feat(live-update): never move properties; mark old instances outdated

Property instances are left out of the replace plan. Once the database is
up to date every instance running at the start is marked outdated, and a
zone that already got a new instance is not started again.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Aaron Kimbrell
2026-09-30 04:03:19 -05:00
parent 8f891229e6
commit 891f21daa9
3 changed files with 103 additions and 15 deletions

View File

@@ -4,6 +4,7 @@
#include <ctime>
#include <filesystem>
#include <map>
#include <set>
#include "BinaryPathFinder.h"
#include "CDActivitiesTable.h"
@@ -31,6 +32,10 @@ namespace {
uint32_t g_UpdateId = 0;
uint32_t g_NextMigrationId = 0;
Clock::time_point g_NextTick{};
// The instances running when the update started: marked outdated once the database is up to date
std::set<std::pair<uint32_t, uint32_t>> g_OldInstances;
bool g_OldMarked = false;
void MarkOldInstances();
std::map<size_t, std::pair<eUnitState, std::string>> g_Logged; // unit index -> what was logged last
ePhase g_LoggedPhase = ePhase::IDLE;
@@ -237,7 +242,13 @@ bool LiveUpdateCoordinator::Start(const std::string& by, LWOOBJID requesterId, i
}
if (!CheckBinaries(error)) return false;
const auto settings = ReadSettings(warnSeconds);
if (!g_Machine.Start(settings, Observe(), Clock::now(), error)) return false;
const auto observed = Observe();
if (!g_Machine.Start(settings, observed, Clock::now(), error)) return false;
g_OldInstances.clear();
for (const auto& world : observed.worlds) {
if (!world.view.shuttingDown) g_OldInstances.insert({ world.view.zoneId, world.view.instanceId });
}
g_OldMarked = false;
g_UpdateId++;
g_Logged.clear();
g_LoggedPhase = ePhase::IDLE;
@@ -295,6 +306,22 @@ void LiveUpdateCoordinator::Update() {
const auto phaseBefore = g_Machine.Phase();
const auto finishedBefore = FinishedUnits();
if (g_Machine.Tick(Observe(), now, g_Actions)) Publish(phaseBefore != g_Machine.Phase() || finishedBefore != FinishedUnits());
MarkOldInstances();
}
namespace {
void MarkOldInstances() {
if (g_OldMarked || g_Machine.Units().empty()) return;
const auto database = g_Machine.Units().front().state;
if (database != eUnitState::STOPPED && database != eUnitState::SKIPPED) return;
g_OldMarked = true;
// Nobody new goes to an instance on the old binaries: requests start new ones. Properties aren't moved; they stop
// once everyone left (OutdatedInstances.h)
const auto marked = Game::im->MarkOutdated([](const Instance& instance) {
return g_OldInstances.contains({ instance.GetMapID(), instance.GetInstanceID() });
});
LOG("Live update %u: %u instance(s) on the old build take nobody new; properties among them stop once empty", g_UpdateId, marked);
}
}
void LiveUpdateCoordinator::Abort(const std::string& why) {

