fix log and change rpc in response to json rpc error (Layr-Labs#474)

Co-authored-by: Bowen Xue <[email protected]>

common/geth/failover.go

@@ -1,6 +1,7 @@
 package geth
 
 import (
+	"net/url"
 	"sync"
 
 	"github.com/Layr-Labs/eigensdk-go/logging"
@@ -9,27 +10,44 @@ import (
 type FailoverController struct {
 	mu             *sync.RWMutex
 	numberRpcFault uint64
+	UrlDomains     []string
 
 	Logger logging.Logger
 }
 
-func NewFailoverController(logger logging.Logger) *FailoverController {
-	return &FailoverController{
-		Logger: logger.With("component", "FailoverController"),
-		mu:     &sync.RWMutex{},
+func NewFailoverController(logger logging.Logger, rpcUrls []string) (*FailoverController, error) {
+	urlDomains := make([]string, len(rpcUrls))
+	for i := 0; i < len(urlDomains); i++ {
+		url, err := url.Parse(rpcUrls[i])
+		if err != nil {
+			return nil, err
+		}
+		urlDomains[i] = url.Hostname()
 	}
+	return &FailoverController{
+		Logger:     logger.With("component", "FailoverController"),
+		mu:         &sync.RWMutex{},
+		UrlDomains: urlDomains,
+	}, nil
 }
 
 // ProcessError attributes the error and updates total number of fault for RPC
 // It returns if RPC should immediately give up
-func (f *FailoverController) ProcessError(err error) bool {
+func (f *FailoverController) ProcessError(err error, rpcIndex int) bool {
 	f.mu.Lock()
 	defer f.mu.Unlock()
 	if err == nil {
 		return false
 	}
 
-	nextEndpoint, action := f.handleError(err)
+	urlDomain := ""
+	if rpcIndex >= len(f.UrlDomains) || rpcIndex < 0 {
+		f.Logger.Error("[FailoverController]", "err", "rpc index is outside of known url")
+	} else {
+		urlDomain = f.UrlDomains[rpcIndex]
+	}
+
+	nextEndpoint, action := f.handleError(err, urlDomain)
 
 	if nextEndpoint == NewRPC {
 		f.numberRpcFault += 1

common/geth/handle_error.go

@@ -22,10 +22,10 @@ const (
 
 // handleHttpError returns a boolean indicating if the current RPC should be rotated
 // the second boolean indicating if should giveup immediately
-func (f *FailoverController) handleHttpError(httpRespError rpc.HTTPError) (NextEndpoint, ImmediateAction) {
+func (f *FailoverController) handleHttpError(httpRespError rpc.HTTPError, urlDomain string) (NextEndpoint, ImmediateAction) {
 	sc := httpRespError.StatusCode
 	// Default to rotation the current RPC, because it allows a higher chance to get the query completed.
-	f.Logger.Info("[HTTP Response Error]", "Status Code", sc, "Error", httpRespError)
+	f.Logger.Info("[HTTP Response Error]", "urlDomain", urlDomain, "statusCode", sc, "err", httpRespError)
 
 	if sc >= 200 && sc < 300 {
 		// 2xx error, however it should not be reachable
@@ -51,30 +51,28 @@ func (f *FailoverController) handleHttpError(httpRespError rpc.HTTPError) (NextE
 // If the error is http, non2xx error would generate HTTP error, https://github.com/ethereum/go-ethereum/blob/master/rpc/http.go#L233
 // but a 2xx http response could contain JSON RPC error, https://github.com/ethereum/go-ethereum/blob/master/rpc/http.go#L181
 // If the error is Websocket or IPC, we only look for JSON error, https://github.com/ethereum/go-ethereum/blob/master/rpc/json.go#L67
-
-func (f *FailoverController) handleError(err error) (NextEndpoint, ImmediateAction) {
+func (f *FailoverController) handleError(err error, urlDomain string) (NextEndpoint, ImmediateAction) {
 
 	var httpRespError rpc.HTTPError
 	if errors.As(err, &httpRespError) {
 		// if error is http error, i.e. non 2xx error, it is handled here
 		// if it is 2xx error, the error message is nil, https://github.com/ethereum/go-ethereum/blob/master/rpc/http.go,
 		// execution does not enter here.
-		return f.handleHttpError(httpRespError)
+		return f.handleHttpError(httpRespError, urlDomain)
 	} else {
 		// it might be http2xx error, websocket error or ipc error. Parse json error code
 		var rpcError rpc.Error
 		if errors.As(err, &rpcError) {
 			ec := rpcError.ErrorCode()
-			f.Logger.Info("[JSON RPC Response Error]", "Error Code", ec, "Error", rpcError)
-			// we always attribute JSON RPC error as sender's fault, i.e no connection rotation
-			return CurrentRPC, Return
+			f.Logger.Warn("[JSON RPC Response Error]", "urlDomain", urlDomain, "errorCode", ec, "err", rpcError)
+			// we always attribute JSON RPC error as receiver's fault, i.e new connection rotation
+			return NewRPC, Return
 		}
 
 		// If no http response or no rpc response is returned, it is a connection issue,
 		// since we can't accurately attribute the network issue to neither sender nor receiver
 		// side. Optimistically, switch rpc client
-		f.Logger.Info("[Default Response Error]", err)
+		f.Logger.Warn("[Default Response Error]", "urlDomain", urlDomain, "err", err)
 		return NewRPC, Retry
 	}
-
 }