From a7a00db86c1b3fe5f132c6bf20a7fc2e06db0d03 Mon Sep 17 00:00:00 2001
From: Dan Fuhry <dan@fuhry.com>
Date: Thu, 3 Apr 2025 23:38:58 -0400
Subject: [PATCH] health-exporter: include node name in labels, publish under
 correct otel-tls service

---
 sd/etcd_factory.go         | 15 +++++++++------
 sd/health_exporter/main.go | 10 +++++++++-
 sd/monitor.go              |  2 ++
 3 files changed, 20 insertions(+), 7 deletions(-)

diff --git a/sd/etcd_factory.go b/sd/etcd_factory.go
index f264ed4..5212ace 100644
--- a/sd/etcd_factory.go
+++ b/sd/etcd_factory.go
@@ -37,7 +37,7 @@ func NewDefaultEtcdClient() (*etcd_client.Client, error) {
 				return nil, err
 			}
 
-			clientSingleton, err = NewEtcdClient(id)
+			clientSingleton, err = NewEtcdClient(id, etcdDiscoveryDomain)
 			if err == nil {
 				break
 			}
@@ -52,14 +52,14 @@ func NewDefaultEtcdClient() (*etcd_client.Client, error) {
 	return clientSingleton, nil
 }
 
-func NewEtcdClient(id mtls.Identity) (*etcd_client.Client, error) {
+func NewEtcdClient(id mtls.Identity, domain string) (*etcd_client.Client, error) {
 	deadline := time.Now().Add(time.Millisecond * time.Duration(etcdStartupTimeoutMs))
 
 	return NewEtcdClientWithDeadline(
-		id, deadline)
+		id, domain, deadline)
 }
 
-func NewEtcdClientWithDeadline(id mtls.Identity, deadline time.Time) (*etcd_client.Client, error) {
+func NewEtcdClientWithDeadline(id mtls.Identity, domain string, deadline time.Time) (*etcd_client.Client, error) {
 	var client *etcd_client.Client
 
 	tlsConfig, err := id.TlsConfig(context.Background())
@@ -68,9 +68,12 @@ func NewEtcdClientWithDeadline(id mtls.Identity, deadline time.Time) (*etcd_clie
 	}
 	pnv := mtls.NewPeerNameVerifier()
 	pnv.AllowFrom(mtls.Service, "etcd")
-	pnv.ConfigureClient(tlsConfig)
+	err = pnv.ConfigureClient(tlsConfig)
+	if err != nil {
+		return nil, err
+	}
 
-	clients, err := srv.GetClient("etcd-client", etcdDiscoveryDomain, "")
+	clients, err := srv.GetClient("etcd-client", domain, "")
 	if err != nil {
 		return nil, fmt.Errorf("failed to discover peers: %v", err)
 	}
diff --git a/sd/health_exporter/main.go b/sd/health_exporter/main.go
index 6b7c87e..bd58ab1 100644
--- a/sd/health_exporter/main.go
+++ b/sd/health_exporter/main.go
@@ -9,6 +9,7 @@ import (
 	"net"
 	"net/http"
 	"os/signal"
+	"strings"
 	"sync"
 	"syscall"
 
@@ -64,6 +65,12 @@ func main() {
 				"protocol=%q,service=%q,monitor_host=%q,monitor_region=%q",
 				s.Protocol, s.Service, s.MonitorHost, s.MonitorRegion)
 
+			if s.Node == "" {
+				nodeName := strings.Split(s.MonitorHost, ".")[0]
+				labels += fmt.Sprintf(",node_name=%q", nodeName)
+			} else {
+				labels += fmt.Sprintf(",node_name=%q", s.Shard)
+			}
 			if s.Shard != "" {
 				labels += fmt.Sprintf(",shard_name=%q", s.Shard)
 			}
@@ -88,7 +95,8 @@ func main() {
 	publisher := &sd.SDPublisher{
 		AdvertisePort: tcpAddr.AddrPort().Port(),
 		Protocol:      sd.ProtocolTCP,
-		Service:       "health-exporter",
+		Service:       "otel-tls",
+		ShardName:     "health-exporter",
 	}
 	publisher.Publish(ctx)
 
diff --git a/sd/monitor.go b/sd/monitor.go
index 579d6ab..656af85 100644
--- a/sd/monitor.go
+++ b/sd/monitor.go
@@ -25,6 +25,7 @@ type SDHealthReport struct {
 	Up            bool
 	Domain        string
 	Shard         string
+	Node          string
 	ShardRegion   string
 	Protocol      string
 	Service       string
@@ -114,6 +115,7 @@ func (sdp *healthCheckingPublisher) updateHealth(ctx context.Context, lease *etc
 		SocketPort:    sdp.publisher.AdvertisePort,
 		Domain:        sdp.publisher.Domain,
 		Shard:         sdp.publisher.ShardName,
+		Node:          sdp.publisher.NodeName,
 		ShardRegion:   strings.Split(sdp.publisher.AdvertiseHost, ".")[1],
 		Protocol:      sdp.healthcheck.ServiceDefinition().Engine,
 		MonitorHost:   hostname.Fqdn(),
-- 
2.50.1