feat(network): lean base pools (#2641)
Some checks are pending
Build / Build (MacOS 15) (push) Waiting to run
Build / Build (MacOS 26) (push) Waiting to run
Build / Build (AlmaLinux 10) (push) Waiting to run
Build / Build (Debian 12) (push) Waiting to run
Build / Build (Debian 13) (push) Waiting to run
Build / Build (Fedora 44) (push) Waiting to run
Build / Build (CentOS 10 Stream) (push) Waiting to run
Build / Build (CentOS 9 Stream) (push) Waiting to run
Build / Build (Ubuntu 26) (push) Waiting to run
Build / Build (Ubuntu 22) (push) Waiting to run
Build / Build (Ubuntu 24) (push) Waiting to run

**Follow-on to #2639 (merged). References a local IORingGroup `1.0.13-preview.11` pack until 1.0.13 (modernuo/IORingGroup#15) is published; do not merge before that switch.**

## Summary

Consumes IORingGroup's lean base pools (modernuo/IORingGroup#15): both network pools now start with one slab, grow a slab at a time with the population, and trim idle slabs back after quiet periods.

- Fixes the transport's send-pool cap: previously only 1024 of the 4096 connections could get a send buffer; connection 1025 was closed at accept.
- Network memory at boot drops from about 96 MB to about 10 MB at the defaults; a full 4096 logged-in connections is about 1.25 GB of base buffers plus the growth budget.
- New settings: `network.initialBufferSlabs` (default 1; slabs of each pool held from boot and the trim floor) and `network.maxBufferSlabs` (default 128; divides the connection maximum into slabs, 32 connections per slab). Both are coerced with a warning; the same value feeds the ring table and the manager so they cannot drift.
- The Debug-only maintenance line includes base-pool capacity and releases.
- `dev-docs/server-requirements.md` rewrites the network memory story and adds the two settings.

## Pre-auth buffers

Every connection starts on the transport's platform-minimum buffers (4 KB receive, 4 KB send) instead of the base pools. It is promoted to full-size buffers (64 KB receive, `network.sendBufferSize` send) when the game server verifies its account — the point where `NetState.Account` is assigned in the `GameServer_AwaitingGameServerLogin` or `GameServer_LoggedIn` state, so a verdict that lands after the parser has moved on still promotes. Nothing ever moves back. The login-server pass stays on the small buffers for its whole lifetime.

Before credentials verify, nothing promotes: the 4 KB send ring is the entire pre-auth send budget, and a connection that overruns it is dropped as exhausted, exactly as the receive side drops a packet header declaring more bytes than the receive buffer can hold (new guard in `HandlePacket`; it also closes the old 65535-byte edge on 64 KB buffers). The stock login sequence sends under 2 KB. After credentials verify, the send path promotes on demand if it ever needs to (unbudgeted, outside the memory ceiling and the shrink bookkeeping — promotion is not growth), and the oversize-packet guard waits on a pending receive promotion or retries a stalled one once for a verified account before disconnecting. A completion that fills the receive buffer arms no receive, so `HandleReceive` now calls `RingSocket.ResumeReceive()` after the parse loop — at 4 KB a burst of small packets fills the buffer in one completion.

Net effect: a flood of unauthenticated connections tops out at about 32 MB across the full 4096-connection cap where the platform minimum is 4 KB (the transport's retained slabs and the base pools used by logged-in players are separate), and never allocates a base-pool slab.

Platform note: on Windows Server 2012 R2 / 2016 the transport's legacy mapping path floors at 64 KB: the pre-auth receive pool is off there (its base is 64 KB), while the pre-auth send buffer starts at 64 KB under the 256 KB base. The server logs the effective sizes at startup.

## Testing

Server.Tests (905) and UOContent.Tests green on the preview pack (one pre-existing `FamiliarAITests` failure from #2644 reproduces on `main`, tracked separately). New tests cover both coercions, that the ring's registration table equals `RequiredRegisteredBuffers` for the configured values, promotion on game-server auth and on a late account, no promotion on the login server, pre-credential overrun ending in exhaustion, post-credential on-demand promotion, the oversize-packet guard through loopback (error, wait, retry with an account, and a promotion made pending mid-parse), and receiving again after a burst fills the initial buffer.
This commit is contained in:
Kamron Batman 2026-09-20 00:24:15 -07:00 committed by GitHub
parent 9d9e672a09
commit 24bcfee554
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
6 changed files with 682 additions and 22 deletions

