Add support for the BMS monitored resource to the Prometheus receiver

VERSION

@@ -1 +1 @@
-PKG_VERSION="2.44.0-bms"
+PKG_VERSION="2.44.0-bms"

cmd/google_cloud_ops_agent_diagnostics/main.go

@@ -44,7 +44,7 @@ func getUserAndMergedConfigs(ctx context.Context, userConfPath string) (*confgen
 func main() {
 	defer func() {
 		if r := recover(); r != nil {
-			log.Fatal("Recovered in run", r)
+			log.Fatalf("Recovering from a panic due to %v", r)
 		}
 	}()
 	if err := run(context.Background()); err != nil {

confgenerator/prometheus.go

@@ -26,14 +26,14 @@ import (
 
 	"github.com/GoogleCloudPlatform/ops-agent/confgenerator/otel"
 	"github.com/GoogleCloudPlatform/ops-agent/confgenerator/resourcedetector"
+	"github.com/GoogleCloudPlatform/ops-agent/internal/platform"
 	"github.com/go-playground/validator/v10"
 	yaml "github.com/goccy/go-yaml"
 	commonconfig "github.com/prometheus/common/config"
 	"github.com/prometheus/common/model"
 	promconfig "github.com/prometheus/prometheus/config"
 	"github.com/prometheus/prometheus/discovery"
 	_ "github.com/prometheus/prometheus/discovery/install" // init() of this package registers service discovery impl.
-	strutil "github.com/prometheus/prometheus/util/strutil"
 )
 
 var (
@@ -60,11 +60,13 @@ func (r PrometheusMetrics) Type() string {
 	return "prometheus"
 }
 
-func (r PrometheusMetrics) Pipelines(_ context.Context) []otel.ReceiverPipeline {
+func (r PrometheusMetrics) Pipelines(ctx context.Context) []otel.ReceiverPipeline {
 	// Get the resource metadata for the instance we're running on.
-	if gceMetadata, ok := MetadataResource.(resourcedetector.GCEResource); ok {
-		// Create a prometheus style mapping for the GCE metadata.
-		gceMetadataMap := createPrometheusStyleGCEMetadata(gceMetadata)
+	if MetadataResource != nil {
+		resourceMetadataMap := MetadataResource.PrometheusStyleMetadata()
+		if p := platform.FromContext(ctx).ResourceOverride; p != nil {
+			resourceMetadataMap = p.PrometheusStyleMetadata()
+		}
 
 		// Add the GCE metadata to the prometheus config.
 		for i := range r.PromConfig.ScrapeConfigs {
@@ -76,8 +78,8 @@ func (r PrometheusMetrics) Pipelines(_ context.Context) []otel.ReceiverPipeline
 					if labels == nil {
 						labels = model.LabelSet{}
 					}
-					for k, v := range gceMetadataMap {
-						// If there are conflicts, the GCE metadata should take precedence.
+					for k, v := range resourceMetadataMap {
+						// If there are conflicts, the resource metadata should take precedence.
 						labels[model.LabelName(k)] = model.LabelValue(v)
 					}
 
@@ -147,50 +149,6 @@ func deepCopy(config promconfig.Config) (promconfig.Config, error) {
 	return copyConfig, nil
 }
 
