fix consul_autopilot_healthy metric emission (#11231)

https://github.com/hashicorp/consul/issues/10730

.changelog/11231.txt

@@ -0,0 +1,3 @@
+```release-note:bug
+telemetry: fixes a bug with Prometheus consul_autopilot_healthy metric where 0 is reported instead of NaN on servers.
+```

agent/consul/autopilot.go

@@ -9,6 +9,7 @@ import (
 	"github.com/hashicorp/raft"
 	autopilot "github.com/hashicorp/raft-autopilot"
 	"github.com/hashicorp/serf/serf"
+	"math"
 
 	"github.com/hashicorp/consul/agent/metadata"
 	"github.com/hashicorp/consul/agent/structs"
@@ -52,6 +53,12 @@ func (d *AutopilotDelegate) NotifyState(state *autopilot.State) {
 		} else {
 			metrics.SetGauge([]string{"autopilot", "healthy"}, 0)
 		}
+	} else {
+
+		// if we are not a leader, emit NaN per
+		// https://www.consul.io/docs/agent/telemetry#autopilot
+		metrics.SetGauge([]string{"autopilot", "healthy"}, float32(math.NaN()))
+
 	}
 }
 
@@ -74,6 +81,8 @@ func (s *Server) initAutopilot(config *Config) {
 		autopilot.WithUpdateInterval(config.ServerHealthInterval),
 		autopilot.WithPromoter(s.autopilotPromoter()),
 	)
+
+	metrics.SetGauge([]string{"autopilot", "healthy"}, float32(math.NaN()))
 }
 
 func (s *Server) autopilotServers() map[raft.ServerID]*autopilot.Server {

agent/metrics_test.go

@@ -0,0 +1,53 @@
+package agent
+
+import (
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"testing"
+)
+
+// TestHTTPHandlers_AgentMetrics_ConsulAutopilotHealthy_Prometheus adds testing around
+// the published autopilot metrics on https://www.consul.io/docs/agent/telemetry#autopilot
+func TestHTTPHandlers_AgentMetrics_ConsulAutopilotHealthy_Prometheus(t *testing.T) {
+	checkForShortTesting(t)
+	// This test cannot use t.Parallel() since we modify global state, ie the global metrics instance
+
+	// don't bootstrap agent so as not to
+	// become a leader
+	hcl := `
+	telemetry = {
+		prometheus_retention_time = "5s",
+		disable_hostname = true
+	}
+	bootstrap = false
+	`
+
+	a := StartTestAgent(t, TestAgent{HCL: hcl})
+	defer a.Shutdown()
+
+	req, err := http.NewRequest("GET", "/v1/agent/metrics?format=prometheus", nil)
+	if err != nil {
+		t.Fatalf("Failed to generate new http request. err: %v", err)
+	}
+
+	respRec := httptest.NewRecorder()
+	_, err = a.srv.AgentMetrics(respRec, req)
+	if err != nil {
+		t.Fatalf("Failed to serve agent metrics. err: %v", err)
+	}
+
+	t.Run("Check consul_autopilot_healthy metric value on startup", func(t *testing.T) {
+		target := "consul_autopilot_healthy NaN"
+		keyValue := strings.Split(target, " ")
+		if !strings.Contains(respRec.Body.String(), target) {
+			t.Fatalf("Could not find the metric \"%s\" with value \"%s\" in the /v1/agent/metrics response", keyValue[0], keyValue[1])
+		}
+	})
+}
+
+func checkForShortTesting(t *testing.T) {
+	if testing.Short() {
+		t.Skip("too slow for testing.Short")
+	}
+}

agent/testagent.go

@@ -185,8 +185,13 @@ func (a *TestAgent) Start(t *testing.T) error {
 		}
 		result, err := config.Load(opts)
 		if result.RuntimeConfig != nil {
+			// If prom metrics need to be enabled, do not disable telemetry
+			if result.RuntimeConfig.Telemetry.PrometheusOpts.Expiration > 0 {
+				result.RuntimeConfig.Telemetry.Disable = false
+			} else {
 				result.RuntimeConfig.Telemetry.Disable = true
 			}
+		}
 		return result, err
 	}
 	bd, err := NewBaseDeps(loader, logOutput)
@@ -195,7 +200,11 @@ func (a *TestAgent) Start(t *testing.T) error {
 	}
 
 	bd.Logger = logger
+	// if we are not testing telemetry things, let's use a "mock" sink for metrics
+	if bd.RuntimeConfig.Telemetry.Disable {
 		bd.MetricsHandler = metrics.NewInmemSink(1*time.Second, time.Minute)
+	}
+
 	a.Config = bd.RuntimeConfig
 
 	agent, err := New(bd)