EF Core: change tracking, Migration, N+1, dan AsNoTracking

1. DbContext dan change tracking

EF Core adalah O/RM. Model-nya "made up of entity classes and a context object that represents a session with the database." ^[1] Context itu adalah DbContext, yang di lesson 0004 Anda daftarkan sebagai Scoped karena ia "designed to represent a short-lived unit-of-work." ^[2]

"Each DbContext instance tracks changes made to entities. These tracked entities in turn drive the changes to the database when SaveChanges is called." ^[2] Setiap Entity yang dilacak punya EntityState:

Definisi state dan tabel aksinya: ^[2]

var p = await db.Products.FirstAsync(x => x.Id == 1);  // Unchanged, snapshot disimpan
p.Price = 15000m;                                       // Modified (property Price)
await db.SaveChangesAsync();                            // UPDATE Products SET Price = @p0 WHERE Id = @p1

Tidak ada Update(). EF Core menyimpan snapshot saat Entity dimuat, dan saat SaveChanges "the application's instance is compared with the snapshot to discover the changes to be persisted." ^[3] "EF Core tracks changes at the property level", jadi UPDATE hanya berisi kolom yang berubah. ^[2]

Padanan Node: Prisma tidak punya change tracker. Di sana Anda menulis prisma.product.update({ where, data }) secara eksplisit. Di EF Core, Anda mengubah objek biasa lalu satu SaveChangesAsync menyimpan semua perubahan dalam unit of work itu. Mirip save() di TypeORM / Mongoose, tapi untuk semua Entity sekaligus.

2. Migration (code-first)

"If you want your EF Core model to be the source of truth, use Migrations." Kalau database yang menjadi sumber kebenaran, pakai reverse engineering (scaffold), yang biasa disebut database-first. ^[4] Product Catalog di ROADMAP memakai code-first.

dotnet ef migrations add AddProductSku   # buat file migration di folder Migrations/
dotnet ef database update                # terapkan ke database lokal

• migrations add: "EF Core compares the current model against a snapshot of the old model to determine the differences, and generates migration source files", yang di-commit seperti kode biasa.
• database update: "EF Core records all applied migrations in a special history table", jadi hanya migration baru yang dijalankan. ^[5]
• Production: cara di atas "is ideal for local development, but is less suitable for production environments." ^[5] Pakai SQL script (dotnet ef migrations script --idempotent) yang bisa direview, karena "A migration may drop a column when the intent was to rename it." ^[6]

Padanan Node: prisma migrate dev (lokal) dan prisma migrate deploy (production), atau migration Knex. Folder Migrations/ + snapshot ≈ folder prisma/migrations/.

3. Masalah N+1

Ada tiga cara memuat relasi (navigation property): eager loading ("loaded from the database as part of the initial query"), explicit loading, dan lazy loading ("transparently loaded from the database when the navigation property is accessed"). ^[7] Lazy loading tidak aktif sendiri; ia perlu package Microsoft.EntityFrameworkCore.Proxies dan UseLazyLoadingProxies(). ^[8]

// lazy loading aktif, 50 Category
foreach (var c in await db.Categories.ToListAsync())   // 1 query
    foreach (var p in c.Products)                      // +1 query PER Category
        Console.WriteLine(p.Name);                     // total 1 + 50 = 51 query

"After the initial query loading all the blogs, we then have another query per blog, loading all its posts; this is sometimes called the N+1 problem, and it can cause very significant performance issues." Cara melihatnya: nyalakan statement logging EF Core, lalu hitung SELECT yang berulang. ^[3]

Solusinya: muat semua yang dibutuhkan dalam satu query.

// Eager loading
var cats = await db.Categories.Include(c => c.Products).ToListAsync();

// Lebih baik untuk API: projection langsung ke DTO (hanya kolom yang perlu)
var dto = await db.Categories
    .Select(c => new CategoryResponse(c.Id, c.Name, c.Products.Select(p => p.Name).ToList()))
    .ToListAsync();

Kalau datanya pasti dibutuhkan, "it is always better to use eager loading, so that EF can fetch all the required data in one roundtrip." Projection dengan Select mengirim "only the needed columns." ^[3] Microsoft bahkan menyarankan: "Because lazy loading makes it extremely easy to inadvertently trigger the N+1 problem, it is recommended to avoid it." ^[3]

N+1 tidak hanya dari lazy loading. Menulis await db.Categories.FindAsync(p.CategoryId) di dalam foreach juga N+1, sama seperti await query Prisma di dalam loop. Prisma: "The n+1 problem occurs when looping through query results and performing one additional query per result", dan solusinya juga include. ^[9] Ingat juga lesson 0003: query baru dieksekusi saat ToListAsync (deferred execution), jadi Include/Select harus ditulis sebelum itu.

4. AsNoTracking

Tracking ada harganya: EF Core memelihara dictionary Entity yang dilacak dan menyimpan snapshot setiap Entity. ^[3] Untuk data yang hanya dibaca, biaya itu sia-sia: "No-tracking queries are useful when the results are used in a read-only scenario. They're generally quicker to execute because there's no need to set up the change tracking information." ^[10]

// GET /api/products: hanya dibaca, langsung jadi DTO
var products = await db.Products.AsNoTracking().ToListAsync();

Jebakan interview: Entity dari AsNoTracking lalu diubah dan SaveChanges dipanggil → tidak ada yang disimpan, tanpa error. Projection Select ke DTO juga tidak dilacak, karena "If the result set doesn't contain any entity types, then no tracking is done." ^[10] Padanan Node: Prisma selalu "no-tracking", karena hasilnya objek biasa.

Ringkasan untuk dicetak: Reference: EF Core · Istilah: GLOSSARY.md

Ada yang belum jelas? Tanyakan ke saya (agent) di terminal, misalnya: "kapan Include justru lebih lambat (cartesian explosion)?" atau "apa beda AsNoTracking dan AsNoTrackingWithIdentityResolution?"

Sumber: [1] Overview of Entity Framework Core · [2] Change Tracking in EF Core · [3] Efficient Querying · [4] Managing Database Schemas · [5] Migrations Overview · [6] Applying Migrations · [7] Loading Related Data · [8] Lazy Loading · [9] Prisma: Query optimization · [10] Tracking vs. No-Tracking Queries