-func createPrometheusStyleGCEMetadata(gceMetadata resourcedetector.GCEResource) map[string]string {
-	metaLabels := map[string]string{
-		"__meta_gce_instance_id":   gceMetadata.InstanceID,
-		"__meta_gce_instance_name": gceMetadata.InstanceName,
-		"__meta_gce_project":       gceMetadata.Project,
-		"__meta_gce_zone":          gceMetadata.Zone,
-		"__meta_gce_network":       gceMetadata.Network,
-		// TODO(b/b/246995894): Add support for subnetwork label.
-		// "__meta_gce_subnetwork":    gceMetadata.Subnetwork,
-		"__meta_gce_public_ip":    gceMetadata.PublicIP,
-		"__meta_gce_private_ip":   gceMetadata.PrivateIP,
-		"__meta_gce_tags":         gceMetadata.Tags,
-		"__meta_gce_machine_type": gceMetadata.MachineType,
-	}
-	prefix := "__meta_gce_"
-	for k, v := range gceMetadata.Metadata {
-		sanitizedKey := "metadata_" + strutil.SanitizeLabelName(k)
-		metaLabels[prefix+sanitizedKey] = strings.ReplaceAll(v, "$", "$$")
-	}
-
-	// Labels are not available using the GCE metadata API.
-	// TODO(b/246995462): Add support for labels.
-	//
-	// for k, v := range gceMetadata.Label {
-	// 	metaLabels[prefix+"label_"+k] = v
-	// }
-
-	for k, v := range gceMetadata.InterfaceIPv4 {
-		sanitizedKey := "interface_ipv4_nic" + strutil.SanitizeLabelName(k)
-		metaLabels[prefix+sanitizedKey] = v
-	}
-
-	// Set the location, namespace and cluster labels.
-	metaLabels["location"] = gceMetadata.Zone
-	metaLabels["namespace"] = gceMetadata.InstanceID
-	metaLabels["cluster"] = "__gce__"
-
-	// Set some curated labels.
-	metaLabels["instance_name"] = gceMetadata.InstanceName
-	metaLabels["machine_type"] = gceMetadata.MachineType
-
-	return metaLabels
-}
-
 func validatePrometheusConfig(sl validator.StructLevel) {
 	promConfig := sl.Current().Interface().(promconfig.Config)
 

confgenerator/resourcedetector/bms_detector.go

@@ -62,6 +62,21 @@ func (r BMSResource) MonitoredResource() *monitoredres.MonitoredResource {
 	}
 }
 
+func (r BMSResource) PrometheusStyleMetadata() map[string]string {
+	metaLabels := map[string]string{
+		"__meta_bms_instance_id": r.InstanceID,
+		"__meta_bms_project":     r.Project,
+		"__meta_bms_location":    r.Location,
+	}
+	// Set the location, namespace and cluster labels.
+	metaLabels["location"] = r.Location
+	metaLabels["namespace"] = r.InstanceID
+	metaLabels["cluster"] = "__bms__"
+	// Set some curated labels.
+	metaLabels["instance_name"] = r.InstanceID
+	return metaLabels
+}
+
 func OnBMS() bool {
 	return os.Getenv(bmsProjectIDEnv) != "" && os.Getenv(bmsLocationEnv) != "" && os.Getenv(bmsInstanceIDEnv) != ""
 }

confgenerator/resourcedetector/detector.go

@@ -25,6 +25,7 @@ type Resource interface {
 	MonitoredResource() *monitoredres.MonitoredResource
 	OTelResourceAttributes() map[string]string
 	ProjectName() string
+	PrometheusStyleMetadata() map[string]string
 }
 
 // Get a resource instance for the current environment;
@@ -59,6 +60,10 @@ func (UnrecognizedPlatformResource) MonitoredResource() *monitoredres.MonitoredR
 	return nil
 }
 
+func (UnrecognizedPlatformResource) PrometheusStyleMetadata() map[string]string {
+	return nil
+}
+
 func GetUnrecognizedPlatformResource() (Resource, error) {
 	return UnrecognizedPlatformResource{}, nil
 }

confgenerator/resourcedetector/gce_detector.go

@@ -14,7 +14,12 @@
 
 package resourcedetector
 
-import "google.golang.org/genproto/googleapis/api/monitoredres"
+import (
+	"strings"
+
+	"github.com/prometheus/prometheus/util/strutil"
+	"google.golang.org/genproto/googleapis/api/monitoredres"
+)
 
 type gceAttribute int
 
@@ -128,6 +133,50 @@ func (r GCEResource) MonitoredResource() *monitoredres.MonitoredResource {
 	}
 }
 