View File

@@ -121,7 +121,7 @@ namespace LiveUpdate {
enum class eWorldPlan : uint8_t {
STOP, // nobody there: shut it down (a new one starts when someone goes there)
REPLACE_EMPTY, // nobody there, but the zone should always have one: start the new one, then stop this one
MOVE, // replace it and move its players (properties save and freeze first)
MOVE, // replace it and move its players (properties are never planned: InReplacePlan)
WAIT_THEN_MOVE, // activity zones and character selection: nobody new goes there; its players leave by
// themselves or, after a while, are moved
};
@@ -146,6 +146,16 @@ namespace LiveUpdate {
return eWorldPlan::MOVE;
}
/**
* Which running instances a live update replaces: all but those already shutting down and properties. A property
* (clone) is never moved: building in progress there isn't saved. It is marked outdated like every other instance
* (master: nobody new goes there) and stops by itself once everyone left (OutdatedInstances.h).
*/
inline bool InReplacePlan(const InstanceMigration::InstanceView& view) {
return !view.shuttingDown && view.cloneId == 0;
}
struct Unit {
eUnitKind kind{};
eUnitState state{ eUnitState::PENDING };
@@ -189,7 +199,7 @@ namespace LiveUpdate {
// Character selection first (logins go to the new one soonest), then the busiest
std::vector<WorldView> worlds;
for (const auto& world : observed.worlds) {
if (!world.view.shuttingDown) worlds.push_back(world);
if (InReplacePlan(world.view)) worlds.push_back(world);
}
std::stable_sort(worlds.begin(), worlds.end(), [](const WorldView& a, const WorldView& b) {
if ((a.view.zoneId == 0) != (b.view.zoneId == 0)) return a.view.zoneId == 0;
@@ -487,10 +497,24 @@ namespace LiveUpdate {
SetState(unit, eUnitState::STOPPING, message);
}
// Whether zone has a public instance on the new build already: one started after the update began (not in the
// plan), running and taking players
bool HasNewInstance(const Observed& observed, uint32_t zone) const {
return std::any_of(observed.worlds.begin(), observed.worlds.end(), [&](const WorldView& world) {
const auto& view = world.view;
if (view.zoneId != zone || view.cloneId != 0 || view.isPrivate || view.outdated || view.shuttingDown || view.draining) return false;
return std::none_of(m_Units.begin(), m_Units.end(), [&](const Unit& unit) {
return unit.kind == eUnitKind::WORLD && unit.zone == view.zoneId && unit.instance == view.instanceId;
});
});
}
void StartWorld(Unit& unit, const Observed& observed, Clock::time_point now, IActions& actions) {
const auto* world = observed.Find(unit.zone, unit.instance);
if (!world || world->view.shuttingDown) return SetState(unit, eUnitState::STOPPED, "It stopped by itself");
unit.players = static_cast<uint32_t>(std::max(world->view.players, 0));
// A zone somebody went to since the update began has its new instance already
if (HasNewInstance(observed, unit.zone)) m_ReplacedZones.insert(unit.zone);
unit.plan = PlanWorld(world->view, world->activityZone, m_Settings.keepZones.contains(unit.zone), m_ReplacedZones.contains(unit.zone));
switch (unit.plan) {
case eWorldPlan::STOP:

