Not found error handling + metrics

disperser/dataapi/metrics.go

@@ -81,6 +81,14 @@ func (g *Metrics) IncrementFailedRequestNum(method string) {
 	}).Inc()
 }
 
+// IncrementNotFoundRequestNum increments the number of not found requests
+func (g *Metrics) IncrementNotFoundRequestNum(method string) {
+	g.NumRequests.With(prometheus.Labels{
+		"status": "not found",
+		"method": method,
+	}).Inc()
+}
+
 // Start starts the metrics server
 func (g *Metrics) Start(ctx context.Context) {
 	g.logger.Info("Starting metrics server at ", "port", g.httpPort)

disperser/dataapi/operator_handlers.go

@@ -2,6 +2,7 @@ package dataapi
 
 import (
 	"context"
+	"errors"
 	"net"
 	"sort"
 	"time"
@@ -106,8 +107,8 @@ func checkIsOnlineAndProcessOperator(operatorStatus OperatorOnlineStatus, operat
 func (s *server) probeOperatorPorts(ctx context.Context, operatorId string) (*OperatorPortCheckResponse, error) {
 	operatorInfo, err := s.subgraphClient.QueryOperatorInfoByOperatorId(context.Background(), operatorId)
 	if err != nil {
-		s.logger.Error("Failed to fetch operator", "error", err)
-		return &OperatorPortCheckResponse{}, err
+		s.logger.Warn("Failed to fetch operator info", "error", err)
+		return &OperatorPortCheckResponse{}, errors.New("not found")
 	}
 
 	retrieverSocket := core.OperatorSocket(operatorInfo.Socket).GetRetrievalSocket()

disperser/dataapi/server.go

@@ -616,10 +616,15 @@ func (s *server) OperatorPortCheck(c *gin.Context) {
 	}))
 	defer timer.ObserveDuration()
 
-	operatorId := c.DefaultQuery("operatorId", "")
+	operatorId := c.DefaultQuery("operator_id", "")
+	s.logger.Info("Checking operator ports", "operatorId", operatorId)
 	portCheckResponse, err := s.probeOperatorPorts(c.Request.Context(), operatorId)
 	if err != nil {
-		s.metrics.IncrementFailedRequestNum("OperatorPortCheck")
+		if err == errNotFound {
+			s.metrics.IncrementNotFoundRequestNum("OperatorPortCheck")
+		} else {
+			s.metrics.IncrementFailedRequestNum("OperatorPortCheck")
+		}
 		errorResponse(c, err)
 		return
 	}