feat: add ironic metrics

cmd/openstack-database-exporter/main.go

@@ -31,6 +31,10 @@ var (
 		"glance.database-url",
 		"Glance database connection URL (oslo.db format)",
 	).Envar("GLANCE_DATABASE_URL").String()
+	ironicDatabaseURL = kingpin.Flag(
+		"ironic.database-url",
+		"Ironic database connection URL (oslo.db format)",
+	).Envar("IRONIC_DATABASE_URL").String()
 	keystoneDatabaseURL = kingpin.Flag(
 		"keystone.database-url",
 		"Keystone database connection URL (oslo.db format)",
@@ -73,6 +77,7 @@ func main() {
 	reg := collector.NewRegistry(collector.Config{
 		CinderDatabaseURL:    *cinderDatabaseURL,
 		GlanceDatabaseURL:    *glanceDatabaseURL,
+		IronicDatabaseURL:    *ironicDatabaseURL,
 		KeystoneDatabaseURL:  *keystoneDatabaseURL,
 		MagnumDatabaseURL:    *magnumDatabaseURL,
 		ManilaDatabaseURL:    *manilaDatabaseURL,

internal/collector/collector.go

@@ -7,6 +7,7 @@ import (
 
 	"github.com/vexxhost/openstack_database_exporter/internal/collector/cinder"
 	"github.com/vexxhost/openstack_database_exporter/internal/collector/glance"
+	"github.com/vexxhost/openstack_database_exporter/internal/collector/ironic"
 	"github.com/vexxhost/openstack_database_exporter/internal/collector/keystone"
 	"github.com/vexxhost/openstack_database_exporter/internal/collector/magnum"
 	"github.com/vexxhost/openstack_database_exporter/internal/collector/manila"
@@ -22,6 +23,7 @@ const (
 type Config struct {
 	CinderDatabaseURL    string
 	GlanceDatabaseURL    string
+	IronicDatabaseURL    string
 	KeystoneDatabaseURL  string
 	MagnumDatabaseURL    string
 	ManilaDatabaseURL    string
@@ -35,6 +37,7 @@ func NewRegistry(cfg Config, logger *slog.Logger) *prometheus.Registry {
 
 	cinder.RegisterCollectors(reg, cfg.CinderDatabaseURL, logger)
 	glance.RegisterCollectors(reg, cfg.GlanceDatabaseURL, logger)
+	ironic.RegisterCollectors(reg, cfg.IronicDatabaseURL, logger)
 	keystone.RegisterCollectors(reg, cfg.KeystoneDatabaseURL, logger)
 	magnum.RegisterCollectors(reg, cfg.MagnumDatabaseURL, logger)
 	manila.RegisterCollectors(reg, cfg.ManilaDatabaseURL, logger)

internal/collector/ironic/baremetal.go

@@ -0,0 +1,71 @@
+package ironic
+
+import (
+	"context"
+	"database/sql"
+	"log/slog"
+
+	"github.com/prometheus/client_golang/prometheus"
+
+	ironicdb "github.com/vexxhost/openstack_database_exporter/internal/db/ironic"
+)
+
+// BaremetalCollector is the umbrella collector for Ironic baremetal metrics
+type BaremetalCollector struct {
+	queries *ironicdb.Queries
+	logger  *slog.Logger
+
+	// Single up metric for baremetal service
+	upMetric *prometheus.Desc
+
+	// Sub-collectors
+	nodesCollector *NodesCollector
+}
+
+// NewBaremetalCollector creates a new umbrella collector for Ironic baremetal service
+func NewBaremetalCollector(db *sql.DB, logger *slog.Logger) *BaremetalCollector {
+	return &BaremetalCollector{
+		queries: ironicdb.New(db),
+		logger: logger.With(
+			"namespace", Namespace,
+			"subsystem", Subsystem,
+			"collector", "baremetal",
+		),
+
+		upMetric: prometheus.NewDesc(
+			prometheus.BuildFQName(Namespace, Subsystem, "up"),
+			"Whether the Ironic baremetal service is up",
+			nil,
+			nil,
+		),
+
+		// Initialize sub-collectors
+		nodesCollector: NewNodesCollector(db, logger),
+	}
+}
+
+// Describe implements prometheus.Collector
+func (c *BaremetalCollector) Describe(ch chan<- *prometheus.Desc) {
+	ch <- c.upMetric
+	c.nodesCollector.Describe(ch)
+}
+
+// Collect implements prometheus.Collector
+func (c *BaremetalCollector) Collect(ch chan<- prometheus.Metric) {
+	ctx := context.Background()
+
+	// Query node metrics once and reuse the result
+	nodes, err := c.queries.GetNodeMetrics(ctx)
+	if err != nil {
+		c.logger.Error("failed to query Ironic database", "error", err)
+		ch <- prometheus.MustNewConstMetric(c.upMetric, prometheus.GaugeValue, 0)
+		return
+	}
+
+	// Collect nodes metrics using the already-fetched data
+	if err := c.nodesCollector.CollectFromRows(ch, nodes); err != nil {
+		c.logger.Error("nodes collector failed", "error", err)
+	}
+
+	ch <- prometheus.MustNewConstMetric(c.upMetric, prometheus.GaugeValue, 1)
+}