View File

@@ -1,5 +1,6 @@
#include <gtest/gtest.h>
#include <algorithm>
#include <map>
#include "LiveUpdateMachine.h"
@@ -203,6 +204,16 @@ TEST(LiveUpdateTest, RoutingSkipsDrainingAndFullInstances) {
world.shuttingDown = false;
world.isPrivate = true;
EXPECT_FALSE(AcceptsNewPlayers(world, 1100, 0, true));
world.isPrivate = false;
// On the old binary or old files: nobody new, friends and properties included
world.outdated = true;
EXPECT_FALSE(AcceptsNewPlayers(world, 1100, 0, false));
EXPECT_FALSE(AcceptsNewPlayers(world, 1100, 0, true));
auto property = World(1150, 3, 1, 42);
EXPECT_TRUE(AcceptsNewPlayers(property, 1150, 42, false));
property.outdated = true;
EXPECT_FALSE(AcceptsNewPlayers(property, 1150, 42, false));
EXPECT_FALSE(AcceptsNewPlayers(property, 1150, 42, true));
// A replacement still starting takes players: they wait for it
auto starting = World(1100, 2, 0);
starting.ready = false;
@@ -325,20 +336,46 @@ TEST(LiveUpdateTest, OldInstanceThatNeverStopsFails) {
EXPECT_EQ(h.WorldUnit(1100, 1).state, eUnitState::FAILED);
}
TEST(LiveUpdateTest, PropertiesAreSavedFirst) {
TEST(LiveUpdateTest, PropertiesAreNeverInThePlan) {
EXPECT_FALSE(InReplacePlan(World(1150, 1, 2, 42)));
EXPECT_FALSE(InReplacePlan(World(1150, 2, 0, 43)));
EXPECT_TRUE(InReplacePlan(World(1100, 3, 2)));
auto stopping = World(1100, 4, 0);
stopping.shuttingDown = true;
EXPECT_FALSE(InReplacePlan(stopping));
Harness h;
h.master.Add(World(1150, 1, 2, 42));
h.master.Add(World(1150, 1, 2, 42)); // busy property
h.master.Add(World(1150, 2, 0, 43)); // empty property
h.master.Add(World(1100, 3, 2));
h.Start();
size_t worlds = 0;
for (const auto& unit : h.machine.Units()) {
if (unit.kind != eUnitKind::WORLD) continue;
worlds++;
EXPECT_EQ(unit.clone, 0u);
}
EXPECT_EQ(worlds, 1u);
h.ServicesComeBack();
ASSERT_EQ(h.master.moves.size(), 1u);
EXPECT_TRUE(h.master.moves[0].prepare);
EXPECT_EQ(h.master.moves[0].prepareWaitSeconds, h.settings.propertyBuildWaitSeconds);
const auto& unit = h.WorldUnit(1150, 1);
EXPECT_EQ(unit.state, eUnitState::PREPARING);
h.machine.OnMigration(Status(1, InstanceMigration::eState::PREPARED));
EXPECT_EQ(unit.state, eUnitState::STARTING);
h.machine.OnMigration(Status(1, InstanceMigration::eState::FAILED));
EXPECT_EQ(unit.state, eUnitState::FAILED);
EXPECT_EQ(h.master.moves[0].zone, 1100u);
EXPECT_FALSE(h.master.moves[0].prepare);
EXPECT_TRUE(h.master.log.end() == std::find_if(h.master.log.begin(), h.master.log.end(), [](const std::string& line) {
return line.find("1150") != std::string::npos;
})) << "nothing is done to a property";
}
TEST(LiveUpdateTest, ZoneWithANewInstanceAlreadyIsNotStartedAgain) {
Harness h;
h.master.Add(World(1000, 1, 0));
h.Start();
// Someone went to the zone after the update began: master started a new instance for them
h.master.Add(World(1000, 50, 1));
h.ServicesComeBack();
h.Tick();
EXPECT_EQ(h.WorldUnit(1000, 1).plan, eWorldPlan::STOP);
EXPECT_TRUE(h.master.Logged("stop world 1000/1"));
EXPECT_TRUE(std::none_of(h.master.log.begin(), h.master.log.end(), [](const std::string& line) { return line.rfind("start world 1000/", 0) == 0; }));
}
TEST(LiveUpdateTest, EmptyInstancesStopOrAreReplaced) {
@@ -602,7 +639,7 @@ TEST(LiveUpdateTest, ShuttingDownInstancesAreLeftOut) {
TEST(LiveUpdateTest, StatusCarriesEveryUnit) {
Harness h;
h.master.Add(World(1150, 4, 2, 42));
h.master.Add(World(1150, 4, 2));
h.Start();
LiveUpdateStatus status;
h.machine.FillStatus(status);
@@ -624,7 +661,7 @@ TEST(LiveUpdateTest, StatusCarriesEveryUnit) {
EXPECT_EQ(read.units[4].kind, eUnitKind::WORLD);
EXPECT_EQ(read.units[4].zoneId, 1150u);
EXPECT_EQ(read.units[4].instanceId, 4u);
EXPECT_EQ(read.units[4].cloneId, 42u);
EXPECT_EQ(read.units[4].cloneId, 0u);
EXPECT_EQ(read.units[4].players, 2u);
EXPECT_EQ(read.units[5].kind, eUnitKind::DASHBOARD);
}