feat: Add active users prometheus metric (#3406)

This  allows deployments using our Prometheus export t determine
the number of active users in the past hour.

The interval is an hour to align with API key last used refresh times.

SSH connections poll to check shutdown time, so this will be accurate
even on long-running connections without dashboard requests.

cli/server.go

@@ -30,6 +30,7 @@ import (
 	"github.com/google/uuid"
 	"github.com/pion/turn/v2"
 	"github.com/pion/webrtc/v3"
+	"github.com/prometheus/client_golang/prometheus"
 	"github.com/prometheus/client_golang/prometheus/promhttp"
 	"github.com/spf13/afero"
 	"github.com/spf13/cobra"
@@ -53,6 +54,7 @@ import (
 	"github.com/coder/coder/coderd/database/databasefake"
 	"github.com/coder/coder/coderd/devtunnel"
 	"github.com/coder/coder/coderd/gitsshkey"
+	"github.com/coder/coder/coderd/prometheusmetrics"
 	"github.com/coder/coder/coderd/telemetry"
 	"github.com/coder/coder/coderd/tracing"
 	"github.com/coder/coder/coderd/turnconn"
@@ -392,6 +394,26 @@ func server() *cobra.Command {
 				defer options.Telemetry.Close()
 			}
 
+			// This prevents the pprof import from being accidentally deleted.
+			_ = pprof.Handler
+			if pprofEnabled {
+				//nolint:revive
+				defer serveHandler(ctx, logger, nil, pprofAddress, "pprof")()
+			}
+			if promEnabled {
+				options.PrometheusRegistry = prometheus.NewRegistry()
+				closeFunc, err := prometheusmetrics.ActiveUsers(ctx, options.PrometheusRegistry, options.Database, 0)
+				if err != nil {
+					return xerrors.Errorf("register active users prometheus metric: %w", err)
+				}
+				defer closeFunc()
+
+				//nolint:revive
+				defer serveHandler(ctx, logger, promhttp.InstrumentMetricHandler(
+					options.PrometheusRegistry, promhttp.HandlerFor(options.PrometheusRegistry, promhttp.HandlerOpts{}),
+				), promAddress, "prometheus")()
+			}
+
 			coderAPI := coderd.New(options)
 			defer coderAPI.Close()
 
@@ -406,17 +428,6 @@ func server() *cobra.Command {
 				}
 			}
 
-			// This prevents the pprof import from being accidentally deleted.
-			_ = pprof.Handler
-			if pprofEnabled {
-				//nolint:revive
-				defer serveHandler(ctx, logger, nil, pprofAddress, "pprof")()
-			}
-			if promEnabled {
-				//nolint:revive
-				defer serveHandler(ctx, logger, promhttp.Handler(), promAddress, "prometheus")()
-			}
-
 			// Since errCh only has one buffered slot, all routines
 			// sending on it must be wrapped in a select/default to
 			// avoid leaving dangling goroutines waiting for the

cli/server_test.go

@@ -1,6 +1,7 @@
 package cli_test
 
 import (
+	"bufio"
 	"context"
 	"crypto/ecdsa"
 	"crypto/elliptic"
@@ -10,6 +11,7 @@ import (
 	"crypto/x509/pkix"
 	"encoding/json"
 	"encoding/pem"
+	"fmt"
 	"math/big"
 	"net"
 	"net/http"
@@ -17,10 +19,13 @@ import (
 	"net/url"
 	"os"
 	"runtime"
+	"strconv"
+	"strings"
 	"testing"
 	"time"
 
 	"github.com/go-chi/chi"
+	"github.com/stretchr/testify/assert"
 	"github.com/stretchr/testify/require"
 	"go.uber.org/goleak"
 
@@ -374,6 +379,56 @@ func TestServer(t *testing.T) {
 		cancelFunc()
 		<-errC
 	})
+	t.Run("Prometheus", func(t *testing.T) {
+		t.Parallel()
+		ctx, cancelFunc := context.WithCancel(context.Background())
+		defer cancelFunc()
+
+		random, err := net.Listen("tcp", "127.0.0.1:0")
+		require.NoError(t, err)
+		_ = random.Close()
+		tcpAddr, valid := random.Addr().(*net.TCPAddr)
+		require.True(t, valid)
+		randomPort := tcpAddr.Port
+
+		root, cfg := clitest.New(t,
+			"server",
+			"--in-memory",
+			"--address", ":0",
+			"--provisioner-daemons", "1",
+			"--prometheus-enable",
+			"--prometheus-address", ":"+strconv.Itoa(randomPort),
+			"--cache-dir", t.TempDir(),
+		)
+		serverErr := make(chan error, 1)
+		go func() {
+			serverErr <- root.ExecuteContext(ctx)
+		}()
+		_ = waitAccessURL(t, cfg)
+
+		var res *http.Response
+		require.Eventually(t, func() bool {
+			req, err := http.NewRequestWithContext(ctx, "GET", fmt.Sprintf("http://127.0.0.1:%d", randomPort), nil)
+			assert.NoError(t, err)
+			res, err = http.DefaultClient.Do(req)
+			return err == nil
+		}, testutil.WaitShort, testutil.IntervalFast)
+
+		scanner := bufio.NewScanner(res.Body)
+		hasActiveUsers := false
+		for scanner.Scan() {
+			// This metric is manually registered to be tracked in the server. That's
+			// why we test it's tracked here.
+			if strings.HasPrefix(scanner.Text(), "coderd_api_active_users_duration_hour") {
+				hasActiveUsers = true
+				continue
+			}
+		}
+		require.NoError(t, scanner.Err())
+		require.True(t, hasActiveUsers)
+		cancelFunc()
+		<-serverErr
+	})
 }
 
 func generateTLSCertificate(t testing.TB) (certPath, keyPath string) {