Update metrics (#147)

* Remove /health and /metrics req logging; closes #127

* Move metrics to metrics package

* Prometheus for index and car lookups; closes #126

* Cleanup metrics; closes #128

cmd-rpc.go

@@ -15,6 +15,7 @@ import (
 	"github.com/allegro/bigcache/v3"
 	"github.com/fsnotify/fsnotify"
 	hugecache "github.com/rpcpool/yellowstone-faithful/huge-cache"
+	"github.com/rpcpool/yellowstone-faithful/metrics"
 	splitcarfetcher "github.com/rpcpool/yellowstone-faithful/split-car-fetcher"
 	"github.com/ryanuber/go-glob"
 	"github.com/urfave/cli/v2"
@@ -202,13 +203,13 @@ func newCmd_rpc() *cli.Command {
 							return nil
 						}()
 						if err != nil {
-							metrics_epochsAvailable.WithLabelValues(fmt.Sprintf("%d", epochNum)).Set(0)
+							metrics.EpochsAvailable.WithLabelValues(fmt.Sprintf("%d", epochNum)).Set(0)
 							klog.Error(err)
 							numFailed.Add(1)
 							// NOTE: DO NOT return the error here, as we want to continue loading other epochs
 							return nil
 						}
-						metrics_epochsAvailable.WithLabelValues(fmt.Sprintf("%d", epochNum)).Set(1)
+						metrics.EpochsAvailable.WithLabelValues(fmt.Sprintf("%d", epochNum)).Set(1)
 						numSucceeded.Add(1)
 						return nil
 					})
@@ -275,7 +276,7 @@ func newCmd_rpc() *cli.Command {
 									return
 								}
 								klog.V(2).Infof("Epoch %d added/replaced in %s", epoch.Epoch(), time.Since(startedAt))
-								metrics_epochsAvailable.WithLabelValues(fmt.Sprintf("%d", epoch.Epoch())).Set(1)
+								metrics.EpochsAvailable.WithLabelValues(fmt.Sprintf("%d", epoch.Epoch())).Set(1)
 							}
 						case fsnotify.Create:
 							{
@@ -298,7 +299,7 @@ func newCmd_rpc() *cli.Command {
 									return
 								}
 								klog.V(2).Infof("Epoch %d added in %s", epoch.Epoch(), time.Since(startedAt))
-								metrics_epochsAvailable.WithLabelValues(fmt.Sprintf("%d", epoch.Epoch())).Set(1)
+								metrics.EpochsAvailable.WithLabelValues(fmt.Sprintf("%d", epoch.Epoch())).Set(1)
 							}
 						case fsnotify.Remove:
 							{
@@ -310,7 +311,7 @@ func newCmd_rpc() *cli.Command {
 									klog.Errorf("error removing epoch for config file %q: %s", event.Name, err.Error())
 								}
 								klog.V(2).Infof("Epoch %d removed in %s", epNumber, time.Since(startedAt))
-								metrics_epochsAvailable.WithLabelValues(fmt.Sprintf("%d", epNumber)).Set(0)
+								metrics.EpochsAvailable.WithLabelValues(fmt.Sprintf("%d", epNumber)).Set(0)
 							}
 						case fsnotify.Rename:
 							klog.V(3).Infof("File %q was renamed; do nothing", event.Name)

http-range.go

@@ -2,9 +2,11 @@ package main
 
 import (
 	"io"
+	"path/filepath"
 	"strings"
 	"time"
 
+	"github.com/rpcpool/yellowstone-faithful/metrics"
 	"k8s.io/klog/v2"
 )
 
