perf: Establish BulkCopyAsync performance baseline

CmdScale.EntityFrameworkCore.TimescaleDB.Benchmarks/CmdScale.EntityFrameworkCore.TimescaleDB.Benchmarks.csproj

@@ -13,6 +13,8 @@
 
   <ItemGroup>
     <PackageReference Include="BenchmarkDotNet" Version="0.15.2" />
+    <PackageReference Include="Testcontainers.PostgreSql" Version="4.7.0" />
+    <PackageReference Include="Z.EntityFramework.Extensions.EFCore" Version="9.103.9.3" />
   </ItemGroup>
 
   <ItemGroup>

CmdScale.EntityFrameworkCore.TimescaleDB.Benchmarks/README.md

@@ -7,59 +7,9 @@ This project uses **BenchmarkDotNet** to measure the performance of high-through
 ## Prerequisites
 
 - .NET 8 SDK or later  
-- Docker and Docker Compose
+- Docker
 
----
-
-## Step 1: Start a Clean Database
-
-The benchmarks require a running TimescaleDB instance. A `docker-compose.yml` file is provided in the project root to simplify this process.
-
-### 🔄 Stop and Reset (if needed)
-
-To ensure you start with a clean slate, run this command to stop any running containers and permanently delete all existing data.
-
-```bash
-docker compose down -v
-```
-
-### ▶️ Start the Database
-
-Launch a new TimescaleDB instance in the background.
-
-```bash
-docker compose up -d
-```
-
-### ✅ Verify Connection String
-
-Ensure the connection string in `BulkCopyToAsyncBenchmarks.cs` matches the settings in your `docker-compose.yml` file (the default should work).
-
----
-
-## Step 2: Apply Migrations
-
-Next, create the necessary tables in the new database using EF Core migrations.
-
-### ➕ Add a Migration
-
-Create a new migration. Run this from the root of the solution.
-
-```bash
-dotnet ef migrations add <YourMigrationName> --project CmdScale.EntityFrameworkCore.TimescaleDB.Example.DataAccess --startup-project CmdScale.EntityFrameworkCore.TimescaleDB.Example
-```
-
-### ⬆️ Update the Database
-
-Apply the migrations to create the hypertable schema.
-
-```bash
-dotnet ef database update --project CmdScale.EntityFrameworkCore.TimescaleDB.Example.DataAccess --startup-project CmdScale.EntityFrameworkCore.TimescaleDB.Example
-```
-
----
-
-## Step 3: Run the Benchmarks
+## Run the Benchmarks
 
 Once the database is set up, you can run the performance tests.
 

CmdScale.EntityFrameworkCore.TimescaleDB.Benchmarks/WriteRecordsBenchmarkBase.cs

@@ -0,0 +1,68 @@
+using BenchmarkDotNet.Attributes;
+using CmdScale.EntityFrameworkCore.TimescaleDB.Example.DataAccess;
+using Microsoft.EntityFrameworkCore;
+using Testcontainers.PostgreSql;
+
+namespace CmdScale.EntityFrameworkCore.TimescaleDB.Benchmarks
+{
+    public abstract class WriteRecordsBenchmarkBase<T> where T : class
+    {
+        public int NumberOfRecords;
+        public int MaxBatchSize;
+        public int NumberOfWorkers;
+
+        private readonly PostgreSqlContainer _dbContainer = new PostgreSqlBuilder()
+            .WithImage("timescale/timescaledb:latest-pg17")
+            .WithDatabase("benchmark_tests_db")
+            .WithUsername("test_user")
+            .WithPassword("test_password")
+            .Build();
+
+        protected string ConnectionString = "";
+        protected readonly List<T> Trades = [];
+        protected TimescaleContext? Context;
+
+        [GlobalSetup]
+        public async Task Setup()
+        {
+            await _dbContainer.StartAsync();
+            ConnectionString = _dbContainer.GetConnectionString();
+
+            DbContextOptionsBuilder<TimescaleContext> optionsBuilder = new();
+            optionsBuilder.UseNpgsql(ConnectionString).UseTimescaleDb();
+            Context = new TimescaleContext(optionsBuilder.Options);
+
+            await Context.Database.MigrateAsync();
+            Console.WriteLine("Migration applied successfully.");
+        }
+
+        [GlobalCleanup]
+        public async Task GlobalCleanup()
+        {
+            await _dbContainer.DisposeAsync();
+        }
+
+        [IterationSetup]
+        public void IterationSetup()
+        {
+            Trades.Clear();
+            var random = new Random();
+            string[] tickers = ["AAPL", "GOOGL", "MSFT", "TSLA", "AMZN"];
+            var baseTimestamp = DateTime.UtcNow.AddMinutes(-30);
+
+            for (int i = 0; i < NumberOfRecords; i++)
+            {
+                var trade = CreateTradeInstance(i, baseTimestamp, tickers[random.Next(tickers.Length)], random);
+                Trades.Add(trade);
+            }
+
+            // Truncate the table before each iteration for a clean slate
+            string tableName = GetTableName();
+            string sql = $"TRUNCATE TABLE \"{tableName}\"";
+            Context!.Database.ExecuteSqlRaw(sql);
+        }
+
+        protected abstract T CreateTradeInstance(int index, DateTime baseTimestamp, string ticker, Random random);
+        protected abstract string GetTableName();
+    }
+}

