111 lines
4.5 KiB
C#
111 lines
4.5 KiB
C#
// Copyright (c) Microsoft. All rights reserved.
|
|
|
|
using Microsoft.Data.SqlClient;
|
|
using Microsoft.Extensions.VectorData;
|
|
using SqlServer.ConformanceTests.Support;
|
|
using VectorData.ConformanceTests;
|
|
using VectorData.ConformanceTests.Support;
|
|
using VectorData.ConformanceTests.Xunit;
|
|
using Xunit;
|
|
|
|
namespace SqlServer.ConformanceTests;
|
|
|
|
public class SqlServerIndexKindTests(SqlServerIndexKindTests.Fixture fixture)
|
|
: IndexKindTests<int>(fixture), IClassFixture<SqlServerIndexKindTests.Fixture>
|
|
{
|
|
// Latest version vector indexes are only available in Azure SQL, not in on-prem SQL Server.
|
|
// They also require at least 100 rows before the vector index can be created,
|
|
// so we override the test to insert data first, then create the index.
|
|
[ConditionalFact]
|
|
[AzureSqlRequired]
|
|
public virtual async Task DiskAnn()
|
|
{
|
|
const string CollectionName = "IndexKindTests_DiskAnn";
|
|
|
|
// Step 1: Create the table using Flat index (no vector index) so we can insert data.
|
|
VectorStoreCollectionDefinition flatDefinition = new()
|
|
{
|
|
Properties =
|
|
[
|
|
new VectorStoreKeyProperty(nameof(SearchRecord.Key), typeof(int)),
|
|
new VectorStoreDataProperty(nameof(SearchRecord.Int), typeof(int)),
|
|
new VectorStoreVectorProperty(nameof(SearchRecord.Vector), typeof(ReadOnlyMemory<float>), dimensions: 3)
|
|
{
|
|
IndexKind = IndexKind.Flat,
|
|
DistanceFunction = DistanceFunction.CosineDistance
|
|
}
|
|
]
|
|
};
|
|
|
|
using var flatCollection = fixture.TestStore.CreateCollection<int, SearchRecord>(CollectionName, flatDefinition);
|
|
await flatCollection.EnsureCollectionDeletedAsync();
|
|
await flatCollection.EnsureCollectionExistsAsync();
|
|
|
|
try
|
|
{
|
|
// Step 2: Insert the 3 test rows + 97 filler rows to meet the 100-row minimum.
|
|
SearchRecord[] testRecords =
|
|
[
|
|
new() { Key = 1, Int = 1, Vector = new([1, 2, 3]) },
|
|
new() { Key = 2, Int = 2, Vector = new([10, 30, 50]) },
|
|
new() { Key = 3, Int = 3, Vector = new([100, 40, 70]) }
|
|
];
|
|
|
|
await flatCollection.UpsertAsync(testRecords);
|
|
|
|
var fillerRecords = Enumerable.Range(100, 97)
|
|
.Select(i => new SearchRecord
|
|
{
|
|
Key = i,
|
|
Int = i,
|
|
Vector = new([i * 0.1f, i * 0.2f, i * 0.3f])
|
|
})
|
|
.ToArray();
|
|
|
|
await flatCollection.UpsertAsync(fillerRecords);
|
|
|
|
// Step 3: Create the DiskANN vector index via raw SQL now that data is in the table.
|
|
using var connection = new SqlConnection(SqlServerTestStore.Instance.ConnectionString);
|
|
await connection.OpenAsync();
|
|
|
|
using (var createIndex = new SqlCommand(
|
|
$"CREATE VECTOR INDEX index_{CollectionName}_Vector ON [{CollectionName}]([Vector]) WITH (METRIC = 'COSINE', TYPE = 'DISKANN');",
|
|
connection))
|
|
{
|
|
await createIndex.ExecuteNonQueryAsync();
|
|
}
|
|
|
|
// Step 4: Create a new collection instance with DiskAnn to route searches through VECTOR_SEARCH().
|
|
VectorStoreCollectionDefinition diskAnnDefinition = new()
|
|
{
|
|
Properties =
|
|
[
|
|
new VectorStoreKeyProperty(nameof(SearchRecord.Key), typeof(int)),
|
|
new VectorStoreDataProperty(nameof(SearchRecord.Int), typeof(int)),
|
|
new VectorStoreVectorProperty(nameof(SearchRecord.Vector), typeof(ReadOnlyMemory<float>), dimensions: 3)
|
|
{
|
|
IndexKind = IndexKind.DiskAnn,
|
|
DistanceFunction = DistanceFunction.CosineDistance
|
|
}
|
|
]
|
|
};
|
|
|
|
using var diskAnnCollection = fixture.TestStore.CreateCollection<int, SearchRecord>(CollectionName, diskAnnDefinition);
|
|
|
|
var result = await diskAnnCollection.SearchAsync(new ReadOnlyMemory<float>([10, 30, 50]), top: 1).SingleAsync();
|
|
|
|
Assert.NotNull(result);
|
|
Assert.Equal(2, result.Record.Int);
|
|
}
|
|
finally
|
|
{
|
|
await flatCollection.EnsureCollectionDeletedAsync();
|
|
}
|
|
}
|
|
|
|
public new class Fixture() : IndexKindTests<int>.Fixture
|
|
{
|
|
public override TestStore TestStore => SqlServerTestStore.Instance;
|
|
}
|
|
}
|