+func (r GCEResource) PrometheusStyleMetadata() map[string]string {
+	metaLabels := map[string]string{
+		"__meta_gce_instance_id":   r.InstanceID,
+		"__meta_gce_instance_name": r.InstanceName,
+		"__meta_gce_project":       r.Project,
+		"__meta_gce_zone":          r.Zone,
+		"__meta_gce_network":       r.Network,
+		// TODO(b/b/246995894): Add support for subnetwork label.
+		// "__meta_gce_subnetwork":    r.Subnetwork,
+		"__meta_gce_public_ip":    r.PublicIP,
+		"__meta_gce_private_ip":   r.PrivateIP,
+		"__meta_gce_tags":         r.Tags,
+		"__meta_gce_machine_type": r.MachineType,
+	}
+	prefix := "__meta_gce_"
+	for k, v := range r.Metadata {
+		sanitizedKey := "metadata_" + strutil.SanitizeLabelName(k)
+		metaLabels[prefix+sanitizedKey] = strings.ReplaceAll(v, "$", "$$")
+	}
+
+	// Labels are not available using the GCE metadata API.
+	// TODO(b/246995462): Add support for labels.
+	//
+	// for k, v := range r.Label {
+	// 	metaLabels[prefix+"label_"+k] = v
+	// }
+
+	for k, v := range r.InterfaceIPv4 {
+		sanitizedKey := "interface_ipv4_nic" + strutil.SanitizeLabelName(k)
+		metaLabels[prefix+sanitizedKey] = v
+	}
+
+	// Set the location, namespace and cluster labels.
+	metaLabels["location"] = r.Zone
+	metaLabels["namespace"] = r.InstanceID
+	metaLabels["cluster"] = "__gce__"
+
+	// Set some curated labels.
+	metaLabels["instance_name"] = r.InstanceName
+	metaLabels["machine_type"] = r.MachineType
+
+	return metaLabels
+}
+
 type GCEResourceBuilderInterface interface {
 	GetResource() (Resource, error)
 }

confgenerator/testdata/goldens/metrics-receiver_prometheus/golden/linux-bms/otel.yaml

@@ -501,23 +501,13 @@ receivers:
         - targets:
           - localhost:1234
           labels:
-            __meta_gce_instance_id: test-instance-id
-            __meta_gce_instance_name: test-instance-name
-            __meta_gce_interface_ipv4_nictest_interface: test-interface-ipv4
-            __meta_gce_machine_type: test-machine-type
-            __meta_gce_metadata_test_escape: $${foo:bar}
-            __meta_gce_metadata_test_key: test-value
-            __meta_gce_network: test-network
-            __meta_gce_private_ip: test-private-ip
-            __meta_gce_project: test-project
-            __meta_gce_public_ip: test-public-ip
-            __meta_gce_tags: test-tag
-            __meta_gce_zone: test-zone
-            cluster: __gce__
-            instance_name: test-instance-name
-            location: test-zone
-            machine_type: test-machine-type
-            namespace: test-instance-id
+            __meta_bms_instance_id: test-bms-instance
+            __meta_bms_location: test-bms-location
+            __meta_bms_project: test-bms-project
+            cluster: __bms__
+            instance_name: test-bms-instance
+            location: test-bms-location
+            namespace: test-bms-instance
     preserve_untyped: true
 service:
   pipelines:

confgenerator/testdata/goldens/metrics-receiver_prometheus_complex/golden/linux-bms/otel.yaml

