From 0a0782bb254a793f9179d959d22dd8da081a5323 Mon Sep 17 00:00:00 2001
From: "l.subbotin" <subveles@gmail.com>
Date: Fri, 18 Oct 2024 11:17:07 +0200
Subject: [PATCH] Added support for getHealth method, introducing
 solana_is_healthy, solana_num_slots_behind (after refactoring)

---
 cmd/solana_exporter/exporter.go |  1 +
 cmd/solana_exporter/health.go   | 85 +++++++++++++++++++++++++++++++++
 pkg/rpc/client.go               |  1 +
 pkg/rpc/gethealth.go            | 34 +++++++++++++
 4 files changed, 121 insertions(+)
 create mode 100644 cmd/solana_exporter/health.go
 create mode 100644 pkg/rpc/gethealth.go

diff --git a/cmd/solana_exporter/exporter.go b/cmd/solana_exporter/exporter.go
index eb7cc2e..4070c0b 100644
--- a/cmd/solana_exporter/exporter.go
+++ b/cmd/solana_exporter/exporter.go
@@ -221,6 +221,7 @@ func main() {
 	ctx, cancel := context.WithCancel(ctx)
 	defer cancel()
 	go slotWatcher.WatchSlots(ctx, collector.slotPace)
+	go collector.WatchHealth(context.Background())
 
 	prometheus.MustRegister(collector)
 	http.Handle("/metrics", promhttp.Handler())
diff --git a/cmd/solana_exporter/health.go b/cmd/solana_exporter/health.go
new file mode 100644
index 0000000..42fe8eb
--- /dev/null
+++ b/cmd/solana_exporter/health.go
@@ -0,0 +1,85 @@
+package main
+
+import (
+	"context"
+	"errors"
+	"fmt"
+	"github.com/asymmetric-research/solana_exporter/pkg/rpc"
+	"github.com/prometheus/client_golang/prometheus"
+	"k8s.io/klog/v2"
+	"strconv"
+	"time"
+)
+
+var (
+	isHealthy = prometheus.NewGauge(prometheus.GaugeOpts{
+		Name: "solana_is_healthy",
+		Help: "Is node healthy",
+	})
+
+	numSlotsBehind = prometheus.NewGauge(prometheus.GaugeOpts{
+		Name: "solana_num_slots_behind",
+		Help: "Number of slots behind",
+	})
+)
+
+func init() {
+	prometheus.MustRegister(isHealthy)
+	prometheus.MustRegister(numSlotsBehind)
+
+}
+func extractNumSlotsBehind(data map[string]any) (int, error) {
+	if val, ok := data["NumSlotsBehind"]; ok {
+		// Type assert if it's a float64 (common for numbers in JSON)
+		switch v := val.(type) {
+		case float64:
+			return int(v), nil
+		case string:
+			// If it's a string, try to convert it to an int
+			num, err := strconv.Atoi(v)
+			if err != nil {
+				return 0, fmt.Errorf("failed to convert string to int: %w", err)
+			}
+			return num, nil
+		default:
+			return 0, fmt.Errorf("unexpected type for NumSlotsBehind: %T", v)
+		}
+	}
+
+	return 0, fmt.Errorf("NumSlotsBehind key not found in data")
+}
+
+func (c *SolanaCollector) WatchHealth(ctx context.Context) {
+	ticker := time.NewTicker(slotPacerSchedule)
+
+	for {
+		<-ticker.C
+
+		// Get current slot height and epoch info
+
+		ctx, cancel := context.WithTimeout(ctx, 5*time.Second)
+		_, err := c.rpcClient.GetHealth(ctx)
+		if err != nil {
+			klog.Infof("failed to fetch info info, retrying: %v", err)
+			cancel()
+			continue
+		}
+		cancel()
+		isNodeHealthy := 1
+		nodeNumSlotsBehind := 0
+		if err != nil {
+			var rpcError *rpc.RPCError
+			if errors.As(err, &rpcError) {
+				if rpcError.Code != 0 {
+					isNodeHealthy = 0
+				}
+				nodeNumSlotsBehind, _ = extractNumSlotsBehind(rpcError.Data)
+
+			}
+
+		}
+		isHealthy.Set(float64(isNodeHealthy))
+
+		numSlotsBehind.Set(float64(nodeNumSlotsBehind))
+	}
+}
diff --git a/pkg/rpc/client.go b/pkg/rpc/client.go
index a4817d1..5db9e63 100644
--- a/pkg/rpc/client.go
+++ b/pkg/rpc/client.go
@@ -73,6 +73,7 @@ type Provider interface {
 	GetLeaderSchedule(ctx context.Context, commitment Commitment, slot int64) (map[string][]int64, error)
 
 	GetBlock(ctx context.Context, commitment Commitment, slot int64, transactionDetails string) (*Block, error)
+	GetHealth(ctx context.Context) (*string, error)
 }
 
 func (c Commitment) MarshalJSON() ([]byte, error) {
diff --git a/pkg/rpc/gethealth.go b/pkg/rpc/gethealth.go
new file mode 100644
index 0000000..423f625
--- /dev/null
+++ b/pkg/rpc/gethealth.go
@@ -0,0 +1,34 @@
+package rpc
+
+import (
+	"context"
+)
+
+type (
+	ErrorData struct {
+		NumSlotsBehind int64 `json:"numSlotsBehind"`
+	}
+
+	GetHealthRpcError struct {
+		Message string    `json:"message"`
+		Data    ErrorData `json:"data"`
+		Code    int64     `json:"code"`
+	}
+
+	getHealthResponse struct {
+		jsonrpc string
+		Result  string   `json:"result"`
+		Error   RPCError `json:"error"`
+		Id      int      `json:"id"`
+	}
+)
+
+// https://docs.solana.com/developing/clients/jsonrpc-api#gethealth
+func (c *Client) GetHealth(ctx context.Context) (*string, error) {
+	var resp response[string]
+
+	if err := getResponse(ctx, c, "getHealth", []any{}, &resp); err != nil {
+		return nil, err
+	}
+	return &resp.Result, nil
+}