internal/collector/ironic/baremetal_test.go

@@ -0,0 +1,71 @@
+package ironic
+
+import (
+	"database/sql"
+	"regexp"
+	"testing"
+
+	"github.com/DATA-DOG/go-sqlmock"
+	ironicdb "github.com/vexxhost/openstack_database_exporter/internal/db/ironic"
+	"github.com/vexxhost/openstack_database_exporter/internal/testutil"
+)
+
+func TestBaremetalCollector(t *testing.T) {
+	tests := []testutil.CollectorTestCase{
+		{
+			Name: "successful collection",
+			SetupMock: func(mock sqlmock.Sqlmock) {
+				rows := sqlmock.NewRows([]string{
+					"uuid", "name", "power_state", "provision_state", "maintenance",
+					"resource_class", "console_enabled", "retired", "retired_reason",
+				}).AddRow(
+					"550e8400-e29b-41d4-a716-446655440000", "node-1", "power on", "active", false,
+					"baremetal", true, false, "",
+				)
+				mock.ExpectQuery(regexp.QuoteMeta(ironicdb.GetNodeMetrics)).WillReturnRows(rows)
+			},
+			ExpectedMetrics: `# HELP openstack_ironic_node Ironic node status
+# TYPE openstack_ironic_node gauge
+openstack_ironic_node{console_enabled="true",id="550e8400-e29b-41d4-a716-446655440000",maintenance="false",name="node-1",power_state="power on",provision_state="active",resource_class="baremetal",retired="false",retired_reason=""} 1
+# HELP openstack_ironic_up Whether the Ironic baremetal service is up
+# TYPE openstack_ironic_up gauge
+openstack_ironic_up 1
+`,
+		},
+		{
+			Name: "database connection failure",
+			SetupMock: func(mock sqlmock.Sqlmock) {
+				mock.ExpectQuery(regexp.QuoteMeta(ironicdb.GetNodeMetrics)).WillReturnError(sql.ErrConnDone)
+			},
+			ExpectedMetrics: `# HELP openstack_ironic_up Whether the Ironic baremetal service is up
+# TYPE openstack_ironic_up gauge
+openstack_ironic_up 0
+`,
+		},
+		{
+			Name: "skips nodes with null uuid",
+			SetupMock: func(mock sqlmock.Sqlmock) {
+				rows := sqlmock.NewRows([]string{
+					"uuid", "name", "power_state", "provision_state", "maintenance",
+					"resource_class", "console_enabled", "retired", "retired_reason",
+				}).AddRow(
+					nil, "node-no-uuid", "power on", "active", false,
+					"baremetal", true, false, "",
+				).AddRow(
+					"550e8400-e29b-41d4-a716-446655440000", "node-1", "power on", "active", false,
+					"baremetal", true, false, "",
+				)
+				mock.ExpectQuery(regexp.QuoteMeta(ironicdb.GetNodeMetrics)).WillReturnRows(rows)
+			},
+			ExpectedMetrics: `# HELP openstack_ironic_node Ironic node status
+# TYPE openstack_ironic_node gauge
+openstack_ironic_node{console_enabled="true",id="550e8400-e29b-41d4-a716-446655440000",maintenance="false",name="node-1",power_state="power on",provision_state="active",resource_class="baremetal",retired="false",retired_reason=""} 1
+# HELP openstack_ironic_up Whether the Ironic baremetal service is up
+# TYPE openstack_ironic_up gauge
+openstack_ironic_up 1
+`,
+		},
+	}
+
+	testutil.RunCollectorTests(t, tests, NewBaremetalCollector)
+}

internal/collector/ironic/ironic.go

@@ -0,0 +1,32 @@
+package ironic
+
+import (
+	"log/slog"
+
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/vexxhost/openstack_database_exporter/internal/db"
+)
+
+// Namespace and subsystem constants
+const (
+	Namespace = "openstack"
+	Subsystem = "ironic"
+)
+
+// RegisterCollectors registers all Ironic collectors with the given database and logger
+func RegisterCollectors(registry *prometheus.Registry, databaseURL string, logger *slog.Logger) {
+	if databaseURL == "" {
+		logger.Info("Collector not loaded", "service", "ironic", "reason", "database URL not configured")
+		return
+	}
+
+	conn, err := db.Connect(databaseURL)
+	if err != nil {
+		logger.Error("Failed to connect to database", "service", "ironic", "error", err)
+		return
+	}
+
+	registry.MustRegister(NewBaremetalCollector(conn, logger))
+
+	logger.Info("Registered collectors", "service", "ironic")
+}