CmdScale.EntityFrameworkCore.TimescaleDB.Benchmarks/WriteRecordsKeylessBenchmarks.cs

@@ -0,0 +1,110 @@
+using BenchmarkDotNet.Attributes;
+using CmdScale.EntityFrameworkCore.TimescaleDB.Abstractions;
+using CmdScale.EntityFrameworkCore.TimescaleDB.Example.DataAccess.Models;
+using Npgsql;
+using NpgsqlTypes;
+
+namespace CmdScale.EntityFrameworkCore.TimescaleDB.Benchmarks
+{
+    [Config(typeof(InProcessConfig))]
+    [MemoryDiagnoser]
+    [ThreadingDiagnoser]
+    public class WriteRecordsKeylessBenchmarks : WriteRecordsBenchmarkBase<Trade>
+    {
+        [Params(100_000, 500_000)]
+        public new int NumberOfRecords;
+
+        [Params(25_000, 50_000, 100_000)]
+        public new int MaxBatchSize;
+
+        [Params(8)]
+        public new int NumberOfWorkers;
+
+        private readonly List<Task> tasks = [];
+        private int totalRecords;
+        private int workerChunkSize;
+
+        protected override Trade CreateTradeInstance(int index, DateTime baseTimestamp, string ticker, Random random)
+        {
+            return new Trade
+            {
+                Timestamp = baseTimestamp.AddMicroseconds(index),
+                Ticker = ticker,
+                Price = (decimal)(100 + random.NextDouble() * 400),
+                Size = random.Next(1, 100)
+            };
+        }
+
+        protected override string GetTableName() => "Trades";
+
+        [Benchmark]
+        public async Task BulkCopyAsync()
+        {
+            TimescaleCopyConfig<Trade> config = new TimescaleCopyConfig<Trade>()
+                .ToTable(GetTableName())
+                .WithWorkers(NumberOfWorkers)
+                .WithBatchSize(MaxBatchSize);
+
+            await Trades.BulkCopyAsync(ConnectionString, config);
+        }
+
+        [Benchmark]
+        public async Task HardcodedBulkCopyAsync()
+        {
+            totalRecords = Trades.Count;
+            workerChunkSize = (int)Math.Ceiling((double)totalRecords / NumberOfWorkers);
+
+            for (int i = 0; i < NumberOfWorkers; i++)
+            {
+                int startIndex = i * workerChunkSize;
+                int currentWorkerDataSize = Math.Min(workerChunkSize, totalRecords - startIndex);
+
+                if (currentWorkerDataSize <= 0)
+                {
+                    break;
+                }
+
+                List<Trade> workerData = [.. Trades.Skip(startIndex).Take(currentWorkerDataSize)];
+                tasks.Add(Task.Run(async () =>
+                {
+                    // Open new connection to DB
+                    using NpgsqlConnection connection = new(ConnectionString);
+                    await connection.OpenAsync();
+
+                    // Command to copy data in a binary format from a client-application 
+                    string copyCommand = "COPY \"Trades\" (\"Time\", \"Value\", \"SegmentId\", \"SignalId\") FROM STDIN (FORMAT BINARY)";
+
+                    for (int j = 0; j < workerData.Count; j += MaxBatchSize)
+                    {
+                        List<Trade> currentBatch = [.. workerData.Skip(j).Take(MaxBatchSize)];
+
+                        // Start a binary import stream
+                        await using NpgsqlBinaryImporter writer = connection.BeginBinaryImport(copyCommand);
+                        foreach (Trade item in currentBatch)
+                        {
+                            // IMPORTANT: Columns must be inserted in the exact order
+                            await writer.WriteAsync(item.Timestamp, NpgsqlDbType.TimestampTz);
+                            await writer.WriteAsync(item.Ticker, NpgsqlDbType.Text);
+                            await writer.WriteAsync(item.Price, NpgsqlDbType.Numeric);
+                            await writer.WriteAsync(item.Size, NpgsqlDbType.Integer);
+                            await writer.WriteAsync(item.Exchange, NpgsqlDbType.Text);
+                        }
+
+                        await writer.CompleteAsync();
+                    }
+                }));
+            }
+
+            await Task.WhenAll(tasks);
+        }
+
+        [Benchmark]
+        public async Task BatchedBulkInsertOptimizedAsync()
+        {
+            foreach (Trade[] chunk in Trades.Chunk(MaxBatchSize))
+            {
+                await Context!.BulkInsertOptimizedAsync(chunk.ToList());
+            }
+        }
+    }
+}