Merge #124409

124409: roachprod: add custom tags for prometheus config r=srosenberg a=nameisbhaskar

few of the labels are missing in prometheus with the change from "gce_sd_configs" to "file_sd_configs". This is because these were extracted from the VM metadata which is not available now. So, the same values are set from roachprod now.

Informs: #124319
Epic: none

Co-authored-by: Bhaskarjyoti Bora <bhaskar.bora@cockroachlabs.com>

pkg/roachprod/promhelperclient/client.go

@@ -77,7 +77,7 @@ func (c *PromClient) UpdatePrometheusTargets(
 	ctx context.Context,
 	promUrl, clusterName string,
 	forceFetchCreds bool,
-	nodes map[int]string,
+	nodes map[int]*NodeInfo,
 	insecure bool,
 	l *logger.Logger,
 ) error {
@@ -157,17 +157,29 @@ type CCParams struct {
 	Labels  map[string]string `yaml:"labels"`
 }
 
+// NodeInfo contains the target and labels for the node
+type NodeInfo struct {
+	Target       string            // Name of the node
+	CustomLabels map[string]string // Custom labels to be added to the cluster config
+}
+
 // createClusterConfigFile creates the cluster config file per node
-func buildCreateRequest(nodes map[int]string, insecure bool) (io.Reader, error) {
+func buildCreateRequest(nodes map[int]*NodeInfo, insecure bool) (io.Reader, error) {
 	configs := make([]*CCParams, 0)
 	for i, n := range nodes {
 		params := &CCParams{
-			Targets: []string{n},
+			Targets: []string{n.Target},
 			Labels: map[string]string{
+				// default labels
 				"node":   strconv.Itoa(i),
 				"tenant": install.SystemInterfaceName,
+				"job":    "cockroachdb",
 			},
 		}
+		// custom labels - this can override the default labels if needed
+		for n, v := range n.CustomLabels {
+			params.Labels[n] = v
+		}
 		configs = append(configs, params)
 	}
 	cb, err := yaml.Marshal(&configs)

pkg/roachprod/promhelperclient/client_test.go

@@ -49,12 +49,16 @@ func TestUpdatePrometheusTargets(t *testing.T) {
 				Body:       io.NopCloser(strings.NewReader("failed")),
 			}, nil
 		}
-		err := c.UpdatePrometheusTargets(ctx, promUrl, "c1", false, map[int]string{1: "n1"}, true, l)
+		err := c.UpdatePrometheusTargets(ctx, promUrl, "c1", false,
+			map[int]*NodeInfo{1: {Target: "n1"}}, true, l)
 		require.NotNil(t, err)
 		require.Equal(t, "request failed with status 400 and error failed", err.Error())
 	})
 	t.Run("UpdatePrometheusTargets succeeds", func(t *testing.T) {
-		nodeInfos := map[int]string{1: "n1", 3: "n3"}
+		nodeInfos := map[int]*NodeInfo{1: {Target: "n1"}, 3: {
+			Target:       "n3",
+			CustomLabels: map[string]string{"custom": "label"},
+		}}
 		c.httpPut = func(ctx context.Context, url string, h *http.Header, body io.Reader) (
 			resp *http.Response, err error) {
 			require.Equal(t, getUrl(promUrl, "c1"), url)
@@ -67,8 +71,12 @@ func TestUpdatePrometheusTargets(t *testing.T) {
 			for _, c := range configs {
 				nodeID, err := strconv.Atoi(c.Labels["node"])
 				require.NoError(t, err)
-				require.Equal(t, nodeInfos[nodeID], c.Targets[0])
+				require.Equal(t, nodeInfos[nodeID].Target, c.Targets[0])
 				require.Equal(t, "system", c.Labels["tenant"])
+				require.Equal(t, "cockroachdb", c.Labels["job"])
+				for k, v := range nodeInfos[nodeID].CustomLabels {
+					require.Equal(t, v, c.Labels[k])
+				}
 			}
 			return &http.Response{
 				StatusCode: 200,

pkg/roachprod/roachprod.go

@@ -794,7 +794,7 @@ func UpdateTargets(
 
 // updatePrometheusTargets updates the prometheus instance cluster config. Any error is logged and ignored.
 func updatePrometheusTargets(ctx context.Context, l *logger.Logger, c *install.SyncedCluster) {
-	nodeIPPorts := make(map[int]string)
+	nodeIPPorts := make(map[int]*promhelperclient.NodeInfo)
 	nodeIPPortsMutex := syncutil.RWMutex{}
 	var wg sync.WaitGroup
 	for _, node := range c.Nodes {
@@ -808,10 +808,10 @@ func updatePrometheusTargets(ctx context.Context, l *logger.Logger, c *install.S
 					l.Errorf("error getting the port for node %d: %v", index, err)
 					return
 				}
-				nodeInfo := fmt.Sprintf("%s:%d", v.PublicIP, desc.Port)
+				nodeInfo := fmt.Sprintf("%s:%d", v.PrivateIP, desc.Port)
 				nodeIPPortsMutex.Lock()
 				// ensure atomicity in map update
-				nodeIPPorts[index] = nodeInfo
+				nodeIPPorts[index] = &promhelperclient.NodeInfo{Target: nodeInfo, CustomLabels: getLabels(v)}
 				nodeIPPortsMutex.Unlock()
 			}(int(node), c.VMs[node-1])
 		}
@@ -826,6 +826,32 @@ func updatePrometheusTargets(ctx context.Context, l *logger.Logger, c *install.S
 	}
 }
 
+// regionRegEx is the regex to extract the region label from zone available as vm property
+var regionRegEx = regexp.MustCompile("(^.+[0-9]+)(-[a-f]$)")
+
+// getLabels returns the labels to be populated in the target configuration in prometheus
+func getLabels(v vm.VM) map[string]string {
+	labels := map[string]string{
+		"cluster":  v.Labels["cluster"],
+		"instance": v.Name,
+		"host_ip":  v.PrivateIP,
+		"project":  v.Project,
+		"zone":     v.Zone,
+	}
+	match := regionRegEx.FindStringSubmatch(v.Zone)
+	if len(match) > 1 {
+		labels["region"] = match[1]
+	}
+	// the following labels are present if the test labels are added before the VM is started
+	if t, ok := v.Labels["test_name"]; ok {
+		labels["test_name"] = t
+	}
+	if t, ok := v.Labels["test_run_id"]; ok {
+		labels["test_run_id"] = t
+	}
+	return labels
+}
+
 // Monitor monitors the status of cockroach nodes in a cluster.
 func Monitor(
 	ctx context.Context, l *logger.Logger, clusterName string, opts install.MonitorOpts,