View file

@ -0,0 +1,236 @@
using System;
using System.Buffers;
using System.Diagnostics;
using System.Network;
using System.Threading;
using Server.Network;
using Xunit;
namespace Server.Tests.Network;
[Collection("Sequential Server Tests")]
public class NetStateRecvGuardTests
{
private const byte TestPacketId = 0xB1;
private static unsafe void RegisterVariableLength()
{
if (IncomingPackets.GetHandler(TestPacketId) == null)
{
IncomingPackets.Register(TestPacketId, 0, false, &NoOp);
}
}
private static void NoOp(NetState state, SpanReader reader)
{
}
// Not 0x73: NetStateSendBufferTests registers its own no-op there first-come-wins, since the
// handler table is process-global, and this test needs its own handler's side effect to fire.
private const byte TestPingPacketId = 0x72;
private static byte _lastPing;
private static unsafe void RegisterNoOpPing()
{
if (IncomingPackets.GetHandler(TestPingPacketId) == null)
{
IncomingPackets.Register(TestPingPacketId, 2, false, &RecordPing);
}
}
private static void RecordPing(NetState state, SpanReader reader) => _lastPing = reader.ReadByte();
private const byte AttachAccountPacketId = 0x71;
private static unsafe void RegisterAttachAccount()
{
if (IncomingPackets.GetHandler(AttachAccountPacketId) == null)
{
IncomingPackets.Register(AttachAccountPacketId, 3, false, &AttachAccount);
}
}
// Stands in for the 0x91 handler: the account attaches mid-parse, so the promotion is pending
// with a receive armed when the next packet in the same completion reaches the guard
private static void AttachAccount(NetState state, SpanReader reader) => state.Account = new MockAccount();
private static void SliceUntil(Func<bool> done)
{
var deadline = Stopwatch.StartNew();
while (!done() && deadline.ElapsedMilliseconds < 5000)
{
NetState.Slice();
Thread.Sleep(5);
}
Assert.True(done());
}
[SkippableFact]
public void HandleReceive_PacketLongerThanTheRecvBuffer_IsAnError()
{
RegisterVariableLength();
var ns = PacketTestUtilities.CreateTestNetState(out var client);
try
{
// A 16-bit length cannot exceed a 64 KiB buffer, so the guard is unreachable there and
// this only ever hit the `< 3` check on a buffer that size.
Skip.If(ns._socket.RecvBuffer.PhysicalSize > ushort.MaxValue);
ns._protocolState = NetState.ProtocolState.GameServer_LoggedIn;
// A variable-length header claiming more than the buffer can ever hold would otherwise
// park the connection with nothing to arm a recv into
var declared = ns._socket.RecvBuffer.PhysicalSize; // one more than the buffer can ever hold
client.Send(new byte[] { TestPacketId, (byte)(declared >> 8), (byte)declared });
SliceUntil(() => ns._protocolState == NetState.ProtocolState.Error);
Assert.Contains("bad state", ns._disconnectReason);
}
finally
{
ns.Dispose();
client.Close();
}
}
[Fact]
public void HandleReceive_PacketThatFits_WaitsForTheRest()
{
RegisterVariableLength();
var ns = PacketTestUtilities.CreateTestNetState(out var client);
try
{
ns._protocolState = NetState.ProtocolState.GameServer_LoggedIn;
var declared = ns._socket.RecvBuffer.PhysicalSize / 2;
client.Send(new byte[] { TestPacketId, (byte)(declared >> 8), (byte)declared });
SliceUntil(() => ns._socket.RecvBuffer.ReadableBytes == 3);
Assert.Equal(NetState.ProtocolState.GameServer_LoggedIn, ns._protocolState);
Assert.True(ns.Running);
Assert.Equal(3, ns._socket.RecvBuffer.ReadableBytes);
}
finally
{
ns.Dispose();
client.Close();
}
}
[SkippableFact]
public void HandleReceive_OversizePacket_WithAnAccount_RetriesPromotionInsteadOfErroring()
{
Skip.If(NetState.InitialRecvBufferSize == 0);
RegisterVariableLength();
