Metric for multiple kymas per subaccount (#1598)

internal/metricsv2/instances_collector.go

@@ -19,6 +19,7 @@ import (
 // - kcp_keb_instances_total - total number of all instances
 // - kcp_keb_global_account_id_instances_total - total number of all instances per global account
 // - kcp_keb_ers_context_license_type_total - count of instances grouped by license types
+// - kcp_keb_sub_account_id_instances_total - total number of instances per subaccount
 type InstancesStatsGetter interface {
 	GetActiveInstanceStats() (internal.InstanceStats, error)
 	GetERSContextStats() (internal.ERSContextStats, error)
@@ -29,6 +30,7 @@ type InstancesCollector struct {
 
 	instancesDesc        *prometheus.Desc
 	instancesPerGAIDDesc *prometheus.Desc
+	instancesPerSAIDDesc *prometheus.Desc
 	licenseTypeDesc      *prometheus.Desc
 	logger               *slog.Logger
 }
@@ -47,6 +49,11 @@ func NewInstancesCollector(statsGetter InstancesStatsGetter, logger *slog.Logger
 			"The total number of instances by Global Account ID",
 			[]string{"global_account_id"},
 			nil),
+		instancesPerSAIDDesc: prometheus.NewDesc(
+			prometheus.BuildFQName(prometheusNamespacev2, prometheusSubsystemv2, "sub_account_id_instances_total"),
+			"The total number of instances by SubAccount ID",
+			[]string{"sub_account_id"},
+			nil),
 		licenseTypeDesc: prometheus.NewDesc(
 			prometheus.BuildFQName(prometheusNamespacev2, prometheusSubsystemv2, "ers_context_license_type_total"),
 			"count of instances grouped by license types",
@@ -58,6 +65,7 @@ func NewInstancesCollector(statsGetter InstancesStatsGetter, logger *slog.Logger
 func (c *InstancesCollector) Describe(ch chan<- *prometheus.Desc) {
 	ch <- c.instancesDesc
 	ch <- c.instancesPerGAIDDesc
+	ch <- c.instancesPerSAIDDesc
 	ch <- c.licenseTypeDesc
 }
 
@@ -72,6 +80,10 @@ func (c *InstancesCollector) Collect(ch chan<- prometheus.Metric) {
 		for globalAccountID, num := range stats.PerGlobalAccountID {
 			collect(ch, c.instancesPerGAIDDesc, num, globalAccountID)
 		}
+
+		for subAccountID, num := range stats.PerSubAcocuntID {
+			collect(ch, c.instancesPerSAIDDesc, num, subAccountID)
+		}
 	}
 
 	stats2, err := c.statsGetter.GetERSContextStats()

internal/model.go

@@ -367,6 +367,7 @@ type OperationStatsV2 struct {
 type InstanceStats struct {
 	TotalNumberOfInstances int
 	PerGlobalAccountID     map[string]int
+	PerSubAcocuntID        map[string]int
 }
 
 // ERSContextStats provides aggregated information regarding ERSContext

internal/storage/dbmodel/operation.go

@@ -53,6 +53,11 @@ type InstanceByGlobalAccountIDStatEntry struct {
 	Total           int
 }
 
+type InstanceBySubAccountIDStatEntry struct {
+	SubAccountID string
+	Total        int
+}
+
 type InstanceERSContextStatsEntry struct {
 	LicenseType sql.NullString
 	Total       int

internal/storage/driver/postsql/instance.go

@@ -545,11 +545,20 @@ func (s *Instance) GetActiveInstanceStats() (internal.InstanceStats, error) {
 
 	result := internal.InstanceStats{
 		PerGlobalAccountID: make(map[string]int),
+		PerSubAcocuntID:    make(map[string]int),
 	}
 	for _, e := range entries {
 		result.PerGlobalAccountID[e.GlobalAccountID] = e.Total
 		result.TotalNumberOfInstances = result.TotalNumberOfInstances + e.Total
 	}
+
+	subEntries, err := s.NewReadSession().GetSubaccountsInstanceStats()
+	if err != nil {
+		return internal.InstanceStats{}, err
+	}
+	for _, e := range subEntries {
+		result.PerSubAcocuntID[e.SubAccountID] = e.Total
+	}
 	return result, nil
 }
 

