Merge #149369

149369: roachprod: scrape workload metrics via promhelperservice r=DarrylWong,herkolategan a=golgeek

This patch adds the workload metrics ports range (2112-2120) to the scrape targets handled by promhelperservice to allow scraping of workload metrics in all Cloud providers.

For releases prior to 25.3 (this patch will be backported to 25.3), workload metrics will only be scraped in GCE by the gce_sd_config Prometheus discovery mechanism.

Epic: none
Release note: None

Co-authored-by: Ludovic Leroux <[email protected]>

Author: craig[bot]

pkg/roachprod/roachprod.go

@@ -836,33 +836,51 @@ func updatePrometheusTargets(
 				l.Errorf("error getting the port for node %d: %v", nodeID, err)
 				return
 			}
+
 			nodeIP := v.PrivateIP
 			if reachability == promhelperclient.Public {
 				nodeIP = v.PublicIP
 			}
-			nodeInfo := fmt.Sprintf("%s:%d", nodeIP, desc.Port)
-			nodeIPPortsMutex.Lock()
+
 			// ensure atomicity in map update
+			nodeIPPortsMutex.Lock()
+			defer nodeIPPortsMutex.Unlock()
+
+			// Hardware metrics
 			if _, ok := nodeIPPorts[nodeID]; !ok {
 				nodeIPPorts[nodeID] = []*promhelperclient.NodeInfo{
 					{
 						Target:       fmt.Sprintf("%s:%d", nodeIP, vm.NodeExporterPort),
-						CustomLabels: createLabels(nodeID, v, "node_exporter", !c.Secure),
+						CustomLabels: createLabels(nodeID, v, "node_exporter", false),
 					},
 					{
 						Target:       fmt.Sprintf("%s:%d", nodeIP, vm.EbpfExporterPort),
-						CustomLabels: createLabels(nodeID, v, "ebpf_exporter", !c.Secure),
+						CustomLabels: createLabels(nodeID, v, "ebpf_exporter", false),
 					},
 				}
 			}
+
+			// CRDB metrics
 			nodeIPPorts[nodeID] = append(
 				nodeIPPorts[nodeID],
 				&promhelperclient.NodeInfo{
-					Target:       nodeInfo,
+					Target:       fmt.Sprintf("%s:%d", nodeIP, desc.Port),
 					CustomLabels: createLabels(nodeID, v, "cockroachdb", !c.Secure),
 				},
 			)
-			nodeIPPortsMutex.Unlock()
+
+			// Workload metrics
+			// TODO (golgeek): find a way to figure out if workload will be running
+			// on the node and only scrape its actual metrics port?
+			for port := vm.WorkloadMetricsPortMin; port <= vm.WorkloadMetricsPortMax; port++ {
+				nodeIPPorts[nodeID] = append(
+					nodeIPPorts[nodeID],
+					&promhelperclient.NodeInfo{
+						Target:       fmt.Sprintf("%s:%d", nodeIP, port),
+						CustomLabels: createLabels(nodeID, v, "workload", false),
+					},
+				)
+			}
 		}(int(node), c.VMs[node-1])
 
 	}
@@ -902,26 +920,26 @@ func createLabels(nodeID int, v vm.VM, job string, insecure bool) map[string]str
 	if t, ok := v.Labels["test_run_id"]; ok {
 		labels["test_run_id"] = t
 	}
+
+	// Default configuration for promhelperservice is HTTPS but insecure exporters
+	// are scraped over HTTP.
+	if insecure {
+		labels["__scheme__"] = "http"
+	}
+
 	switch job {
 	case "cockroachdb":
 		labels["__metrics_path__"] = "/_status/vars"
-		if insecure {
-			labels["__scheme__"] = "http"
-		}
 	case "node_exporter":
 		labels["__metrics_path__"] = vm.NodeExporterMetricsPath
-		// node_exporter is always scraped over http
-		labels["__scheme__"] = "http"
-
-		// Node ID is exposed by cockroachdb metrics, we add it to node_exporter
-		labels["node_id"] = strconv.Itoa(nodeID)
-
 	case "ebpf_exporter":
 		labels["__metrics_path__"] = vm.EbpfExporterMetricsPath
-		// ebpf_exporter is always scraped over http
-		labels["__scheme__"] = "http"
+	case "workload":
+		labels["__metrics_path__"] = vm.WorkloadMetricsPath
+	}
 
-		// Node ID is exposed by cockroachdb metrics, we add it to ebpf_exporter
+	// Node ID is exposed by cockroachdb metrics, let's add it to other exporters.
+	if job != "cockroachdb" {
 		labels["node_id"] = strconv.Itoa(nodeID)
 	}
 

pkg/roachprod/vm/startup.go

@@ -35,6 +35,13 @@ const (
 	// EbpfExporterMetricsPath is the path that EbpfExporter serves metrics on.
 	EbpfExporterMetricsPath = "/metrics"
 
+	// WorkloadMetricsPortMin and WorkloadMetricsPortMax are the range of ports
+	// used by workload to expose metrics.
+	WorkloadMetricsPortMin = 2112
+	WorkloadMetricsPortMax = 2120
+	// WorkloadMetricsPath is the path that workload serves metrics on.
+	WorkloadMetricsPath = "/metrics"
+
 	// DefaultSharedUser is the default user that is shared across all VMs.
 	DefaultSharedUser = "ubuntu"
 	// InitializedFile is the base name of the initialization paths defined below.