Kueue v1alpha1 API

spec defines the desired state of the CapacityProvider.

status defines the capacity observed and published by the provider.

spec defines the desired state of the DynamicQuotaOrchestrator.

status defines the observed state of the DynamicQuotaOrchestrator.

spec defines the preemption rules of the PreemptionConfig.

providers lists CapacityProvider objects consumed by this DQO.

name identifies a CapacityProvider.

effectiveCapacityMultiplier specifies the multiplier applied to the discovered capacity from this provider. It defaults to 1.

subtreeRootQuotaRef identifies the root of the quota subtree.

kind indicates the kind of the quota node, i.e. ClusterQueue or Cohort.

name indicates the name of the quota node, i.e. ClusterQueue or Cohort.

flavors reports capacity for each flavor and resource. The list must contain at least one flavor. It should contain the same flavors as spec.orchestratedFlavors. An empty resources map means zero capacity for all resources of the flavor.

name identifies the ResourceFlavor whose capacity is reported.

resources contains total capacity by resource name. The provider orchestrates all resources of the flavor: a resource that is not listed, including when the map is empty, has zero capacity.

name identifies the ResourceFlavor managed by this provider.

apiGroup is the group for the resource being referenced.

kind is the type of the resource being referenced.

name is the name of the resource being referenced.

orchestratedFlavors identifies the ResourceFlavors for which this provider may publish capacity. DQO ignores entries in status.capacity.flavors whose names are not listed here. The provider orchestrates all resources of the listed flavors: capacity that is not reported for a listed flavor is zero.

controllerName identifies the controller publishing capacity. This field is immutable.

parameters optionally references implementation-specific configuration. DQO does not read or validate the referenced object.

conditions represents the current state of this provider.

capacity is the normalized capacity published by the provider.

capacityDiscovery specifies capacity aggregation.

capacityDistribution specifies how aggregated capacity is distributed. When omitted, the DQO is discovery-only: it reports aggregated capacity but does not write effectiveQuotas status.

conditions represents the current state of the DQO.

effectiveCapacity is the combined capacity from the referenced providers. Each provider's capacity is multiplied by its effectiveCapacityMultiplier. DQO clears this field if any referenced provider is missing or does not have CapacitySynchronized=True.

flavors contains the total capacity for each flavor and resource. It includes all flavors listed in spec.orchestratedFlavors of the referenced providers, up to 128 distinct flavors. If the referenced providers together orchestrate more distinct flavors, the DQO sets EffectiveCapacityComputed=False with reason AggregationFailed and clears status.effectiveCapacity. An empty resources map means zero capacity for all resources of the flavor.

name identifies the ResourceFlavor whose capacity is reported.

resources contains total capacity by resource name. A resource that is not listed, including when the map is empty, has zero capacity for the flavor.

trigger specifies the prerequisite for contributing candidates.

Possible values are:

• Always: contributes matching candidates unconditionally.
• InsufficientQuota: contributes matching candidates only if preempting baseline candidates does not yield sufficient quota to admit the preemptor workload.
• QuotaFeasibleAndInsufficientTopology: contributes matching candidates only if quota is feasible for the entire preemptor under at least one eligible flavor assignment (after preempting baseline candidates and any candidates from InsufficientQuota rules), but the workload cannot be admitted because no eligible flavor assignment satisfies its topology requirements.

Baseline candidates are the deduplicated union of:

• candidates selected by the preemptor's ClusterQueue.spec.preemption policy;
• candidates selected by applicable rules in the referenced PreemptionConfig whose activationPolicy.trigger is Always.

key is the label key that stores the integer value in the workload that will be used for candidate selection.

fallbackValue is used when a workload does not have the label key or the value under the key cannot be parsed as an integer. If not specified, workloads without the label or with a label value not parsable as int are treated as incomparable, and therefore excluded from preemption candidates.

comparison defines how the candidate's label value compares to the preemptor's.

minValue specifies the lowest label value a candidate workload can have to be considered for preemption. If not specified, no lower bound is enforced.

maxValue specifies the highest label value a candidate workload can have to be considered for preemption. If not specified, no upper bound is enforced.

scope specifies the queue or cohort relation boundary of candidates to the preemptor workload.

clusterQueueSelector defines label selector constraints on candidate ClusterQueues. Accepts all if not set.

labelSelector defines label selector constraints on candidate Workloads. Accepts all if not set.

numericLabels defines rules for filtering candidates using custom numeric labels on the Workload resource. Multiple numeric labels are joined using AND-rule (all have to be satisfied). Accepts all if not set.

priority defines the requirements for the priority of candidates. Workloads not matching those requirements will not be considered as preemption candidates. If nil, no priority requirements are enforced.

name is the identifier of the preemption rule.

preemptorSelector is a label selector indicating which workloads can trigger preemptions using this rule. Accepts all workloads if not set.

activationPolicy determines when this rule contributes matching candidates to preemption evaluation.

candidateSelectors specifies the selection rules for workloads that are candidates for preemption. Candidates resulting from multiple selectors are summed into one set. No selectors result in an empty candidate set, thereby disallowing any preemptions with this rule.

mode specifies whether priority comparison uses base or boosted (effective) priority.

comparison defines how the candidate's priority compares to the preemptor's priority. For example, "LessThan" means that only workloads with lower priority will be allowed as preemption candidates.

rules specifies preemption candidate selection rules.