bitswap/httpnet: collect metrics

Author: hsanjuan

bitswap/network/httpnet/httpnet.go

@@ -129,6 +129,8 @@ type httpnet struct {
 	supportsHave       bool
 	insecureSkipVerify bool
 	allowlist          map[string]struct{}
+
+	metrics *metrics
 }
 
 // New returns a BitSwapNetwork supported by underlying IPFS host.
@@ -143,6 +145,7 @@ func New(host host.Host, opts ...Option) *httpnet {
 		maxIdleConns:       DefaultMaxIdleConns,
 		supportsHave:       DefaultSupportsHave,
 		insecureSkipVerify: DefaultInsecureSkipVerify,
+		metrics:            newMetrics(),
 	}
 
 	for _, opt := range opts {

bitswap/network/httpnet/metrics.go

@@ -0,0 +1,131 @@
+package httpnet
+
+import (
+	"context"
+
+	imetrics "github.com/ipfs/go-metrics-interface"
+)
+
+var durationHistogramBuckets = []float64{0.05, 0.1, 0.25, 0.5, 1, 2, 5, 10, 30, 60, 120, 240, 480, 960, 1920}
+
+type ctxKeyT string
+
+var ctxKey ctxKeyT = ctxKeyT(imetrics.CtxScopeKey)
+
+func requestsInFlight(ctx context.Context) imetrics.Gauge {
+	return imetrics.NewCtx(ctx, "httpnet_requests_in_flight", "Current number of in-flight requests").Gauge()
+}
+
+func requestsTotal(ctx context.Context) imetrics.Counter {
+	return imetrics.NewCtx(ctx, "httpnet_requests_total", "Total request count").Counter()
+}
+
+func requestsFailure(ctx context.Context) imetrics.Counter {
+	return imetrics.NewCtx(ctx, "httpnet_requests_failure", "Failed (no response, dial error etc) requests count").Counter()
+}
+
+func requestsBodyFailure(ctx context.Context) imetrics.Counter {
+	return imetrics.NewCtx(ctx, "httpnet_requests_body_failure", "Failure count when reading response body").Counter()
+}
+
+func statusNotFound(ctx context.Context) imetrics.Counter {
+	return imetrics.NewCtx(ctx, "httpnet_status_404", "Request count with NotFound status").Counter()
+}
+
+func statusGone(ctx context.Context) imetrics.Counter {
+	return imetrics.NewCtx(ctx, "httpnet_status_410", "Request count with Gone status").Counter()
+}
+
+func statusForbidden(ctx context.Context) imetrics.Counter {
+	return imetrics.NewCtx(ctx, "httpnet_status_403", "Request count with Forbidden status").Counter()
+}
+
+func statusUnavailableForLegalReasons(ctx context.Context) imetrics.Counter {
+	return imetrics.NewCtx(ctx, "httpnet_status_451", "Request count with Unavailable For Legal Reasons status").Counter()
+}
+
+func statusOK(ctx context.Context) imetrics.Counter {
+	return imetrics.NewCtx(ctx, "httpnet_status_200", "Request count with OK status").Counter()
+}
+
+func statusTooManyRequests(ctx context.Context) imetrics.Counter {
+	return imetrics.NewCtx(ctx, "httpnet_status_429", "Request count with Too Many Requests status").Counter()
+}
+
+func statusServiceUnavailable(ctx context.Context) imetrics.Counter {
+	return imetrics.NewCtx(ctx, "httpnet_status_503", "Request count with Service Unavailable status").Counter()
+}
+
+func statusInternalServerError(ctx context.Context) imetrics.Counter {
+	return imetrics.NewCtx(ctx, "httpnet_status_500", "Request count with Internal Server Error status").Counter()
+}
+
+func statusOthers(ctx context.Context) imetrics.Counter {
+	return imetrics.NewCtx(ctx, "httpnet_status_others", "Request count with other status codes").Counter()
+}
+
+func requestTime(ctx context.Context) imetrics.Histogram {
+	return imetrics.NewCtx(ctx, "httpnet_request_duration_seconds", "Histogram of request durations").Histogram(durationHistogramBuckets)
+}
+
+type metrics struct {
+	RequestsInFlight                 imetrics.Gauge
+	RequestsTotal                    imetrics.Counter
+	RequestsFailure                  imetrics.Counter
+	RequestsBodyFailure              imetrics.Counter
+	StatusNotFound                   imetrics.Counter
+	StatusGone                       imetrics.Counter
+	StatusForbidden                  imetrics.Counter
+	StatusUnavailableForLegalReasons imetrics.Counter
+	StatusOK                         imetrics.Counter
+	StatusTooManyRequests            imetrics.Counter
+	StatusServiceUnavailable         imetrics.Counter
+	StatusInternalServerError        imetrics.Counter
+	StatusOthers                     imetrics.Counter
+	RequestTime                      imetrics.Histogram
+}
+
+func newMetrics() *metrics {
+	ctx := context.WithValue(context.Background(), ctxKey, "exchange")
+
+	return &metrics{
+		RequestsInFlight:                 requestsInFlight(ctx),
+		RequestsTotal:                    requestsTotal(ctx),
+		RequestsFailure:                  requestsFailure(ctx),
+		RequestsBodyFailure:              requestsBodyFailure(ctx),
+		StatusNotFound:                   statusNotFound(ctx),
+		StatusGone:                       statusGone(ctx),
+		StatusForbidden:                  statusForbidden(ctx),
+		StatusUnavailableForLegalReasons: statusUnavailableForLegalReasons(ctx),
+		StatusOK:                         statusOK(ctx),
+		StatusTooManyRequests:            statusTooManyRequests(ctx),
+		StatusServiceUnavailable:         statusServiceUnavailable(ctx),
+		StatusInternalServerError:        statusInternalServerError(ctx),
+		StatusOthers:                     statusOthers(ctx),
+		RequestTime:                      requestTime(ctx),
+	}
+}
+
+func (m *metrics) updateStatusCounter(statusCode int) {
+	m.RequestsTotal.Inc()
+	switch statusCode {
+	case 404:
+		m.StatusNotFound.Inc()
+	case 410:
+		m.StatusGone.Inc()
+	case 403:
+		m.StatusForbidden.Inc()
+	case 451:
+		m.StatusUnavailableForLegalReasons.Inc()
+	case 200:
+		m.StatusOK.Inc()
+	case 429:
+		m.StatusTooManyRequests.Inc()
+	case 503:
+		m.StatusServiceUnavailable.Inc()
+	case 500:
+		m.StatusInternalServerError.Inc()
+	default:
+		m.StatusOthers.Inc()
+	}
+}

