From 78da15566351c20b57eff50907844607a06d9a10 Mon Sep 17 00:00:00 2001
From: Roland <33993199+rolznz@users.noreply.github.com>
Date: Thu, 13 Nov 2025 19:27:52 +0700
Subject: [PATCH] Fix: pool reconnect (#200)

* fix: bump ping timeout for poor connections

* fix: always close relay on ping failure

* fix: add multiple ping attempts before closing relay connection

* fix: only 3 ping attempts

* fix: pool check for relay context done

* fix: include relay URL in debug log

* fix: do not return after closing relay on max ping attempts

* fix: compile error
---
 connection.go |  2 +-
 relay.go      | 22 +++++++++++++++++-----
 2 files changed, 18 insertions(+), 6 deletions(-)

diff --git a/connection.go b/connection.go
index f6857b2..56ca19d 100644
--- a/connection.go
+++ b/connection.go
@@ -59,7 +59,7 @@ func (c *Connection) Close() error {
 
 // Ping sends a ping message to the websocket connection.
 func (c *Connection) Ping(ctx context.Context) error {
-	ctx, cancel := context.WithTimeoutCause(ctx, time.Millisecond*800, errors.New("ping took too long"))
+	ctx, cancel := context.WithTimeoutCause(ctx, time.Millisecond*10000, errors.New("ping took too long"))
 	defer cancel()
 	return c.conn.Ping(ctx)
 }
diff --git a/relay.go b/relay.go
index f3d1320..58ecd4c 100644
--- a/relay.go
+++ b/relay.go
@@ -9,7 +9,6 @@ import (
 	"log"
 	"net/http"
 	"strconv"
-	"strings"
 	"sync"
 	"sync/atomic"
 	"time"
@@ -168,6 +167,7 @@ func (r *Relay) ConnectWithTLS(ctx context.Context, tlsConfig *tls.Config) error
 
 	// queue all write operations here so we don't do mutex spaghetti
 	go func() {
+		pingAttempt := 0
 		for {
 			select {
 			case <-r.connectionContext.Done():
@@ -180,12 +180,24 @@ func (r *Relay) ConnectWithTLS(ctx context.Context, tlsConfig *tls.Config) error
 				return
 
 			case <-ticker.C:
+				debugLogf("{%s} pinging relay", r.URL)
 				err := r.Connection.Ping(r.connectionContext)
-				if err != nil && !strings.Contains(err.Error(), "failed to wait for pong") {
-					InfoLogger.Printf("{%s} error writing ping: %v; closing websocket", r.URL, err)
-					r.Close() // this should trigger a context cancelation
-					return
+				if err != nil {
+					pingAttempt++
+					debugLogf("{%s} error writing ping (attempt %d): %v", r.URL, pingAttempt, err)
+
+					if pingAttempt >= 3 {
+						debugLogf("{%s} error writing ping after multiple attempts; closing websocket", r.URL)
+						err = r.Close() // this should trigger a context cancelation
+						if err != nil {
+							debugLogf("{%s} failed to close relay: %v", r.URL, err)
+						}
+					}
+					continue
 				}
+				// ping was OK
+				debugLogf("{%s} ping OK", r.URL)
+				pingAttempt = 0
 
 			case writeRequest := <-r.writeQueue:
 				// all write requests will go through this to prevent races