From 49cf6c14f3126f0415897b2ace7fe59659fce391 Mon Sep 17 00:00:00 2001
From: Yehonal <147092+Yehonal@users.noreply.github.com>
Date: Tue, 14 Jul 2026 23:43:38 +0200
Subject: [PATCH] fix(apps/startup-scripts): guard against duplicate/too-early
 server starts (#26149)

Co-authored-by: YehonalBot <yehonalbot@users.noreply.github.com>
---
 apps/startup-scripts/src/starter              | 172 +++++++++++++++++-
 .../test/test_startup_scripts.bats            |  80 ++++++++
 2 files changed, 251 insertions(+), 1 deletion(-)

diff --git a/apps/startup-scripts/src/starter b/apps/startup-scripts/src/starter
index 4e58e8e7b..5585e7eba 100755
--- a/apps/startup-scripts/src/starter
+++ b/apps/startup-scripts/src/starter
@@ -45,6 +45,174 @@ if [ ! -f "$BINARY" ]; then
     exit 1
 fi
 
+function config_value() {
+    local key="$1"
+
+    if [ -z "${CONFIG_ABS:-}" ] || [ ! -f "$CONFIG_ABS" ]; then
+        return 1
+    fi
+
+    awk -F= -v key="$key" '
+        {
+            configKey = $1
+            gsub(/^[[:space:]]+|[[:space:]]+$/, "", configKey)
+        }
+        configKey == key {
+            value = $0
+            sub(/^[^=]*=/, "", value)
+            sub(/[[:space:]]*[#;].*$/, "", value)
+            gsub(/"/, "", value)
+            gsub(/^[[:space:]]+|[[:space:]]+$/, "", value)
+            print value
+            exit
+        }
+    ' "$CONFIG_ABS"
+}
+
+function command_contains_config() {
+    local command_line="$1"
+
+    if [ -z "${CONFIG_ABS:-}" ]; then
+        return 0
+    fi
+
+    case "$command_line" in
+        *" -c $CONFIG_ABS"*|*" -c \"$CONFIG_ABS\""*)
+            return 0
+            ;;
+    esac
+
+    return 1
+}
+
+function find_existing_instances() {
+    local line pid command_line
+
+    while IFS= read -r line; do
+        pid="${line%% *}"
+        command_line="${line#* }"
+
+        [ -z "$pid" ] && continue
+        [ "$pid" = "$$" ] && continue
+        [ "$pid" = "$PPID" ] && continue
+
+        case "$command_line" in
+            *"/starter "*|*"pgrep "*)
+                continue
+                ;;
+        esac
+
+        if command_contains_config "$command_line"; then
+            printf '%s\n' "$line"
+        fi
+    done < <(pgrep -fa "$EXECPATH" 2>/dev/null || true)
+}
+
+function listening_on_port() {
+    local port="$1"
+
+    command -v ss >/dev/null 2>&1 || return 1
+    ss -H -ltn "sport = :$port" 2>/dev/null | grep -q .
+}
+
+function guarded_ports() {
+    local port soap_enabled soap_port
+
+    if [ "$BINFILE" != "worldserver" ]; then
+        return 0
+    fi
+
+    port="$(config_value "WorldServerPort" || true)"
+    if [ -n "$port" ]; then
+        printf '%s\n' "$port"
+    fi
+
+    soap_enabled="$(config_value "SOAP.Enabled" || true)"
+    soap_port="$(config_value "SOAP.Port" || true)"
+    if [ "$soap_enabled" = "1" ] && [ -n "$soap_port" ]; then
+        printf '%s\n' "$soap_port"
+    fi
+}
+
+function find_listening_guard_ports() {
+    local port
+
+    while IFS= read -r port; do
+        [ -z "$port" ] && continue
+        if listening_on_port "$port"; then
+            printf '%s\n' "$port"
+        fi
+    done < <(guarded_ports)
+}
+
+function wait_for_startup_guard() {
+    local timeout="${AC_STARTUP_GUARD_TIMEOUT:-90}"
+    local interval="${AC_STARTUP_GUARD_INTERVAL:-2}"
+    local elapsed=0
+    local instances ports
+    local sleep_for
+    local warned=0
+
+    if [ "${AC_STARTUP_SKIP_GUARD:-0}" = "1" ]; then
+        return 0
+    fi
+
+    case "$timeout" in
+        ''|*[!0-9]*)
+            echo "Startup guard: invalid AC_STARTUP_GUARD_TIMEOUT '$timeout', using 90."
+            timeout=90
+            ;;
+    esac
+
+    case "$interval" in
+        ''|*[!0-9]*)
+            echo "Startup guard: invalid AC_STARTUP_GUARD_INTERVAL '$interval', using 2."
+            interval=2
+            ;;
+    esac
+
+    if [ "$interval" -lt 1 ]; then
+        echo "Startup guard: AC_STARTUP_GUARD_INTERVAL must be at least 1, using 1."
+        interval=1
+    fi
+
+    while true; do
+        instances="$(find_existing_instances)"
+        ports="$(find_listening_guard_ports)"
+
+        if [ -z "$instances" ] && [ -z "$ports" ]; then
+            return 0
+        fi
+
+        if [ "$elapsed" -ge "$timeout" ]; then
+            echo "Startup guard: refusing to start duplicate $BINFILE after waiting ${timeout}s."
+            if [ -n "$instances" ]; then
+                echo "Startup guard: existing matching process(es):"
+                echo "$instances"
+            fi
+            if [ -n "$ports" ]; then
+                echo "Startup guard: configured port(s) still listening: $(echo "$ports" | paste -sd ',' -)"
+            fi
+            exit 0
+        fi
+
+        if [ "$warned" -eq 0 ]; then
+            echo "Startup guard: waiting for existing $BINFILE instance or configured port(s) to clear..."
+            warned=1
+        fi
+
+        sleep_for="$interval"
+        if [ $((elapsed + sleep_for)) -gt "$timeout" ]; then
+            sleep_for=$((timeout - elapsed))
+        fi
+
+        [ "$sleep_for" -le 0 ] && continue
+
+        sleep "$sleep_for"
+        elapsed=$((elapsed + sleep_for))
+    done
+}
+
 # Create crashes directory if it doesn't exist
 mkdir -p "$CRASHES_PATH"
 
