Khi nào dùng `bulk_create`, `bulk_update`, `QuerySet.update()` thay vì loop `save()`?

Loop instance.save() cho 10k row = 10k INSERT round-trip → tốn vài chục giây và ép DB ngộp connection. Bulk operations gom thành 1-N câu SQL, nhanh hơn 10–100 lần.

Model.objects.bulk_create(objs, batch_size=1000) chạy 1 INSERT cho mỗi batch, có ignore_conflicts=True (skip duplicate) và update_conflicts=True (upsert kiểu Postgres). Model.objects.bulk_update(objs, fields, batch_size=1000) chạy 1 UPDATE ... CASE WHEN cho mỗi batch — object phải có sẵn pk. QuerySet.update(**kwargs) chạy 1 UPDATE SQL cho mọi row match filter, không cần load object về Python.

# Sai: 10k query
for row in rows:
    Item(sku=row.sku, price=row.price).save()

# Đúng: 10 query (batch_size=1000)
Item.objects.bulk_create(
    [Item(sku=r.sku, price=r.price) for r in rows],
    batch_size=1000,
)

# Cập nhật cùng giá trị, không cần load object
Item.objects.filter(category='archived').update(visible=False)

Bulk operation bỏ qua signal và bỏ qua Model.save() override — nếu logic phụ thuộc signal (audit log, invalidate cache) thì phải tự handle sau bulk. auto_now cũng không chạy với update(); truyền updated_at=timezone.now() thẳng vào kwargs khi cần.

Looping instance.save() for 10k rows = 10k round-trip INSERTs → tens of seconds wasted and the DB drowns in connections. Bulk operations collapse into 1-N SQL statements → 10–100× faster.

• Model.objects.bulk_create(objs, batch_size=1000) — one INSERT per batch. Bypasses save() overrides, pre_save/post_save signals, auto_now. Supports ignore_conflicts=True (skip duplicates) and update_conflicts=True (Postgres-style upsert).
• Model.objects.bulk_update(objs, fields, batch_size=1000) — one UPDATE ... CASE WHEN per batch. Objects must already have a pk.
• QuerySet.update(kwargs)** — one UPDATE SQL for every matching row; same value or an F() expression. No need to load objects into Python.

# Wrong: 10k queries
for row in rows:
    Item(sku=row.sku, price=row.price).save()

# Right: 10 queries (batch_size=1000)
Item.objects.bulk_create(
    [Item(sku=r.sku, price=r.price) for r in rows],
    batch_size=1000,
)

# Same value across rows, no need to load objects
Item.objects.filter(category='archived').update(visible=False)

Pitfall: Bulk operations skip signals and skip Model.save() overrides — if your logic relies on signals (audit log, cache invalidation), handle it manually after the bulk. auto_now also does not fire with update(); pass updated_at=timezone.now() explicitly when needed.