internal/storage/driver/postsql/instance_test.go

@@ -112,10 +112,10 @@ func TestInstance(t *testing.T) {
 
 		// populate database with samples
 		fixInstances := []internal.Instance{
-			*fixInstance(instanceData{val: "A1", globalAccountID: "A"}),
-			*fixInstance(instanceData{val: "A2", globalAccountID: "A", deletedAt: time.Time{}}),
-			*fixInstance(instanceData{val: "A3", globalAccountID: "A"}),
-			*fixInstance(instanceData{val: "C1", globalAccountID: "C"}),
+			*fixInstance(instanceData{val: "A1", globalAccountID: "A", subAccountID: "sub-01"}),
+			*fixInstance(instanceData{val: "A2", globalAccountID: "A", subAccountID: "sub-02", deletedAt: time.Time{}}),
+			*fixInstance(instanceData{val: "A3", globalAccountID: "A", subAccountID: "sub-02"}),
+			*fixInstance(instanceData{val: "C1", globalAccountID: "C", subAccountID: "sub-01"}),
 			*fixInstance(instanceData{val: "C2", globalAccountID: "C", deletedAt: time.Now()}),
 			*fixInstance(instanceData{val: "B1", globalAccountID: "B", deletedAt: time.Now()}),
 		}
@@ -155,6 +155,7 @@ func TestInstance(t *testing.T) {
 		assert.Equal(t, internal.InstanceStats{
 			TotalNumberOfInstances: 3,
 			PerGlobalAccountID:     map[string]int{"A": 2, "C": 1},
+			PerSubAcocuntID:        map[string]int{"sub-01": 2},
 		}, stats)
 		assert.Equal(t, 3, numberOfInstancesA)
 		assert.Equal(t, 1, numberOfInstancesC)

internal/storage/postsql/factory.go

@@ -37,6 +37,7 @@ type ReadSession interface {
 	ListOperationsByType(operationType internal.OperationType) ([]dbmodel.OperationDTO, dberr.Error)
 	GetOperationStats() ([]dbmodel.OperationStatEntry, error)
 	GetActiveInstanceStats() ([]dbmodel.InstanceByGlobalAccountIDStatEntry, error)
+	GetSubaccountsInstanceStats() ([]dbmodel.InstanceBySubAccountIDStatEntry, error)
 	GetERSContextStats() ([]dbmodel.InstanceERSContextStatsEntry, error)
 	GetNumberOfInstancesForGlobalAccountID(globalAccountID string) (int, error)
 	GetRuntimeStateByOperationID(operationID string) (dbmodel.RuntimeStateDTO, dberr.Error)

internal/storage/postsql/read.go

@@ -697,6 +697,30 @@ func (r readSession) GetActiveInstanceStats() ([]dbmodel.InstanceByGlobalAccount
 	return rows, err
 }
 
+func (r readSession) GetSubaccountsInstanceStats() ([]dbmodel.InstanceBySubAccountIDStatEntry, error) {
+	var rows []dbmodel.InstanceBySubAccountIDStatEntry
+	var stmt *dbr.SelectStmt
+	filter := dbmodel.InstanceFilter{
+		States: []dbmodel.InstanceState{dbmodel.InstanceNotDeprovisioned},
+	}
+	// Find and join the last operation for each instance matching the state filter(s).
+	// Last operation is found with the greatest-n-per-group problem solved with OUTER JOIN, followed by a (INNER) JOIN to get instance columns.
+	stmt = r.session.
+		Select(fmt.Sprintf("%s.sub_account_id", InstancesTableName), "count(*) as total").
+		From(InstancesTableName).
+		Join(dbr.I(OperationTableName).As("o1"), fmt.Sprintf("%s.instance_id = o1.instance_id", InstancesTableName)).
+		LeftJoin(dbr.I(OperationTableName).As("o2"), fmt.Sprintf("%s.instance_id = o2.instance_id AND o1.created_at < o2.created_at AND o2.state NOT IN ('%s', '%s')", InstancesTableName, orchestration.Pending, orchestration.Canceled)).
+		Where("o2.created_at IS NULL").
+		Where("deleted_at = '0001-01-01T00:00:00.000Z'").
+		Where(fmt.Sprintf("o1.state NOT IN ('%s', '%s')", orchestration.Pending, orchestration.Canceled)).
+		Where(buildInstanceStateFilters("o1", filter)).
+		GroupBy(fmt.Sprintf("%s.sub_account_id", InstancesTableName)).
+		Having(fmt.Sprintf("count(*) > 1"))
+
+	_, err := stmt.Load(&rows)
+	return rows, err
+}
+
 func (r readSession) GetERSContextStats() ([]dbmodel.InstanceERSContextStatsEntry, error) {
 	var rows []dbmodel.InstanceERSContextStatsEntry
 	// group existing instances by license_Type from the last operation that is not pending or canceled