bitswap/network/httpnet/msg_sender.go

@@ -207,9 +207,13 @@ func (sender *httpMsgSender) tryURL(ctx context.Context, u *senderURL, entry bsm
 
 	log.Debugf("%s %q", method, req.URL)
 	atomic.AddUint64(&sender.ht.stats.MessagesSent, 1)
+	reqStart := time.Now()
+	sender.ht.metrics.RequestsInFlight.Inc()
 	resp, err := sender.ht.client.Do(req)
 	if err != nil {
 		err = fmt.Errorf("error making request to %q: %w", req.URL, err)
+		sender.ht.metrics.RequestsFailure.Inc()
+		sender.ht.metrics.RequestsInFlight.Dec()
 		log.Debug(err)
 		// Something prevents us from making a request.  We cannot
 		// dial, or setup the connection perhaps.  This counts as
@@ -241,13 +245,22 @@ func (sender *httpMsgSender) tryURL(ctx context.Context, u *senderURL, entry bsm
 	if err != nil {
 		// treat this as server error
 		err = fmt.Errorf("error reading body from %q: %w", req.URL, err)
+		sender.ht.metrics.RequestsBodyFailure.Inc()
+		sender.ht.metrics.RequestsInFlight.Dec()
 		log.Debug(err)
 		return &senderError{
 			Type: typeServer,
 			Err:  err,
 		}
 	}
+	reqDuration := time.Since(reqStart)
+
+	sender.ht.metrics.RequestsInFlight.Dec()
+	sender.ht.metrics.RequestTime.Observe(float64(reqDuration) / float64(time.Second))
+	sender.ht.metrics.updateStatusCounter(resp.StatusCode)
+
 	sender.ht.connEvtMgr.OnMessage(sender.peer)
+
 	switch resp.StatusCode {
 	// Valid responses signaling unavailability of the
 	// content.