refactor: scope host details extractor to vmware only

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

@@ -178,15 +178,17 @@ spec:
 apiVersion: cortex.cloud/v1alpha1
 kind: Knowledge
 metadata:
-  name: host-details
+  name: vmware-host-details
 spec:
   schedulingDomain: nova
   extractor:
-    name: sap_host_details_extractor
+    name: vmware_host_details_extractor
   description: |
-    This knowledge extracts SAP-specific host details information.
+    This knowledge extracts details of VMware compute hosts. KVM hosts are
+    served by a dedicated CRD and ironic hosts are excluded.
   recency: "60s"
   dependencies:
     datasources:
       - name: placement-resource-provider-traits
+      - name: placement-resource-provider-inventory-usages
       - name: nova-hypervisors

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

@@ -173,7 +173,7 @@ spec:
       - name: identity-projects
       - name: identity-domains
     knowledges:
-      - name: host-details
+      - name: vmware-host-details
   description: |
     This KPI tracks the resource utilization of projects running VMs on VMware hosts.
 ---
@@ -203,7 +203,7 @@ spec:
   impl: vmware_host_capacity_kpi
   dependencies:
     knowledges:
-      - name: host-details
+      - name: vmware-host-details
       - name: host-utilization
   description: |
     This KPI tracks the capacity and utilization of VMware hosts in terms of CPU, RAM, and disk resources.

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

@@ -9,7 +9,6 @@ spec:
   impl: kvm_host_capacity_kpi
   dependencies:
     knowledges:
-      - name: host-details
       - name: host-utilization
   description: |
     This KPI tracks the total, utilized, reserved and failover capacity of KVM hosts.

internal/knowledge/extractor/controller_test.go

@@ -450,7 +450,7 @@ func TestKnowledgeReconciler_Reconcile_SupportedExtractors(t *testing.T) {
 		"vm_life_span_histogram_extractor",
 		"host_az_extractor",
 		"host_pinned_projects_extractor",
-		"sap_host_details_extractor",
+		"vmware_host_details_extractor",
 	}
 
 	for _, extractorName := range supportedExtractors {

internal/knowledge/extractor/plugins/compute/host_details.go → internal/knowledge/extractor/plugins/compute/vmware_host_details.go

@@ -14,19 +14,17 @@ import (
 	"sigs.k8s.io/controller-runtime/pkg/client"
 )
 
-type HostDetails struct {
+// VMwareHostDetails describes a single VMware compute host. Only VMware hosts
+// are extracted here (KVM hosts are served by a dedicated CRD, and ironic hosts
+// are excluded); the extractor SQL filters everything else out.
+type VMwareHostDetails struct {
 	// Name of the OpenStack compute host.
 	ComputeHost string `db:"compute_host"`
 	// Availability zone of the compute host.
 	AvailabilityZone string `db:"availability_zone"`
 	// CPU Architecture of the compute host.
 	// Can be "cascade-lake" or "sapphire-rapids"
 	CPUArchitecture string `db:"cpu_architecture"`
-	// Hypervisor type of the compute host.
-	HypervisorType string `db:"hypervisor_type"`
-	// Hypervisor family of the compute host.
-	// Can be "kvm" or "vmware"
-	HypervisorFamily string `db:"hypervisor_family"`
 	// Amount of VMs currently running on the compute host.
 	RunningVMs int `db:"running_vms"`
 	// Type of workload running on the compute host.
@@ -45,28 +43,28 @@ type HostDetails struct {
 	// Physical size category of a single host inside a VMware building block,
 	// derived from the memory inventory and rounded to the nearest TiB
 	// (e.g. "4TiB"), or to the nearest GiB (e.g. "512GiB") when smaller than
-	// 1 TiB. Only meaningful for VMware hosts; "unknown" otherwise.
+	// 1 TiB. "unknown" when the memory inventory is missing.
 	PhysicalHostSize string `db:"physical_host_size"`
 }
 
-type HostDetailsExtractor struct {
+type VMwareHostDetailsExtractor struct {
 	// Common base for all extractors that provides standard functionality.
 	plugins.BaseExtractor[
-		struct{},    // No options passed through yaml config
-		HostDetails, // Feature model
+		struct{},          // No options passed through yaml config
+		VMwareHostDetails, // Feature model
 	]
 }
 
