feat: add provisionerd prometheus metrics (#4909)

provisionerd/provisionerd.go

@@ -11,6 +11,8 @@ import (
 	"time"
 
 	"github.com/hashicorp/yamux"
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/promauto"
 	"github.com/spf13/afero"
 	"go.opentelemetry.io/otel/attribute"
 	semconv "go.opentelemetry.io/otel/semconv/v1.11.0"
@@ -41,9 +43,10 @@ type Provisioners map[string]sdkproto.DRPCProvisionerClient
 
 // Options provides customizations to the behavior of a provisioner daemon.
 type Options struct {
-	Filesystem afero.Fs
-	Logger     slog.Logger
-	Tracer     trace.TracerProvider
+	Filesystem     afero.Fs
+	Logger         slog.Logger
+	TracerProvider trace.TracerProvider
+	Metrics        *Metrics
 
 	ForceCancelInterval time.Duration
 	UpdateInterval      time.Duration
@@ -66,14 +69,19 @@ func New(clientDialer Dialer, opts *Options) *Server {
 	if opts.Filesystem == nil {
 		opts.Filesystem = afero.NewOsFs()
 	}
-	if opts.Tracer == nil {
-		opts.Tracer = trace.NewNoopTracerProvider()
+	if opts.TracerProvider == nil {
+		opts.TracerProvider = trace.NewNoopTracerProvider()
+	}
+	if opts.Metrics == nil {
+		reg := prometheus.NewRegistry()
+		mets := NewMetrics(reg)
+		opts.Metrics = &mets
 	}
 
 	ctx, ctxCancel := context.WithCancel(context.Background())
 	daemon := &Server{
 		opts:   opts,
-		tracer: opts.Tracer.Tracer(tracing.TracerName),
+		tracer: opts.TracerProvider.Tracer(tracing.TracerName),
 
 		clientDialer: clientDialer,
 
@@ -103,6 +111,42 @@ type Server struct {
 	activeJob    *runner.Runner
 }
 
+type Metrics struct {
+	Runner runner.Metrics
+}
+
+func NewMetrics(reg prometheus.Registerer) Metrics {
+	auto := promauto.With(reg)
+	durationToFloatMs := func(d time.Duration) float64 {
+		return float64(d.Milliseconds())
+	}
+
+	return Metrics{
+		Runner: runner.Metrics{
+			ConcurrentJobs: auto.NewGaugeVec(prometheus.GaugeOpts{
+				Namespace: "coderd",
+				Subsystem: "provisionerd",
+				Name:      "jobs_current",
+			}, []string{"provisioner"}),
+			JobTimings: auto.NewHistogramVec(prometheus.HistogramOpts{
+				Namespace: "coderd",
+				Subsystem: "provisionerd",
+				Name:      "job_timings_ms",
+				Buckets: []float64{
+					durationToFloatMs(1 * time.Second),
+					durationToFloatMs(10 * time.Second),
+					durationToFloatMs(30 * time.Second),
+					durationToFloatMs(1 * time.Minute),
+					durationToFloatMs(5 * time.Minute),
+					durationToFloatMs(10 * time.Minute),
+					durationToFloatMs(30 * time.Minute),
+					durationToFloatMs(1 * time.Hour),
+				},
+			}, []string{"provisioner", "status"}),
+		},
+	}
+}
+
 // Connect establishes a connection to coderd.
 func (p *Server) connect(ctx context.Context) {
 	// An exponential back-off occurs when the connection is failing to dial.
@@ -282,6 +326,7 @@ func (p *Server) acquireJob(ctx context.Context) {
 		p.opts.UpdateInterval,
 		p.opts.ForceCancelInterval,
 		p.tracer,
+		p.opts.Metrics.Runner,
 	)
 
 	go p.activeJob.Run()

provisionerd/runner/runner.go

@@ -16,6 +16,7 @@ import (
 	"time"
 
 	"github.com/google/uuid"
+	"github.com/prometheus/client_golang/prometheus"
 	"github.com/spf13/afero"
 	"go.opentelemetry.io/otel/codes"
 	semconv "go.opentelemetry.io/otel/semconv/v1.11.0"
@@ -34,6 +35,7 @@ const (
 
 type Runner struct {
 	tracer              trace.Tracer
+	metrics             Metrics
 	job                 *proto.AcquiredJob
 	sender              JobUpdater
 	logger              slog.Logger
@@ -65,6 +67,12 @@ type Runner struct {
 	okToSend bool
 }
 
+type Metrics struct {
+	ConcurrentJobs *prometheus.GaugeVec
+	// JobTimings also counts the total amount of jobs.
+	JobTimings *prometheus.HistogramVec
+}
+
 type JobUpdater interface {
 	UpdateJob(ctx context.Context, in *proto.UpdateJobRequest) (*proto.UpdateJobResponse, error)
 	FailJob(ctx context.Context, in *proto.FailedJob) error
@@ -82,6 +90,7 @@ func NewRunner(
 	updateInterval time.Duration,
 	forceCancelInterval time.Duration,
 	tracer trace.Tracer,
+	metrics Metrics,
 ) *Runner {
 	m := new(sync.Mutex)
 
@@ -91,6 +100,7 @@ func NewRunner(
 
 	return &Runner{
 		tracer:              tracer,
+		metrics:             metrics,
 		job:                 job,
 		sender:              updater,
 		logger:              logger.With(slog.F("job_id", job.JobId)),
@@ -120,9 +130,22 @@ func NewRunner(
 // that goroutine on the context passed into Fail(), and it marks okToSend false to signal us here
 // that this function should not also send a terminal message.
 func (r *Runner) Run() {
+	start := time.Now()
 	ctx, span := r.startTrace(r.notStopped, tracing.FuncName())
 	defer span.End()
 
+	concurrentGauge := r.metrics.ConcurrentJobs.WithLabelValues(r.job.Provisioner)
+	concurrentGauge.Inc()
+	defer func() {
+		status := "success"
+		if r.failedJob != nil {
+			status = "failed"
+		}
+
+		concurrentGauge.Dec()
+		r.metrics.JobTimings.WithLabelValues(r.job.Provisioner, status).Observe(float64(time.Since(start).Milliseconds()))
+	}()
+
 	r.mutex.Lock()
 	defer r.mutex.Unlock()
 	defer r.stop()