var ns = PacketTestUtilities.CreateTestNetState(out var client);
try
{
// _protocolState is still AwaitingSeed here, so the Account setter's gate does not fire
ns.Account = new MockAccount();
Assert.Equal(NetState.InitialRecvBufferSize, ns._socket.RecvBuffer.PhysicalSize);
ns._protocolState = NetState.ProtocolState.GameServer_LoggedIn;
var declared = NetState.InitialRecvBufferSize; // as much as the initial buffer can ever hold
client.Send(new byte[] { TestPacketId, (byte)(declared >> 8), (byte)declared });
SliceUntil(() => ns._socket.RecvBuffer.ReadableBytes == 3);
Assert.Equal(NetState.ProtocolState.GameServer_LoggedIn, ns._protocolState);
Assert.True(ns.Running);
// The deferred promotion applies at this next completion; the header is then delivered
// whole to the 0xB1 no-op handler
client.Send(new byte[declared - 3]);
SliceUntil(
() => ns._socket.RecvBuffer.ReadableBytes == 0 &&
ns._socket.RecvBuffer.PhysicalSize == NetState.RecvBufferSize
);
Assert.True(ns.Running);
Assert.Equal(NetState.ProtocolState.GameServer_LoggedIn, ns._protocolState);
}
finally
{
ns.Dispose();
client.Close();
}
}
[SkippableFact]
public void HandleReceive_AfterABurstFillsTheInitialBuffer_KeepsReceiving()
{
Skip.If(NetState.InitialRecvBufferSize == 0);
RegisterNoOpPing();
var ns = PacketTestUtilities.CreateTestNetState(out var client);
try
{
ns._protocolState = NetState.ProtocolState.GameServer_LoggedIn;
// One segment of pings fills the 4 KiB buffer (capacity is PhysicalSize - 1, odd, so the
// last ping straddles two completions); the parser consumes them all
var capacity = ns._socket.RecvBuffer.PhysicalSize - 1;
var pings = new byte[capacity + 1];
for (var i = 0; i < pings.Length; i += 2)
{
pings[i] = TestPingPacketId;
pings[i + 1] = (byte)(i / 2);
}
client.Send(pings);
SliceUntil(() => ns._socket.RecvBuffer.ReadableBytes == 0 && ns._receivedData);
Assert.True(ns.Running);
// Without a re-arm this ping never arrives
client.Send(new byte[] { TestPingPacketId, 0xEE });
SliceUntil(() => _lastPing == 0xEE);
Assert.True(ns.Running);
}
finally
{
ns.Dispose();
client.Close();
}
}
[SkippableFact]
public void HandleReceive_AccountAttachedMidParse_ThenOversizeHeader_WaitsForThePendingPromotion()
{
Skip.If(NetState.InitialRecvBufferSize == 0);
RegisterAttachAccount();
RegisterVariableLength();
var ns = PacketTestUtilities.CreateTestNetState(out var client);
try
{
ns._protocolState = NetState.ProtocolState.GameServer_LoggedIn;
var declared = ns._socket.RecvBuffer.PhysicalSize;
// One send, one completion on loopback: the account packet promotes (deferred: a receive is
// armed), then the oversize header reaches the guard with that promotion pending. If the OS
// ever splits this into two completions instead, the first assertion block still holds — the
// header then hits the guard after the promotion already applied and simply waits on the
// 64 KiB buffer, so the test cannot flake, it just exercises the weaker path on that run.
client.Send(new byte[] { AttachAccountPacketId, 0x00, 0x00, TestPacketId, (byte)(declared >> 8), (byte)declared });
SliceUntil(() => ns.Account != null);
Assert.True(ns.Running);
Assert.Equal(NetState.ProtocolState.GameServer_LoggedIn, ns._protocolState);
Assert.Equal(3, ns._socket.RecvBuffer.ReadableBytes); // the header waits, not rejected
// The next completion applies the promotion and the whole packet is delivered
client.Send(new byte[declared - 3]);
SliceUntil(
() => ns._socket.RecvBuffer.ReadableBytes == 0 &&
ns._socket.RecvBuffer.PhysicalSize == NetState.RecvBufferSize
);
Assert.True(ns.Running);
}
finally
{
ns.Dispose();
client.Close();
}
}
}

