload, which does
231+
not fit within the unusued quota, is eligible to issue preemptions
232+
when the requests of the workload are below the
233+
resource flavor's nominal quota or borrowWithinCohort is enabled
234+
on the Cluster Queue.
235+
Fairsharing means that ClusterQueues with pending Workloads can preempt other Workloads
236+
in their cohort until the preempting ClusterQueue
237+
obtains an equal or weighted share of the borrowable resources.
238+
The borrowable resources are the unused nominal quota
239+
of all the ClusterQueues in the cohort.
240+
FairSharing is a more heavy weight algorithm.
227241
type: string
228242
enum:
229243
- Classical
@@ -235,21 +249,23 @@ spec:
235249
This field is optional.
236250
type: object
237251
properties:
238-
queueLabelPolicy:
252+
policy:
239253
description: |-
240-
queueLabelPolicy controls whether or not Kueue reconciles
254+
policy controls whether or not Kueue reconciles
241255
jobs that don't set the label kueue.x-k8s.io/queue-name.
242256
The allowed values are QueueNameRequired and QueueNameOptional.
243-
If set to QueueNameRequired, then those jobs will be suspended and never started unless
244-
they are assigned a queue and eventually admitted. This also applies to
245-
jobs created before starting the kueue controller.
257+
QueueNameOptional means that workloads will be suspended on
258+
creation and a label will be added via a mutating webhook.
259+
QueueNameRequired means that workloads that are managed
260+
by Kueue must have a the label kueue.x-k8s.io/queue-name.
261+
If this label is not present on the workload, then Kueue will
262+
ignore this workload.
246263
Defaults to QueueNameRequired; therefore, those jobs are not managed and if they are created
247264
unsuspended, they will start immediately.
248265
type: string
249266
enum:
250267
- QueueNameRequired
251268
- QueueNameOptional
252-
>>>>>>> 22bc930 (add FairSharing, QueueLabelPolicy and GangScheduling)
253269
logLevel:
254270
description: |-
255271
logLevel is an intent based logging for an overall component. It does not give fine grained control, but it is a

0 commit comments

Comments
 (0)