From 5bcf03ec7fcc412c2259f341f6c7f3f86631466b Mon Sep 17 00:00:00 2001
From: Stefan Boberg <stefan.boberg@epicgames.com>
Date: Thu, 9 Apr 2026 13:26:08 +0200
Subject: Fix ZenServerState stale entry detection on PID reuse (k8s) (#932)

- Detect stale shared-memory entries whose PID matches the current process but predate our registration (m_OurEntry == nullptr)
- Sweep() now reclaims such entries instead of skipping them
- Lookup() and LookupByEffectivePort() skip stale same-PID entries
- Fixes startup failure on k8s where PID 1 is always reused after an unclean shutdown
---
 src/zenutil/zenserverprocess.cpp | 31 +++++++++++++++++++++++++++++++
 1 file changed, 31 insertions(+)

(limited to 'src/zenutil/zenserverprocess.cpp')

diff --git a/src/zenutil/zenserverprocess.cpp b/src/zenutil/zenserverprocess.cpp
index 2b27b2d8b..9a282a848 100644
--- a/src/zenutil/zenserverprocess.cpp
+++ b/src/zenutil/zenserverprocess.cpp
@@ -267,6 +267,8 @@ ZenServerState::InitializeReadOnly()
 ZenServerState::ZenServerEntry*
 ZenServerState::Lookup(int DesiredListenPort) const
 {
+	const uint32_t OurPid = GetCurrentProcessId();
+
 	for (int i = 0; i < m_MaxEntryCount; ++i)
 	{
 		uint16_t EntryPort = m_Data[i].DesiredListenPort;
@@ -274,6 +276,14 @@ ZenServerState::Lookup(int DesiredListenPort) const
 		{
 			if (DesiredListenPort == 0 || (EntryPort == DesiredListenPort))
 			{
+				// If the entry's PID matches our own but we haven't registered yet,
+				// this is a stale entry from a previous process incarnation (e.g. PID 1
+				// reuse after unclean shutdown in k8s). Skip it.
+				if (m_Data[i].Pid == OurPid && m_OurEntry == nullptr)
+				{
+					continue;
+				}
+
 				std::error_code _;
 				if (IsProcessRunning(m_Data[i].Pid, _))
 				{
@@ -289,6 +299,8 @@ ZenServerState::Lookup(int DesiredListenPort) const
 ZenServerState::ZenServerEntry*
 ZenServerState::LookupByEffectivePort(int Port) const
 {
+	const uint32_t OurPid = GetCurrentProcessId();
+
 	for (int i = 0; i < m_MaxEntryCount; ++i)
 	{
 		uint16_t EntryPort = m_Data[i].EffectiveListenPort;
@@ -296,6 +308,11 @@ ZenServerState::LookupByEffectivePort(int Port) const
 		{
 			if (EntryPort == Port)
 			{
+				if (m_Data[i].Pid == OurPid && m_OurEntry == nullptr)
+				{
+					continue;
+				}
+
 				std::error_code _;
 				if (IsProcessRunning(m_Data[i].Pid, _))
 				{
@@ -358,12 +375,26 @@ ZenServerState::Sweep()
 
 	ZEN_ASSERT(m_IsReadOnly == false);
 
+	const uint32_t OurPid = GetCurrentProcessId();
+
 	for (int i = 0; i < m_MaxEntryCount; ++i)
 	{
 		ZenServerEntry& Entry = m_Data[i];
 
 		if (Entry.DesiredListenPort)
 		{
+			// If the entry's PID matches our own but we haven't registered yet,
+			// this is a stale entry from a previous process incarnation (e.g. PID 1
+			// reuse after unclean shutdown in k8s). Reclaim it.
+			if (Entry.Pid == OurPid && m_OurEntry == nullptr)
+			{
+				ZEN_CONSOLE_DEBUG("Sweep - pid {} matches current process but no registration yet, reclaiming stale entry (port {})",
+								  Entry.Pid.load(),
+								  Entry.DesiredListenPort.load());
+				Entry.Reset();
+				continue;
+			}
+
 			std::error_code ErrorCode;
 			if (Entry.Pid != 0 && IsProcessRunning(Entry.Pid, ErrorCode) == false)
 			{
-- 
cgit v1.2.3