View file

@ -1,4 +1,5 @@
using System;
using System.Buffers;
using System.Collections.Generic;
using System.Diagnostics;
using System.Net.Sockets;
@ -12,9 +13,11 @@ namespace Server.Tests.Network;
[Collection("Sequential Server Tests")]
public class NetStateSendBufferTests
{
// Authentication is the 0x91 credentials check on the game server; that is where buffers promote
private static NetState CreateAuthenticatedNetState(out Socket client)
{
var ns = PacketTestUtilities.CreateTestNetState(out client);
ns._protocolState = NetState.ProtocolState.GameServer_AwaitingGameServerLogin;
ns.Account = new MockAccount();
return ns;
}
@ -26,6 +29,183 @@ public class NetStateSendBufferTests
return data;
}
private static unsafe void RegisterNoOpPing()
{
if (IncomingPackets.GetHandler(0x73) == null)
{
IncomingPackets.Register(0x73, 2, false, &NoOp);
}
}
private static void NoOp(NetState state, SpanReader reader)
{
}
private static void SliceUntil(Func<bool> done)
{
var deadline = Stopwatch.StartNew();
while (!done() && deadline.ElapsedMilliseconds < 5000)
{
NetState.Slice();
Thread.Sleep(5);
}
Assert.True(done());
}
[SkippableFact]
public void Account_InTheGameServerState_PromotesBothBuffers()
{
Skip.If(NetState.InitialSendBufferSize == 0);
RegisterNoOpPing();
var ns = PacketTestUtilities.CreateTestNetState(out var client);
try
{
Assert.Equal(NetState.InitialSendBufferSize, ns._socket.SendBuffer.PhysicalSize);
Assert.Equal(NetState.InitialRecvBufferSize, ns._socket.RecvBuffer.PhysicalSize);
ns._protocolState = NetState.ProtocolState.GameServer_AwaitingGameServerLogin;
ns.Account = new MockAccount();
// Send promotes at once; nothing was in flight so no buffer retires
Assert.Equal(NetState.SendBufferSize, ns._socket.SendBuffer.PhysicalSize);
Assert.False(ns._sendBufferGrown);
// Recv promotes at the next completion
ns._protocolState = NetState.ProtocolState.GameServer_LoggedIn;
client.Send(new byte[] { 0x73, 0x01 });
SliceUntil(() => ns._socket.RecvBuffer.PhysicalSize == NetState.RecvBufferSize);
Assert.True(ns.Running);
}
finally
{
ns.Dispose();
client.Close();
}
}
[SkippableFact]
public void Account_OnTheLoginServer_KeepsTheInitialBuffers()
{
Skip.If(NetState.InitialSendBufferSize == 0);
var ns = PacketTestUtilities.CreateTestNetState(out var client);
try
{
ns._protocolState = NetState.ProtocolState.LoginServer_AwaitingLogin;
ns.Account = new MockAccount();
Assert.Equal(NetState.InitialSendBufferSize, ns._socket.SendBuffer.PhysicalSize);
Assert.Equal(NetState.InitialRecvBufferSize, ns._socket.RecvBuffer.PhysicalSize);
}
finally
{
ns.Dispose();
client.Close();
}
}
[SkippableFact]
public void Send_BeyondTheInitialBuffer_BeforeCredentials_IsExhausted()
{
// Nothing promotes before credentials verify: the 4 KiB ring is the whole pre-auth budget
Skip.If(NetState.InitialSendBufferSize == 0);
var ns = PacketTestUtilities.CreateTestNetState(out var client);
try
{
var data = Pattern(NetState.InitialSendBufferSize + 512, 7);
ns.Send(data);
Assert.Contains("exhausted", ns._disconnectReason, StringComparison.OrdinalIgnoreCase);
Assert.Equal(NetState.InitialSendBufferSize, ns._socket.SendBuffer.PhysicalSize);
// A graceful server-side close half-closes and waits for the peer's own FIN; closing the
// peer here stands in for the client eventually going away, so Running can be observed.
// Once it completes the NetState is disposed and _socket goes null, so nothing below
// this point may touch it.
client.Close();
SliceUntil(() => !ns.Running);
}
finally
{
ns.Dispose();
client.Close();
}
}
[SkippableFact]
public void Send_BeyondTheInitialBuffer_WithCredentials_PromotesOnDemand()
{
// The late-verdict path from Account_AssignedAfterTheStateFlipped_StillPromotes: an account
// attached while the setter's own gate does not fire must still let the send path promote
Skip.If(NetState.InitialSendBufferSize == 0);
var ns = PacketTestUtilities.CreateTestNetState(out var client);
try
{
ns.Account = new MockAccount(); // _protocolState is still AwaitingSeed: the setter does not promote
var data = Pattern(NetState.InitialSendBufferSize + 512, 7);
ns.Send(data);
Assert.True(ns.Running);
Assert.Equal(string.Empty, ns._disconnectReason);
Assert.Equal(NetState.SendBufferSize, ns._socket.SendBuffer.PhysicalSize);
Assert.False(ns._sendBufferGrown); // promotion is not growth: nothing to shrink later
Assert.Equal(data, ReadAll(client, data.Length));
}
finally
{
ns.Dispose();
client.Close();
}
}
[SkippableFact]
public void Send_BacklogBeforeCredentials_IsExhaustedInsteadOfPromoted()
{
Skip.If(NetState.InitialSendBufferSize == 0);
var ns = PacketTestUtilities.CreateTestNetState(out var client);
try
{
// No Slice() between sends: nothing drains, so a second write finds a non-empty initial buffer
var half = Pattern(NetState.InitialSendBufferSize / 2, 11);
ns.Send(half);
ns.Send(half);
ns.Send(half); // cannot fit, buffer not empty, no account: refused
Assert.Equal(NetState.InitialSendBufferSize, ns._socket.SendBuffer.PhysicalSize);
Assert.Contains("exhausted", ns._disconnectReason, StringComparison.OrdinalIgnoreCase);
// A graceful server-side close waits for the queued bytes to drain and then for the
// peer's own FIN; closing the peer here stands in for the client eventually going away.
// Once it completes the NetState is disposed and _socket goes null, so nothing below
// this point may touch it.
client.Close();
SliceUntil(() => !ns.Running);
}
finally
{
ns.Dispose();
client.Close();
}
}
[SkippableFact]
public void Account_AssignedAfterTheStateFlipped_StillPromotes()
{
Skip.If(NetState.InitialSendBufferSize == 0);
var ns = PacketTestUtilities.CreateTestNetState(out var client);
try
{
ns._protocolState = NetState.ProtocolState.GameServer_LoggedIn;
ns.Account = new MockAccount();
Assert.Equal(NetState.SendBufferSize, ns._socket.SendBuffer.PhysicalSize);
}
finally
{
ns.Dispose();
client.Close();
}
}
// A random prefix lands under the target; zeros (2 bits each) walk it up a byte at a time
private static byte[] CompressesToExactly(int compressedLength, int seed)
{
@ -448,4 +628,58 @@ public class NetStateSendBufferTests
// growth off; budget untouched
Assert.Equal(1L, NetState.CoerceSendBufferGrowthBudget(1, sendBufferSize, sendBufferSize));
}
[Fact]
public void CoerceMaxBufferSlabs_ClampsToOneSlabPerConnection()
{
var maxConnections = NetState.SocketManager.MaxSockets;
// fewer than one slab is meaningless
Assert.Equal(1, NetState.CoerceMaxBufferSlabs(0));
Assert.Equal(1, NetState.CoerceMaxBufferSlabs(-4));
// more slabs than connections cannot make a slab any smaller
Assert.Equal(maxConnections, NetState.CoerceMaxBufferSlabs(maxConnections * 2));
Assert.Equal(128, NetState.CoerceMaxBufferSlabs(128));
}
[Fact]
public void CoerceInitialBufferSlabs_ClampsToTheSlabCount()
{
var slabs = NetState.BasePoolSlabCount(128);
Assert.True(slabs > 1);
Assert.Equal(1, NetState.CoerceInitialBufferSlabs(0, slabs));
Assert.Equal(1, NetState.CoerceInitialBufferSlabs(-1, slabs));
// a pool never holds more slabs than it has
Assert.Equal(slabs, NetState.CoerceInitialBufferSlabs(slabs + 1, slabs));
Assert.Equal(4, NetState.CoerceInitialBufferSlabs(4, slabs));
}
[Fact]
public void Ring_RegisteredBufferTable_MatchesTheConfiguredSlabCount()
{
var manager = NetState.SocketManager;
var maxBufferSlabs = NetState.CoerceMaxBufferSlabs(ServerConfiguration.GetSetting("network.maxBufferSlabs", NetState.DefaultMaxBufferSlabs));
// A table smaller than this throws at manager construction. Connections start on the
// transport minimum until the game server promotes them, so the formula must count those too.
// Mirrors what production passes: the request, not the effective size the manager may coerce
// down to (RequiredRegisteredBuffers treats a non-zero request as a pool either way).
Assert.Equal(
RingSocketManager.RequiredRegisteredBuffers(
manager.MaxSockets,
NetState.SendBufferSize,
manager.MaxSendBufferSize,
manager.SendBufferGrowthBudget,
maxBufferSlabs,
IORingBuffer.MinimumSize,
IORingBuffer.MinimumSize
),
NetState.Ring.MaxRegisteredBuffers
);
}
}