From 374500ae4c8edb3d5231fb82c9b1147957082fea Mon Sep 17 00:00:00 2001
From: pschork <354473+pschork@users.noreply.github.com>
Date: Tue, 21 May 2024 12:58:35 -0700
Subject: [PATCH] Improved reachability check 404 handling (#571)

---
 disperser/dataapi/queried_operators_handlers.go |  6 +++---
 node/node.go                                    | 12 +++++++++---
 2 files changed, 12 insertions(+), 6 deletions(-)

diff --git a/disperser/dataapi/queried_operators_handlers.go b/disperser/dataapi/queried_operators_handlers.go
index 0714c1016a..b2ff89d78b 100644
--- a/disperser/dataapi/queried_operators_handlers.go
+++ b/disperser/dataapi/queried_operators_handlers.go
@@ -169,8 +169,8 @@ func ValidOperatorIP(address string, logger logging.Logger) bool {
 func (s *server) probeOperatorPorts(ctx context.Context, operatorId string) (*OperatorPortCheckResponse, error) {
 	operatorInfo, err := s.subgraphClient.QueryOperatorInfoByOperatorId(context.Background(), operatorId)
 	if err != nil {
-		s.logger.Warn("failed to fetch operator info", "error", err)
-		return &OperatorPortCheckResponse{}, errors.New("not found")
+		s.logger.Warn("failed to fetch operator info", "operatorId", operatorId, "error", err)
+		return &OperatorPortCheckResponse{}, errors.New("operator info not found")
 	}
 
 	operatorSocket := core.OperatorSocket(operatorInfo.Socket)
@@ -190,7 +190,7 @@ func (s *server) probeOperatorPorts(ctx context.Context, operatorId string) (*Op
 	}
 
 	// Log the online status
-	s.logger.Info("operator port check response", portCheckResponse)
+	s.logger.Info("operator port check response", "response", portCheckResponse)
 
 	// Send the metadata to the results channel
 	return portCheckResponse, nil
diff --git a/node/node.go b/node/node.go
index a9ecd88726..76f2b60e4d 100644
--- a/node/node.go
+++ b/node/node.go
@@ -12,6 +12,7 @@ import (
 	"net/http"
 	"net/url"
 	"os"
+	"strings"
 	"sync"
 	"time"
 
@@ -477,7 +478,7 @@ func (n *Node) checkNodeReachability() {
 		return
 	}
 
-	checkUrl, err := url.Parse(fmt.Sprintf("%s/api/v1/operators-info/port-check?operator_id=%s", n.Config.DataApiUrl, n.Config.ID.Hex()))
+	checkUrl, err := url.Parse(fmt.Sprintf("%s/api/v1/operators-info/port-check?operator_id=%s", strings.TrimSuffix(n.Config.DataApiUrl, "/"), n.Config.ID.Hex()))
 	if err != nil {
 		n.Logger.Error("Reachability check failed - invalid check url", err, "checkUrl", checkUrl.String())
 		return
@@ -497,7 +498,12 @@ func (n *Node) checkNodeReachability() {
 			n.Logger.Error("Reachability check request failed", err)
 			continue
 		} else if resp.StatusCode == 404 {
-			n.Logger.Error("Reachability check failed - operator id not found", "status", resp.StatusCode, "operator_id", n.Config.ID.Hex())
+			body, _ := io.ReadAll(resp.Body)
+			if string(body) == "404 page not found" {
+				n.Logger.Error("Invalid reachability check url", "checkUrl", checkUrl.String())
+			} else {
+				n.Logger.Warn("Reachability check operator id not found", "status", resp.StatusCode, "operator_id", n.Config.ID.Hex())
+			}
 			continue
 		} else if resp.StatusCode != 200 {
 			n.Logger.Error("Reachability check request failed", "status", resp.StatusCode)
@@ -506,7 +512,7 @@ func (n *Node) checkNodeReachability() {
 
 		data, err := io.ReadAll(resp.Body)
 		if err != nil {
-			n.Logger.Error("Failed to read reachability check response", err, "data", resp.Body)
+			n.Logger.Error("Failed to read reachability check response", err)
 			continue
 		}