@@ -494,23 +494,13 @@ receivers:
         - targets:
           - localhost:1234
           labels:
-            __meta_gce_instance_id: test-instance-id
-            __meta_gce_instance_name: test-instance-name
-            __meta_gce_interface_ipv4_nictest_interface: test-interface-ipv4
-            __meta_gce_machine_type: test-machine-type
-            __meta_gce_metadata_test_escape: $${foo:bar}
-            __meta_gce_metadata_test_key: test-value
-            __meta_gce_network: test-network
-            __meta_gce_private_ip: test-private-ip
-            __meta_gce_project: test-project
-            __meta_gce_public_ip: test-public-ip
-            __meta_gce_tags: test-tag
-            __meta_gce_zone: test-zone
-            cluster: __gce__
-            instance_name: test-instance-name
-            location: test-zone
-            machine_type: test-machine-type
-            namespace: test-instance-id
+            __meta_bms_instance_id: test-bms-instance
+            __meta_bms_location: test-bms-location
+            __meta_bms_project: test-bms-project
+            cluster: __bms__
+            instance_name: test-bms-instance
+            location: test-bms-location
+            namespace: test-bms-instance
       - job_name: drop
         honor_timestamps: true
         scrape_interval: 10s
@@ -553,23 +543,13 @@ receivers:
         - targets:
           - 0.0.0.0:9100
           labels:
-            __meta_gce_instance_id: test-instance-id
-            __meta_gce_instance_name: test-instance-name
-            __meta_gce_interface_ipv4_nictest_interface: test-interface-ipv4
-            __meta_gce_machine_type: test-machine-type
-            __meta_gce_metadata_test_escape: $${foo:bar}
-            __meta_gce_metadata_test_key: test-value
-            __meta_gce_network: test-network
-            __meta_gce_private_ip: test-private-ip
-            __meta_gce_project: test-project
-            __meta_gce_public_ip: test-public-ip
-            __meta_gce_tags: test-tag
-            __meta_gce_zone: test-zone
-            cluster: __gce__
-            instance_name: test-instance-name
-            location: test-zone
-            machine_type: test-machine-type
-            namespace: test-instance-id
+            __meta_bms_instance_id: test-bms-instance
+            __meta_bms_location: test-bms-location
+            __meta_bms_project: test-bms-project
+            cluster: __bms__
+            instance_name: test-bms-instance
+            location: test-bms-location
+            namespace: test-bms-instance
     preserve_untyped: true
 service:
   pipelines:

confgenerator/testdata/goldens/metrics-receiver_prometheus_default_replacement/golden/linux-bms/otel.yaml

@@ -501,23 +501,13 @@ receivers:
         - targets:
           - localhost:1234
           labels:
-            __meta_gce_instance_id: test-instance-id
-            __meta_gce_instance_name: test-instance-name
-            __meta_gce_interface_ipv4_nictest_interface: test-interface-ipv4
-            __meta_gce_machine_type: test-machine-type
-            __meta_gce_metadata_test_escape: $${foo:bar}
-            __meta_gce_metadata_test_key: test-value
-            __meta_gce_network: test-network
-            __meta_gce_private_ip: test-private-ip
-            __meta_gce_project: test-project
-            __meta_gce_public_ip: test-public-ip
-            __meta_gce_tags: test-tag
-            __meta_gce_zone: test-zone
-            cluster: __gce__
-            instance_name: test-instance-name
-            location: test-zone
-            machine_type: test-machine-type
-            namespace: test-instance-id
+            __meta_bms_instance_id: test-bms-instance
+            __meta_bms_location: test-bms-location
+            __meta_bms_project: test-bms-project
+            cluster: __bms__
+            instance_name: test-bms-instance
+            location: test-bms-location
+            namespace: test-bms-instance
     preserve_untyped: true
 service:
   pipelines:

confgenerator/testdata/goldens/metrics-receiver_prometheus_multi_replacement_regex/golden/linux-bms/otel.yaml

@@ -501,23 +501,13 @@ receivers:
         - targets:
           - localhost:1234
           labels:
-            __meta_gce_instance_id: test-instance-id
-            __meta_gce_instance_name: test-instance-name
-            __meta_gce_interface_ipv4_nictest_interface: test-interface-ipv4
-            __meta_gce_machine_type: test-machine-type
-            __meta_gce_metadata_test_escape: $${foo:bar}
-            __meta_gce_metadata_test_key: test-value
-            __meta_gce_network: test-network
-            __meta_gce_private_ip: test-private-ip
-            __meta_gce_project: test-project
-            __meta_gce_public_ip: test-public-ip
-            __meta_gce_tags: test-tag
-            __meta_gce_zone: test-zone
-            cluster: __gce__
-            instance_name: test-instance-name
-            location: test-zone
-            machine_type: test-machine-type
-            namespace: test-instance-id
+            __meta_bms_instance_id: test-bms-instance
+            __meta_bms_location: test-bms-location
+            __meta_bms_project: test-bms-project
+            cluster: __bms__
+            instance_name: test-bms-instance
+            location: test-bms-location
+            namespace: test-bms-instance
     preserve_untyped: true
 service:
   pipelines: