Tayra

Appearance

Querying with Blind Indexes ​

A blind index is only useful if you can efficiently query by it. This page covers how to compute a blind index for a search term and use it in EF Core and Marten queries.

Computing a Blind Index ​

Use ITayra.ComputeBlindIndexAsync() to compute a blind index hash for a search term:

By property expression (recommended):

cs
// Compute a blind index hash using a strongly-typed property expression
string emailHash = await biTayra.ComputeBlindIndexAsync(
    "jane@example.com", "EmailHash", typeof(BlindIndexedCustomer));
Console.WriteLine($"  Typed hash: {emailHash}");
anchor

By hash name directly:

cs
// Compute a blind index hash by index name and entity type
string hash = await biTayra.ComputeBlindIndexAsync(
    "jane@example.com", "EmailHash", typeof(BlindIndexedCustomer));
Console.WriteLine($"  Raw hash: {hash}");
anchor

Both methods apply the configured transforms in order before computing the HMAC. The resulting hash string is URL-safe Base64.

EF Core ​

Save ​

When using the EF Core interceptor, the blind index is computed automatically during SaveChanges. You do not need to call EncryptAsync manually if transparent encryption is enabled.

cs
// With transparent encryption enabled, blind indexes are computed automatically
// during SaveChanges — no manual call to EncryptAsync is required.
//
// var customer = new ProtectedCustomer
// {
//     Id = Guid.NewGuid(),
//     Name = "Jane Doe",
//     Email = "jane@example.com",
// };
// dbContext.Customers.Add(customer);
// await dbContext.SaveChangesAsync();
// // customer.EmailHash is now populated with the HMAC
Console.WriteLine("  Blind index is computed automatically during SaveChanges.");
anchor

Query by Blind Index and Decrypt ​

cs
// Compute the hash for the search term
string searchHash = await blindIndexer.ComputeHashAsync(
    "jane@example.com", "EmailHash", typeof(ProtectedCustomer));

// Query by the companion column
// var customer = await dbContext.Customers
//     .Where(c => c.EmailHash == searchHash)
//     .FirstOrDefaultAsync();

// The interceptor decrypts automatically on materialization
// Console.WriteLine($"Found: {customer?.Name}");
Console.WriteLine($"  EF Core search hash: {searchHash}");
anchor

Use AsNoTracking for read-only queries

When you only need to read and display data, AsNoTracking() avoids the overhead of change tracking and is safe to combine with Tayra's materialization interceptor.

Add a Database Index ​

Blind index queries are only efficient if the database has an index on the companion column. Add one in your migration or OnModelCreating:

cs
// In your DbContext.OnModelCreating, add a database index on the companion column
// so blind-index queries are efficient (avoids full table scans):
modelBuilder.Entity<ProtectedCustomer>()
    .HasIndex(c => c.EmailHash);
anchor

Marten ​

Save ​

Marten stores documents as JSONB. The companion property is included in the document JSON and can be queried with LINQ.

cs
// EncryptAsync computes the blind index before encryption.
// The EmailHash companion property is populated automatically.

// await tayra.EncryptAsync(customer);

// Store the document — EmailHash is included in the JSONB.
// await using var session = store.LightweightSession();
// session.Store(customer);
// await session.SaveChangesAsync();
Console.WriteLine("\nBlind index save: EmailHash is populated before the document is stored.");
anchor

Query by Companion Property ​

cs
// Compute the search hash for the lookup term (same transforms applied as during save)
// var blindIndexer = provider.GetRequiredService<IBlindIndexer>();
// string searchHash = await blindIndexer.ComputeHashAsync(
//     "jane@example.com", "EmailHash", typeof(MartenCustomer));

// Query by the companion property in the JSONB document
// await using var querySession = store.QuerySession();
// var customer = await querySession.Query<MartenCustomer>()
//     .Where(c => c.EmailHash == searchHash)
//     .FirstOrDefaultAsync();

// Decrypt the loaded document to restore plaintext PII
// await tayra.DecryptAsync(customer);
Console.WriteLine("Blind index query: filter by EmailHash, then decrypt on load.");
anchor

Create a Marten index on the companion property

Use Marten's Index<T>() API in StoreOptions to add a GIN or B-tree index on the companion JSONB field for efficient querying.

Compound Blind Index Queries ​

When you need to search by multiple encrypted fields simultaneously, compute a hash for each and combine them in a single query:

cs
// For compound indexes, the hash is computed automatically by EncryptAsync
// when you call it on the entity — all fields are combined into one HMAC.
// Retrieve the stored hash from the entity to build your query:

// var entity = new IndexedCustomer { FirstName = "Jane", LastName = "Doe", ... };
// await tayra.EncryptAsync(entity);
// string fullNameHash = entity.FullNameIndex;

// Query the companion column:
// db.Query<IndexedCustomer>().Where(c => c.FullNameIndex == fullNameHash)
Console.WriteLine("  Compound blind index hash computed during EncryptAsync.");
anchor

Compound indexes on multiple companion columns

If you frequently query by two companion columns together, consider creating a composite database index on both columns (FirstNameHash, LastNameHash) to avoid index intersection overhead.

Paginated Lookups ​

Blind indexes support exact-match lookups only. You cannot perform prefix searches, range queries, or LIKE comparisons on a blind index hash. For paginated result sets, use the blind index to identify matching rows, then apply additional ordering:

cs
// Blind indexes support exact-match lookups only.
// Use the hash to filter, then apply ordering and paging.
string pagedEmailHash = await biTayra.ComputeBlindIndexAsync(
    "jane@example.com", "EmailHash", typeof(BlindIndexedCustomer));

// var results = await dbContext.Customers
//     .Where(c => c.EmailHash == pagedEmailHash)
//     .OrderBy(c => c.Id)
//     .Skip(page * pageSize)
//     .Take(pageSize)
//     .ToListAsync();
Console.WriteLine($"  Paginated hash computed: {pagedEmailHash}");
anchor

Performance Notes ​

  • HMAC computation is fast. HMAC-SHA256 on a short string is sub-microsecond on modern hardware. It does not meaningfully impact write throughput.
  • The HMAC key is cached. After the first call, the key is held in the DefaultCryptoEngine memory cache. Subsequent calls to ComputeBlindIndexAsync are memory-only operations.
  • Add a database index. Without an index on the companion column, every query becomes a full table scan. This is by far the biggest performance factor.
  • Companion columns add storage. A SHA-256 hash stored as URL-safe Base64 is 43 characters. Budget accordingly for large tables.

See Also ​