Skalowanie z Redis Backplane

Problem wielu serwerów

Czym jest backplane?

Dodawanie backplane Redis

// dotnet add package Microsoft.AspNetCore.SignalR.StackExchangeRedis

builder.Services.AddSignalR()
    .AddStackExchangeRedis("localhost:6379", options =>
    {
        options.Configuration.ChannelPrefix =
            RedisChannel.Literal("myapp"); // namespace your channels
    });

// Or from config:
// .AddStackExchangeRedis(builder.Configuration.GetConnectionString("Redis")!);

Jak działa Redis Pub/Sub w SignalR

// Server A: client sends message
// Hub on Server A calls:
await Clients.Group("room-42").ReceiveMessage(user, msg);

// Internally SignalR publishes to Redis:
// PUBLISH signalr/myapp/group/room-42 <serialized message>

// Redis delivers to Server B and C
// Both servers find connections in "room-42" and push to them

Opcje konfiguracji

builder.Services.AddSignalR()
    .AddStackExchangeRedis(opts =>
    {
        opts.Configuration = ConfigurationOptions.Parse(
            builder.Configuration["Redis:ConnectionString"]!);
        opts.Configuration.Password      = builder.Configuration["Redis:Password"];
        opts.Configuration.Ssl           = true;
        opts.Configuration.AbortOnConnectFail = false;
        opts.Configuration.ConnectRetry  = 3;
    });

Azure SignalR Service

// dotnet add package Microsoft.Azure.SignalR

builder.Services.AddSignalR()
    .AddAzureSignalR(builder.Configuration["Azure:SignalR:ConnectionString"]!);

// That's it — Azure manages all connections and backplane
// Your server scales to zero when not needed

Sesje sticky (rozwiązanie awaryjne dla transportu innego niż WebSocket)

# Nginx: sticky session config (IP hash)
upstream signalr_servers {
    ip_hash; # ensures same client hits same server
    server server1:5000;
    server server2:5000;
    server server3:5000;
}

# Or use cookie-based sticky sessions:
# sticky cookie srv_id expires=1h;

# WebSockets don't need sticky sessions —
# the connection is long-lived on one server

Grupy na wielu serwerach

// This works correctly across servers:
public async Task JoinRoom(string room)
{
    // Adds to Redis-backed group
    await Groups.AddToGroupAsync(Context.ConnectionId, room);

    // Clients.Group sends via Redis to all servers
    await Clients.Group(room).UserJoined(Context.User!.Identity!.Name!);
}
// No code changes needed — the Redis backplane handles distribution

Monitorowanie Redis Pub/Sub

# redis-cli: monitor all SignalR channels
redis-cli PSUBSCRIBE "myapp*"

# Check active channel subscribers
redis-cli PUBSUB CHANNELS "myapp*"

# Check number of subscribers per channel
redis-cli PUBSUB NUMSUB "myapp/all"

Obsługa awarii Redis

builder.Services.AddSignalR()
    .AddStackExchangeRedis(opts =>
    {
        opts.Configuration = ConfigurationOptions.Parse(redisConn);
        opts.Configuration.AbortOnConnectFail = false; // don't crash app
        opts.Configuration.ConnectRetry       = 5;
        opts.Configuration.ReconnectRetryPolicy =
            new ExponentialRetry(5000, maxDeltaBackoffMilliseconds: 60000);
    });

Praktyczny przykład: skalowalna architektura czatu

// Architecture:
// [Browser] -> [Load Balancer (any server, WebSocket)] ->
//   [SignalR Server 1] -+
//   [SignalR Server 2] -+-> [Redis Pub/Sub] -> all servers
//   [SignalR Server 3] -+

// Server code is identical on all instances:
builder.Services.AddSignalR().AddStackExchangeRedis(redisConn);
app.MapHub<ChatHub>("/hubs/chat");
// Scale replicas: kubectl scale deployment chat --replicas=3

Szybkie sprawdzenie

Podsumowanie: skalowanie za pomocą backplane Redis