-//go:embed host_details.sql
-var hostDetailsQuery string
+//go:embed vmware_host_details.sql
+var vmwareHostDetailsQuery string
 
-// Extract the traits of a compute host from the database.
-func (e *HostDetailsExtractor) Extract() ([]plugins.Feature, error) {
+// Extract the details of the VMware compute hosts from the database.
+func (e *VMwareHostDetailsExtractor) Extract() ([]plugins.Feature, error) {
 	if e.DB == nil {
 		return nil, errors.New("database connection is not initialized")
 	}
-	var hostDetails []HostDetails
-	if _, err := e.DB.Select(&hostDetails, hostDetailsQuery); err != nil {
+	var hostDetails []VMwareHostDetails
+	if _, err := e.DB.Select(&hostDetails, vmwareHostDetailsQuery); err != nil {
 		return nil, err
 	}
 

internal/knowledge/extractor/plugins/compute/host_details.sql → internal/knowledge/extractor/plugins/compute/vmware_host_details.sql

@@ -1,8 +1,11 @@
+-- Details of VMware compute hosts. Only VMware hosts are considered here: KVM
+-- hosts are served by a dedicated CRD, and ironic hosts are excluded. The
+-- WHERE clause keeps rows whose service host matches the VMware naming
+-- convention (nova-compute-%) while dropping ironic ones (nova-compute-ironic-%).
 WITH host_traits AS (
     SELECT
         h.id AS hypervisor_id,
         h.service_host,
-        h.hypervisor_type,
         h.running_vms,
         h.state,
         h.status,
@@ -11,7 +14,9 @@ WITH host_traits AS (
     FROM openstack_hypervisors h
     LEFT JOIN openstack_resource_provider_traits t
         ON h.id = t.resource_provider_uuid
-    GROUP BY h.id, h.service_host, h.hypervisor_type, h.running_vms, h.state, h.status, h.service_disabled_reason
+    WHERE h.service_host LIKE 'nova-compute-%'
+        AND h.service_host NOT LIKE 'nova-compute-ironic-%'
+    GROUP BY h.id, h.service_host, h.running_vms, h.state, h.status, h.service_disabled_reason
 ),
 -- Physical memory size (in MiB) of a single host inside a VMware building block.
 -- A building block pools multiple physical hosts into one resource provider, so
@@ -37,12 +42,6 @@ SELECT
         WHEN ht.traits LIKE '%CUSTOM_HW_SAPPHIRE_RAPIDS%' THEN 'sapphire-rapids'
         ELSE 'cascade-lake'
     END AS cpu_architecture,
-    ht.hypervisor_type,
-    CASE
-        WHEN ht.service_host LIKE 'nova-compute-%' THEN 'vmware'
-        WHEN ht.service_host LIKE 'node%-bb%' THEN 'kvm'
-        ELSE 'unknown'
-    END AS hypervisor_family,
     CASE
         WHEN ht.traits LIKE '%CUSTOM_HANA_EXCLUSIVE_HOST%' THEN 'hana'
         ELSE 'general-purpose'
@@ -67,10 +66,11 @@ SELECT
         WHEN ht.state != 'up' THEN '[down] ' || COALESCE(ht.service_disabled_reason, '--')
         ELSE NULL
     END AS disabled_reason,
-    -- Physical host size category. The size is first rounded to whole GiB; if
-    -- that is at least 1024 GiB it is reported in TiB ("<n>TiB"), otherwise in
-    -- GiB ("<n>GiB"). Rounding to GiB first avoids a value like 1023.6 GiB being
-    -- shown as "1024GiB" instead of "1TiB".
+    -- Physical host size category of a single host inside the building block.
+    -- The size is first rounded to whole GiB; if that is at least 1024 GiB it is
+    -- reported in TiB, otherwise in GiB. Rounding to GiB
+    -- first avoids a value like 1023.6 GiB being shown as "1024GiB" instead of
+    -- "1TiB". "unknown" when the memory inventory is missing.
     CASE
         WHEN hpm.physical_size_mb IS NULL THEN 'unknown'
         WHEN ROUND(hpm.physical_size_mb / 1024.0) >= 1024
@@ -79,4 +79,4 @@ SELECT
     END AS physical_host_size
 FROM host_traits ht
 LEFT JOIN host_physical_memory hpm
-    ON ht.hypervisor_id = hpm.resource_provider_uuid;
+    ON ht.hypervisor_id = hpm.resource_provider_uuid;