Filter disabled -> filter maintenance + filter status conditions

Author: PhilippMatthes

helm/bundles/cortex-nova/templates/pipelines.yaml

@@ -64,7 +64,8 @@ spec:
     - ref: {name: filter-has-requested-traits}
     - ref: {name: filter-has-accelerators}
     - ref: {name: filter-correct-az}
-    - ref: {name: filter-disabled}
+    - ref: {name: filter-status-conditions}
+    - ref: {name: filter-maintenance}
     - ref: {name: filter-external-customer}
     - ref: {name: filter-packed-virtqueue}
     - ref: {name: filter-project-aggregates}

helm/bundles/cortex-nova/templates/steps.yaml

@@ -209,42 +209,36 @@ metadata:
   name: filter-correct-az
 spec:
   operator: cortex-nova
-  # TODO: Remove this database reference once the scheduler
-  # step doesn't need it anymore.
-  databaseSecretRef:
-    name: cortex-nova-postgres
-    namespace: {{ .Release.Namespace }}
   type: filter
   impl: filter_correct_az
   description: |
     This step will filter out hosts whose aggregate information indicates they
     are not placed in the requested availability zone.
-  knowledges:
-    # TODO: Remove this dependency since this is a filter which should
-    # not depend on a potentially non-recent knowledge.
-    # This should be changed to use the hypervisor CRD.
-    - name: host-az
 ---
 apiVersion: cortex.cloud/v1alpha1
 kind: Step
 metadata:
-  name: filter-disabled
+  name: filter-status-conditions
 spec:
   operator: cortex-nova
-  # TODO: Remove this database reference once the scheduler
-  # step doesn't need it anymore.
-  databaseSecretRef:
-    name: cortex-nova-postgres
-    namespace: {{ .Release.Namespace }}
   type: filter
-  impl: filter_disabled
+  impl: filter_status_conditions
   description: |
-    This step will filter out hosts for which the hypervisor status is
-    `disabled`, the hypervisor state is `down`, or the trait
-    `COMPUTE_STATUS_DISABLED` is assigned.
-  # TODO: This step currently depends on traits directly from the datasources.
-  # This should be changed to use the hypervisor CRD.
-  knowledges: []
+    This step will filter out hosts for which the hypervisor status conditions
+    do not meet the expected values, for example, that the hypervisor is ready
+    and not disabled.
+---
+apiVersion: cortex.cloud/v1alpha1
+kind: Step
+metadata:
+  name: filter-maintenance
+spec:
+  operator: cortex-nova
+  type: filter
+  impl: filter_maintenance
+  description: |
+    This step will filter out hosts that are currently in maintenance mode that
+    prevents scheduling, for example, manual maintenance or termination.
 ---
 apiVersion: cortex.cloud/v1alpha1
 kind: Step

internal/scheduling/decisions/nova/pipeline_controller_test.go

@@ -285,11 +285,11 @@ func TestDecisionPipelineController_InitPipeline(t *testing.T) {
 			steps: []v1alpha1.Step{
 				{
 					ObjectMeta: metav1.ObjectMeta{
-						Name: "filter_disabled",
+						Name: "filter_status_conditions",
 					},
 					Spec: v1alpha1.StepSpec{
 						Type: v1alpha1.StepTypeFilter,
-						Impl: "filter_disabled",
+						Impl: "filter_status_conditions",
 					},
 				},
 			},
@@ -319,7 +319,7 @@ func TestDecisionPipelineController_InitPipeline(t *testing.T) {
 					},
 					Spec: v1alpha1.StepSpec{
 						Type: v1alpha1.StepTypeFilter,
-						Impl: "filter_disabled",
+						Impl: "filter_status_conditions",
 						Opts: runtime.RawExtension{
 							Raw: []byte(`{"scope":{"host_capabilities":{"any_of_trait_infixes":["TEST_TRAIT"]}}}`),
 						},
@@ -337,7 +337,7 @@ func TestDecisionPipelineController_InitPipeline(t *testing.T) {
 					},
 					Spec: v1alpha1.StepSpec{
 						Type: v1alpha1.StepTypeFilter,
-						Impl: "filter_disabled",
+						Impl: "filter_status_conditions",
 						Opts: runtime.RawExtension{
 							Raw: []byte(`invalid json`),
 						},

internal/scheduling/decisions/nova/plugins/filters/filter_disabled.go

@@ -0,0 +1,65 @@
+// Copyright SAP SE
+// SPDX-License-Identifier: Apache-2.0
+
+package filters
+
+import (
+	"context"
+	"log/slog"
+
+	api "github.com/cobaltcore-dev/cortex/api/delegation/nova"
+	"github.com/cobaltcore-dev/cortex/internal/scheduling/lib"
+	hv1 "github.com/cobaltcore-dev/openstack-hypervisor-operator/api/v1"
+)
+
+type FilterMaintenanceStep struct {
+	lib.BaseStep[api.ExternalSchedulerRequest, lib.EmptyStepOpts]
+}
+
+// Check that the maintenance spec of the hypervisor doesn't prevent scheduling.
+func (s *FilterMaintenanceStep) Run(traceLog *slog.Logger, request api.ExternalSchedulerRequest) (*lib.StepResult, error) {
+	result := s.PrepareResult(request)
+
+	hvs := &hv1.HypervisorList{}
+	if err := s.Client.List(context.Background(), hvs); err != nil {
+		traceLog.Error("failed to list hypervisors", "error", err)
+		return nil, err
+	}
+
+	flagsPreventingScheduling := map[string]bool{
+		hv1.MaintenanceUnset:       false,
+		hv1.MaintenanceManual:      true,
+		hv1.MaintenanceAuto:        false,
+		hv1.MaintenanceHA:          false,
+		hv1.MaintenanceTermination: true,
+	}
+
+	var hostsReady = make(map[string]struct{})
+	for _, hv := range hvs.Items {
+		preventScheduling, ok := flagsPreventingScheduling[hv.Spec.Maintenance]
+		if !ok {
+			traceLog.Info(
+				"hypervisor has unknown maintenance flag, filtering host",
+				"host", hv.Name, "maintenance", hv.Spec.Maintenance,
+			)
+			continue
+		}
+		if preventScheduling {
+			traceLog.Info(
+				"hypervisor maintenance flag prevents scheduling, filtering host",
+				"host", hv.Name, "maintenance", hv.Spec.Maintenance,
+			)
+			continue
+		}
+		hostsReady[hv.Name] = struct{}{}
+	}
+
+	traceLog.Info("hosts passing maintenance filter", "hosts", hostsReady)
+	for host := range result.Activations {
+		if _, ok := hostsReady[host]; ok {
+			continue
+		}
+		delete(result.Activations, host)
+	}
+	return result, nil
+}