@@ -55,6 +223,8 @@ cd "$BINPATH" || {
 
 EXECPATH=$(realpath "$BINFILE")
 
+wait_for_startup_guard
+
 if [ "$GDB_ENABLED" -eq 1 ]; then
     echo "Starting $EXECPATH with GDB enabled"
 
@@ -148,4 +318,4 @@ else
             exec "$EXECPATH" ${CONFIG_ABS:+-c "$CONFIG_ABS"}
         fi
     fi
-fi
\ No newline at end of file
+fi
diff --git a/apps/startup-scripts/test/test_startup_scripts.bats b/apps/startup-scripts/test/test_startup_scripts.bats
index c7a90c7f6..82d49eba0 100755
--- a/apps/startup-scripts/test/test_startup_scripts.bats
+++ b/apps/startup-scripts/test/test_startup_scripts.bats
@@ -50,6 +50,86 @@ teardown() {
     [[ "$output" =~ "Test server starting" ]]
 }
 
+@test "starter: should refuse duplicate binary/config start" {
+    cat > "$TEST_DIR/bin/duplicate-server" << 'EOF'
+#!/usr/bin/env bash
+sleep 30
+EOF
+    chmod +x "$TEST_DIR/bin/duplicate-server"
+
+    "$TEST_DIR/bin/duplicate-server" -c "$TEST_DIR/test-server.conf" &
+    local duplicate_pid=$!
+
+    AC_STARTUP_GUARD_TIMEOUT=1 AC_STARTUP_GUARD_INTERVAL=1 run timeout 5s "$SCRIPT_DIR/starter" "$TEST_DIR/bin" "duplicate-server" "" "$TEST_DIR/test-server.conf" "" "" 0
+    kill "$duplicate_pid" 2>/dev/null || true
+    wait "$duplicate_pid" 2>/dev/null || true
+
+    debug_on_failure
+    [ "$status" -eq 0 ]
+    [[ "$output" =~ "Startup guard: refusing to start duplicate duplicate-server" ]]
+    [[ "$output" =~ "existing matching process" ]]
+}
+
+@test "starter: should refuse occupied configured worldserver ports" {
+    if ! command -v nc >/dev/null 2>&1; then
+        skip "nc is required for this test"
+    fi
+
+    cat > "$TEST_DIR/bin/worldserver" << 'EOF'
+#!/usr/bin/env bash
+echo "worldserver should not start"
+EOF
+    chmod +x "$TEST_DIR/bin/worldserver"
+
+    local listen_port=40123
+    cat > "$TEST_DIR/worldserver.conf" << EOF
+WorldServerPort = $listen_port
+SOAP.Enabled = 0
+EOF
+    nc -l 127.0.0.1 "$listen_port" >/dev/null &
+    local listener_pid=$!
+
+    AC_STARTUP_GUARD_TIMEOUT=1 AC_STARTUP_GUARD_INTERVAL=1 run timeout 5s "$SCRIPT_DIR/starter" "$TEST_DIR/bin" "worldserver" "" "$TEST_DIR/worldserver.conf" "" "" 0
+    kill "$listener_pid" 2>/dev/null || true
+    wait "$listener_pid" 2>/dev/null || true
+
+    debug_on_failure
+    [ "$status" -eq 0 ]
+    [[ "$output" =~ "Startup guard: refusing to start duplicate worldserver" ]]
+    [[ "$output" =~ "configured port(s) still listening: $listen_port" ]]
+}
+
+@test "starter: should parse config keys literally for startup guard" {
+    if ! command -v nc >/dev/null 2>&1; then
+        skip "nc is required for this test"
+    fi
+
+    cat > "$TEST_DIR/bin/worldserver" << 'EOF'
+#!/usr/bin/env bash
+echo "worldserver started"
+EOF
+    chmod +x "$TEST_DIR/bin/worldserver"
+
+    local listen_port=40124
+    cat > "$TEST_DIR/worldserver.conf" << EOF
+WorldServerPort = 40125
+SOAPXEnabled = 1
+SOAPXPort = $listen_port
+SOAP.Enabled = 0
+EOF
+    nc -l 127.0.0.1 "$listen_port" >/dev/null &
+    local listener_pid=$!
+
+    AC_STARTUP_GUARD_TIMEOUT=1 AC_STARTUP_GUARD_INTERVAL=1 run timeout 5s "$SCRIPT_DIR/starter" "$TEST_DIR/bin" "worldserver" "" "$TEST_DIR/worldserver.conf" "" "" 0
+    kill "$listener_pid" 2>/dev/null || true
+    wait "$listener_pid" 2>/dev/null || true
+
+    debug_on_failure
+    [ "$status" -eq 0 ]
+    [[ "$output" =~ "worldserver started" ]]
+    [[ ! "$output" =~ "Startup guard: refusing" ]]
+}
+
 # ===== SIMPLE RESTARTER TESTS =====
 
 @test "simple-restarter: should fail with missing parameters" {