internal/collector/ironic/nodes.go

@@ -0,0 +1,140 @@
+package ironic
+
+import (
+	"context"
+	"database/sql"
+	"log/slog"
+
+	"github.com/prometheus/client_golang/prometheus"
+	ironicdb "github.com/vexxhost/openstack_database_exporter/internal/db/ironic"
+)
+
+// maxLabelLength is the maximum length for free-form label values to prevent
+// unbounded cardinality and memory usage in Prometheus.
+const maxLabelLength = 128
+
+// truncateLabel truncates a string to maxLabelLength.
+func truncateLabel(s string) string {
+	if len(s) > maxLabelLength {
+		return s[:maxLabelLength]
+	}
+	return s
+}
+
+// NodesCollector collects metrics about Ironic nodes
+type NodesCollector struct {
+	queries *ironicdb.Queries
+	logger  *slog.Logger
+
+	// Individual node metrics
+	nodeMetric *prometheus.Desc
+}
+
+// NewNodesCollector creates a new NodesCollector
+func NewNodesCollector(db *sql.DB, logger *slog.Logger) *NodesCollector {
+	return &NodesCollector{
+		queries: ironicdb.New(db),
+		logger: logger.With(
+			"namespace", Namespace,
+			"subsystem", Subsystem,
+			"collector", "nodes",
+		),
+
+		nodeMetric: prometheus.NewDesc(
+			prometheus.BuildFQName(Namespace, Subsystem, "node"),
+			"Ironic node status",
+			[]string{
+				"id", "name", "power_state", "provision_state",
+				"resource_class", "maintenance", "console_enabled", "retired", "retired_reason",
+			},
+			nil,
+		),
+	}
+}
+
+// Describe implements prometheus.Collector
+func (c *NodesCollector) Describe(ch chan<- *prometheus.Desc) {
+	ch <- c.nodeMetric
+}
+
+// Collect queries the database and collects node metrics into the provided channel.
+func (c *NodesCollector) Collect(ch chan<- prometheus.Metric) error {
+	ctx := context.Background()
+
+	nodes, err := c.queries.GetNodeMetrics(ctx)
+	if err != nil {
+		c.logger.Error("failed to get node metrics", "error", err)
+		return err
+	}
+
+	return c.CollectFromRows(ch, nodes)
+}
+
+// CollectFromRows emits node metrics from pre-fetched rows.
+func (c *NodesCollector) CollectFromRows(ch chan<- prometheus.Metric, nodes []ironicdb.GetNodeMetricsRow) error {
+	for _, node := range nodes {
+		// Skip nodes with empty UUID to avoid duplicate label sets
+		if !node.Uuid.Valid || node.Uuid.String == "" {
+			nodeName := ""
+			if node.Name.Valid {
+				nodeName = node.Name.String
+			}
+			c.logger.Debug("skipping node with empty UUID", "name", nodeName)
+			continue
+		}
+		// Individual node status metric
+		maintenance := "false"
+		if node.Maintenance.Valid && node.Maintenance.Bool {
+			maintenance = "true"
+		}
+
+		powerState := "unknown"
+		if node.PowerState.Valid {
+			powerState = node.PowerState.String
+		}
+
+		provisionState := "unknown"
+		if node.ProvisionState.Valid {
+			provisionState = node.ProvisionState.String
+		}
+
+		resourceClass := "unknown"
+		if node.ResourceClass.Valid {
+			resourceClass = node.ResourceClass.String
+		}
+
+		consoleEnabled := "false"
+		if node.ConsoleEnabled.Valid && node.ConsoleEnabled.Bool {
+			consoleEnabled = "true"
+		}
+
+		retired := "false"
+		if node.Retired.Valid && node.Retired.Bool {
+			retired = "true"
+		}
+
+		retiredReason := truncateLabel(node.RetiredReason)
+
+		name := ""
+		if node.Name.Valid {
+			name = node.Name.String
+		}
+
+		nodeUUID := node.Uuid.String
+
+		// Emit individual node metric
+		metric, err := prometheus.NewConstMetric(
+			c.nodeMetric,
+			prometheus.GaugeValue,
+			1,
+			nodeUUID, name, powerState, provisionState, resourceClass, maintenance, consoleEnabled, retired, retiredReason,
+		)
+		if err != nil {
+			c.logger.Error("failed to create node metric", "error", err)
+			continue
+		}
+		ch <- metric
+	}
+
+	return nil
+}