[tmpnet] Move monitoring label handling to node

Author: maru-ava

tests/fixture/tmpnet/check_monitoring.go

@@ -13,6 +13,7 @@ import (
 	"math"
 	"net/http"
 	"net/url"
+	"os"
 	"strconv"
 	"strings"
 	"time"
@@ -271,13 +272,11 @@ func getSelectors(networkUUID string) (string, error) {
 
 	// Fall back to using Github labels as selectors
 	selectors := []string{}
-	githubLabels := githubLabelsFromEnv()
-	for label := range githubLabels {
-		value := githubLabels[label]
-		if len(value) == 0 {
-			continue
+	for _, label := range githubLabels {
+		value := os.Getenv(strings.ToUpper(label))
+		if len(value) > 0 {
+			selectors = append(selectors, fmt.Sprintf(`%s="%s"`, label, value))
 		}
-		selectors = append(selectors, fmt.Sprintf(`%s="%s"`, label, value))
 	}
 	if len(selectors) == 0 {
 		return "", errors.New("no GH_* env vars set to use for selectors")

tests/fixture/tmpnet/node.go

@@ -13,6 +13,7 @@ import (
 	"net"
 	"net/http"
 	"net/netip"
+	"os"
 	"strconv"
 	"strings"
 	"time"
@@ -37,6 +38,16 @@ var (
 	errMissingTLSKeyForNodeID = fmt.Errorf("failed to ensure node ID: missing value for %q", config.StakingTLSKeyContentKey)
 	errMissingCertForNodeID   = fmt.Errorf("failed to ensure node ID: missing value for %q", config.StakingCertContentKey)
 	errInvalidKeypair         = fmt.Errorf("%q and %q must be provided together or not at all", config.StakingTLSKeyContentKey, config.StakingCertContentKey)
+
+	// Labels expected to be available in the environment when running in GitHub Actions
+	githubLabels = []string{
+		"gh_repo",
+		"gh_workflow",
+		"gh_run_id",
+		"gh_run_number",
+		"gh_run_attempt",
+		"gh_job_id",
+	}
 )
 
 // NodeRuntime defines the methods required to support running a node.
@@ -418,3 +429,26 @@ func (n *Node) WaitForHealthy(ctx context.Context) error {
 		}
 	}
 }
+
+// getMonitoringLabels retrieves the map of labels and their values to be
+// applied to metrics and logs collected from the node.
+func (n *Node) getMonitoringLabels() map[string]string {
+	labels := map[string]string{
+		// Explicitly setting an instance label avoids the default
+		// behavior of using the node's URI since the URI isn't
+		// guaranteed stable (e.g. port may change after restart).
+		"instance":          n.GetUniqueID(),
+		"network_uuid":      n.network.UUID,
+		"node_id":           n.NodeID.String(),
+		"is_ephemeral_node": strconv.FormatBool(n.IsEphemeral),
+		"network_owner":     n.network.Owner,
+	}
+	// Include the values of github labels if available
+	for _, label := range githubLabels {
+		value := os.Getenv(strings.ToUpper(label))
+		if len(value) > 0 {
+			labels[label] = value
+		}
+	}
+	return labels
+}

tests/fixture/tmpnet/process_runtime.go

@@ -16,7 +16,6 @@ import (
 	"os"
 	"os/exec"
 	"path/filepath"
-	"strconv"
 	"strings"
 	"syscall"
 	"time"
@@ -322,17 +321,7 @@ func getProcess(pid int) (*os.Process, error) {
 // Write monitoring configuration enabling collection of metrics and logs from the node.
 func (p *ProcessRuntime) writeMonitoringConfig() error {
 	// Ensure labeling that uniquely identifies the node and its network
-	commonLabels := FlagsMap{
-		// Explicitly setting an instance label avoids the default
-		// behavior of using the node's URI since the URI isn't
-		// guaranteed stable (e.g. port may change after restart).
-		"instance":          p.node.GetUniqueID(),
-		"network_uuid":      p.node.network.UUID,
-		"node_id":           p.node.NodeID.String(),
-		"is_ephemeral_node": strconv.FormatBool(p.node.IsEphemeral),
-		"network_owner":     p.node.network.Owner,
-	}
-	commonLabels.SetDefaults(githubLabelsFromEnv())
+	commonLabels := p.node.getMonitoringLabels()
 
 	prometheusConfig := []ConfigMap{
 		{
@@ -479,17 +468,3 @@ func watchLogFileForFatal(ctx context.Context, cancelWithCause context.CancelCau
 		}
 	}
 }
-
-func githubLabelsFromEnv() map[string]string {
-	return map[string]string{
-		// prometheus/promtail ignore empty values so including these
-		// labels with empty values outside of a github worker (where
-		// the env vars will not be set) should not be a problem.
-		"gh_repo":        os.Getenv("GH_REPO"),
-		"gh_workflow":    os.Getenv("GH_WORKFLOW"),
-		"gh_run_id":      os.Getenv("GH_RUN_ID"),
-		"gh_run_number":  os.Getenv("GH_RUN_NUMBER"),
-		"gh_run_attempt": os.Getenv("GH_RUN_ATTEMPT"),
-		"gh_job_id":      os.Getenv("GH_JOB_ID"),
-	}
-}