@@ -43,6 +45,15 @@ func (r *readCloserWrapper) ReadAt(p []byte, off int64) (n int, err error) {
 			// if has suffix .index, then it's an index file
 			if strings.HasSuffix(r.name, ".index") {
 				prefix = icon + azureBG("[READ-INDEX]")
+				// get the index name, which is the part before the .index suffix, after the last .
+				indexName := strings.TrimSuffix(r.name, ".index")
+				// split the index name by . and get the last part
+				byDot := strings.Split(indexName, ".")
+				if len(byDot) > 0 {
+					indexName = byDot[len(byDot)-1]
+				}
+				// TODO: distinguish between remote and local index reads
+				metrics.IndexLookupHistogram.WithLabelValues(indexName).Observe(float64(took.Seconds()))
 			}
 			// if has suffix .car, then it's a car file
 			if strings.HasSuffix(r.name, ".car") || r.isSplitCar {
@@ -51,6 +62,9 @@ func (r *readCloserWrapper) ReadAt(p []byte, off int64) (n int, err error) {
 				} else {
 					prefix = icon + purpleBG("[READ-CAR]")
 				}
+				carName := filepath.Base(r.name)
+				// TODO: distinguish between remote and local index reads
+				metrics.CarLookupHistogram.WithLabelValues(carName).Observe(float64(took.Seconds()))
 			}
 			klog.V(5).Infof(prefix+" %s:%d+%d (%s)\n", (r.name), off, len(p), took)
 		}

metrics/metrics.go

@@ -0,0 +1,90 @@
+package metrics
+
+import (
+	"github.com/prometheus/client_golang/prometheus"
+	"github.com/prometheus/client_golang/prometheus/promauto"
+)
+
+var RpcRequestByMethod = promauto.NewCounterVec(
+	prometheus.CounterOpts{
+		Name: "rpc_requests_by_method",
+		Help: "RPC requests by method",
+	},
+	[]string{"method"},
+)
+
+var EpochsAvailable = promauto.NewGaugeVec(
+	prometheus.GaugeOpts{
+		Name: "epoch_available",
+		Help: "Epochs available",
+	},
+	[]string{"epoch"},
+)
+
+var StatusCode = promauto.NewCounterVec(
+	prometheus.CounterOpts{
+		Name: "status_code",
+		Help: "Status code",
+	},
+	[]string{"code"},
+)
+
+var MethodToCode = promauto.NewCounterVec(
+	prometheus.CounterOpts{
+		Name: "method_to_code",
+		Help: "Method to code",
+	},
+	[]string{"method", "code"},
+)
+
+var MethodToSuccessOrFailure = promauto.NewCounterVec(
+	prometheus.CounterOpts{
+		Name: "method_to_success_or_failure",
+		Help: "Method to success or failure",
+	},
+	[]string{"method", "status"},
+)
+
+var MethodToNumProxied = promauto.NewCounterVec(
+	prometheus.CounterOpts{
+		Name: "method_to_num_proxied",
+		Help: "Method to num proxied",
+	},
+	[]string{"method"},
+)
+
+// - Version information of this binary
+var Version = promauto.NewGaugeVec(
+	prometheus.GaugeOpts{
+		Name: "version",
+		Help: "Version information of this binary",
+	},
+	[]string{"started_at", "tag", "commit", "compiler", "goarch", "goos", "goamd64", "vcs", "vcs_revision", "vcs_time", "vcs_modified"},
+)
+
+var IndexLookupHistogram = promauto.NewHistogramVec(
+	prometheus.HistogramOpts{
+		Name:    "index_lookup_latency_histogram",
+		Help:    "Index lookup latency",
+		Buckets: prometheus.ExponentialBuckets(0.000001, 10, 10),
+	},
+	[]string{"index_type"},
+)
+
+var CarLookupHistogram = promauto.NewHistogramVec(
+	prometheus.HistogramOpts{
+		Name:    "car_lookup_latency_histogram",
+		Help:    "Car lookup latency",
+		Buckets: prometheus.ExponentialBuckets(0.000001, 10, 10),
+	},
+	[]string{"car"},
+)
+
+var RpcResponseLatencyHistogram = promauto.NewHistogramVec(
+	prometheus.HistogramOpts{
+		Name:    "rpc_response_latency_histogram",
+		Help:    "RPC response latency histogram",
+		Buckets: prometheus.ExponentialBuckets(0.000001, 10, 10),
+	},
+	[]string{"rpc_method"},
+)