AllTopologyAware returns the subset of registered backends that implement TopologyAwareBackend, keyed by name.

AllTopologyAware returns the subset of registered backends that implement TopologyAwareBackend, keyed by name.

ApplyYAMLData applies YAML data to Kubernetes.

CountPodsByPhase counts pods by their phase and returns a PodPhaseCount.

DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new LastError.

Delete explicitly deletes all PCLQ-level ResourceClaims owned by this PodClique. This is required because the PCLQ finalizer's verifyNoResourcesAwaits

DeletePCS deletes a PodCliqueSet by name. NotFound errors are ignored.

ExpectCreations records resource creation expectations for uids for a controlled resource identified by a controlleeKey.

FindPCSGConfigByName finds the matching PodCliqueScalingGroupConfig by PCSG name without requiring the full PCSG object. This is useful in contexts (e

GetAvailableIndices returns the `requiredIndicesCount` available indices for pods. Extracts indices from active pod hostnames and returns the availabl

GetClient returns the unified K8s client struct

GetDescription returns the description from the reconcile step.

GetExistingResourceNames returns the names of PCLQ-level ResourceClaims by selecting on the grove.io/podclique label that Sync stamps on each RC.

GetLogger returns the logger registered with the fake manager

GetPCLQPods lists all Pods controlled by the given PodClique. It queries the podControllerUIDIndexField field index (registered by RegisterPodControll

GetPodGang fetches a PodGang by name and namespace.

HasAnyPackDomain reports whether the constraint has any required or preferred pack domain.

HasErrors returns true if any tasks encountered errors.

NewSetBy converts a slice into a set using keyFunc to derive the key.

RCName returns the deterministic ResourceClaim name for a given base ref and optional replica index. Callers must ensure replicaIndex is non-nil when

RecordErrors records errors encountered during a reconcile run.

Register registers the webhooks with the controller manager.

Setup initializes the shared cluster by connecting to an existing Kubernetes cluster. The cluster must be created beforehand with: - Grove operator d

ShortCircuitReconcileFlow returns true if the reconcile flow should be short-circuited and not continue.

ValidatePCSOnCreate validates all MNNVL annotations on a PodCliqueSet during creation: PCS-level metadata and each PodCliqueTemplateSpec in the spec.

ValidatePodCliqueSet runs scheduler-specific validations on the PodCliqueSet (e.g. TAS required but not supported).

VerifyPCSGReplicasInTopologyDomain verifies that each PCSG replica's pods are in the same topology domain.

WaitForCacheSync blocks until all started informers' caches were synced or the stop channel gets closed.

WaitForCount waits for a specific number of pods to be created with the given label selector. Returns the pod list once the expected count is reached.

buildResource constructs a PodClique resource from templates, setting up metadata, labels, dependencies and environment variables. When pclqExists is

castToPodCliqueSet attempts to cast a runtime.Object to a PodCliqueSet.

Build helm override tuples from grove tuning options. Each tuple is ``(helm_flag, "key=value")`` where ``helm_flag`` is either ``"--set"`` (t

containerHasGPU checks if a single container requests GPU resources. TODO: This check is incomplete - it only looks at Resources.Limits and Resources.

createDeleteTasks creates deletion tasks for PodCliques belonging to specific PCSG replica indices

createPlaceholderSecretIfNotExists creates the webhook TLS secret if it does not already exist. This is needed because the upstream cert-controller (O

-- Helpers -- drainEvents reads all available events from the channel without blocking.

emptyPodClique creates a new empty PodClique resource with basic metadata

fakeBinaryProfile builds a minimal valid google.v1.Profile in binary protobuf. This mirrors what Pyroscope returns with Content-Type: application/prot

--- Helper / assertion functions --- getComputeDomain attempts to get a ComputeDomain by name. Returns error if not found.

getOperatorNamespaceFromFile reads the operator's namespace from the specified file path. This is extracted for testability.

getPodsForClique returns the names of all pods belonging to the specified clique.

Helper function to get the Ready condition's LastTransitionTime from a pod According to the sample files, this is the correct timestamp to check for s

groupPCLQsByLabel groups PodCliques by the value of the specified label key

newUpdateTracker creates a new update tracker

notifyIfAllParentsReady will notify consumers if all PCLQ pods are ready.

pcsgName returns the FQN for a PCSG owned by `testPCSName` at the given replica index, using the same scheme that componentutils.GetExpectedPCSGFQNsPe

pcsgScaleOutPodRefs: 4 pods (sga=3).

reconcileStatus updates the PodCliqueScalingGroup status with current replica counts and conditions

recordError adds an error to the sync flow result

refreshReadyPodsOfPodClique updates the ready pod count for the PodClique that owns the given pod.

requireClusterConfig returns the cached cluster MNNVL configuration, detecting it on first call. This function is safe to call from multiple tests - d

rollbackNodeLabels reverses label mutations on the specified nodes: - Restores original values for labels that existed before mutation. - Removes labe

runScaleTest provides the shared scaffolding (cluster setup, tracker, pprof, output dir, result export) used by every scale test in this package.

skipUnless skips the test if the cluster configuration doesn't match the required values. Usage: config.skipUnless(t, crdSupported, featureEnabled) Us

start begins watching pod events and blocks until the watcher is ready. Uses tc.Ctx, tc.Client, tc.Namespace, and tc.Workload.GetLabelSelector() for w

triggerDeletionOfPodCliques executes concurrent deletion tasks for PodCliques and handles any resulting errors

validatePodCliqueSetWithBackend resolves the scheduler backend for the PCS and runs backend-specific validation. All cliques share the same (resolved)

validateResourceSharingSpecs validates the common fields (Name, Namespace, Scope) across all levels.

verifySinglePCSReplicaUpdatedFirst verifies that a single PCS replica is fully updated before another replica begins. This function ensures strict re

AddDependencies adds all the `to` PodCliques that depend on the `from` PodClique.

BuildSchedulerReferenceMap builds a map from scheduler name to SchedulerTopologyReference pointer.

CleanupStalePerReplicaRCs deletes stale PerReplica ResourceClaims in a single server-side DeleteCollection call. It builds a unified label selector th

CreateOperatorRegistry initializes the operator registry for the PodCliqueSet reconciler

DeepCopy is an autogenerated deepcopy function, copying the receiver, creating a new PodCliqueScalingGroup.

DeleteExpectations removes all expectations stored against a controlleeKey from the store.

DeleteResourceClaim deletes a ResourceClaim by name. NotFound errors are ignored.

EnsureResourceClaim ensures a ResourceClaim exists with the given spec, labels, and owner. ResourceClaim.spec is immutable in Kubernetes, so existing

GenerateBasePodGangName generates a base PodGang name for a PodCliqueSet replica. This is used for PodGangs that are not part of scaled scaling group

GetConditionStatus looks up the conditionType from a slice of conditions. If the condition is found then it returns the status else it will return nil

GetContainerStatusIfTerminatedErroneously gets the first occurrence of corev1.ContainerStatus (across init, sidecar and main containers) that has a no

GetKAIPodGroupsForPCS retrieves all KAI PodGroups for a given PodCliqueSet by label selector.

GetLogs returns a log stream request for a pod container.

GetOrDefault returns the backend registered under name, or the default backend if name is empty.

IsEmptyStringType returns true if value (which is a string or has an underline type string) is empty or contains only whitespace characters.

IsManagedByGrove checks if the pod is managed by Grove by inspecting its labels. All grove managed resources will have a standard set of default label

IsResourceTerminating checks if a deletion timestamp is set. If it is set it returns true else false.