- Book: Decoupled PHP — Clean and Hexagonal Architecture for Applications That Outlive the Framework
- Also by me: Thinking in Go (2-book series) — Complete Guide to Go Programming + Hexagonal Architecture in Go
- My project: Hermes IDE | GitHub — an IDE for developers who ship with Claude Code and other AI coding tools
- Me: xgabriel.com | GitHub
You have an Order. It owns line items and a shipping address. In the
database that is three tables: orders, order_items,
order_addresses. The aggregate is one thing in the domain and three
rows-with-children in storage.
Now look at what most codebases do with that. The service loads the
order, loops the items, saves each one in its own statement. Halfway
through, a unique-constraint violation throws. The order header is
already committed. Two items are in. One address is missing. You have a
row in orders that no part of the system considers valid, and no
single place to point at when you ask how it got there.
The aggregate is supposed to be a consistency boundary. Saving its
parts in separate, independently-committing operations breaks that
boundary on the way to disk. This post is about closing it: one
repository, one transactional save across every table, and a read path
that rebuilds the whole object from rows without leaking the ORM into
the domain.
The aggregate the domain sees
The domain class has no idea it lives in three tables. It holds its
children directly and guards their invariants.
<?php
declare(strict_types=1);
namespace App\Domain\Order;
use App\Domain\Shared\Money;
final class Order
{
/** @var list<LineItem> */
private array $items;
private function __construct(
private readonly OrderId $id,
private readonly CustomerId $customerId,
array $items,
private Address $shipTo,
private OrderStatus $status,
) {
if ($items === []) {
throw new InvalidOrder('order needs an item');
}
$this->items = array_values($items);
}
public static function place(
OrderId $id,
CustomerId $customerId,
array $items,
Address $shipTo,
): self {
return new self(
$id, $customerId, $items,
$shipTo, OrderStatus::Placed,
);
}
public static function restore(
OrderId $id,
CustomerId $customerId,
array $items,
Address $shipTo,
OrderStatus $status,
): self {
return new self(
$id, $customerId, $items, $shipTo, $status,
);
}
public function total(): Money
{
$sum = Money::zero($this->items[0]->price->currency);
foreach ($this->items as $item) {
$sum = $sum->plus($item->lineTotal());
}
return $sum;
}
public function id(): OrderId { return $this->id; }
public function status(): OrderStatus { return $this->status; }
/** @return list<LineItem> */
public function items(): array { return $this->items; }
public function shipTo(): Address { return $this->shipTo; }
public function customerId(): CustomerId
{
return $this->customerId;
}
}
LineItem and Address are small value objects in the same namespace.
The point: the items are loaded the moment the order is. There is no
proxy, no getItems() that fires a query on first access. A loaded
Order is fully present in memory or it does not exist yet.
The port the use case depends on
The use case asks for an OrderRepository. The interface speaks domain
types and hides every table.
<?php
declare(strict_types=1);
namespace App\Application\Port;
use App\Domain\Order\Order;
use App\Domain\Order\OrderId;
interface OrderRepository
{
public function save(Order $order): void;
public function findById(OrderId $id): ?Order;
}
One save, one find. The caller never sees that "save" touches three
tables. It hands over the aggregate and trusts the repository to land
the whole thing or none of it.
The write path: one transaction, three tables
Here is the adapter against raw PDO. No ORM, so nothing is hidden. You
can see the transaction open, the delete-then-insert for children, and
the commit.
<?php
declare(strict_types=1);
namespace App\Infrastructure\Persistence\Pdo;
use App\Application\Port\OrderRepository;
use App\Domain\Order\Order;
use App\Domain\Order\OrderId;
use PDO;
final readonly class PdoOrderRepository implements
OrderRepository
{
public function __construct(private PDO $pdo) {}
public function save(Order $order): void
{
$this->pdo->beginTransaction();
try {
$this->upsertHeader($order);
$this->replaceItems($order);
$this->replaceAddress($order);
$this->pdo->commit();
} catch (\Throwable $e) {
$this->pdo->rollBack();
throw $e;
}
}
private function upsertHeader(Order $order): void
{
$sql = 'INSERT INTO orders
(id, customer_id, status)
VALUES (:id, :cust, :status)
ON DUPLICATE KEY UPDATE
status = VALUES(status)';
$this->pdo->prepare($sql)->execute([
'id' => $order->id()->value,
'cust' => $order->customerId()->value,
'status' => $order->status()->value,
]);
}
private function replaceItems(Order $order): void
{
$del = $this->pdo->prepare(
'DELETE FROM order_items WHERE order_id = :id'
);
$del->execute(['id' => $order->id()->value]);
$ins = $this->pdo->prepare(
'INSERT INTO order_items
(order_id, sku, qty, price_minor, currency)
VALUES (:oid, :sku, :qty, :price, :cur)'
);
foreach ($order->items() as $item) {
$ins->execute([
'oid' => $order->id()->value,
'sku' => $item->sku,
'qty' => $item->quantity,
'price' => $item->price->amountMinor,
'cur' => $item->price->currency,
]);
}
}
private function replaceAddress(Order $order): void
{
$a = $order->shipTo();
$sql = 'INSERT INTO order_addresses
(order_id, line1, city, country)
VALUES (:id, :l1, :city, :country)
ON DUPLICATE KEY UPDATE
line1 = VALUES(line1),
city = VALUES(city),
country = VALUES(country)';
$this->pdo->prepare($sql)->execute([
'id' => $order->id()->value,
'l1' => $a->line1,
'city' => $a->city,
'country' => $a->country,
]);
}
}
Three things earn their place here.
The whole write is in one transaction. If the address insert fails, the
header and items roll back with it. The database never holds a
half-written aggregate. That is the consistency boundary, honored at the
storage edge.
Children are replaced, not diffed. DELETE every order_item for this
order, then re-insert the current set. You skip the bookkeeping of
"which item is new, which changed, which was removed." The aggregate in
memory is the source of truth; the rows become a copy of it. For a
small child collection this is the simplest correct option, and it is
within one transaction, so no reader ever sees the gap between delete
and insert.
The header upsert means save works for both a new order and an edited
one. The use case calls save and does not branch on whether the order
existed.
The read path: rebuild the whole from rows
Loading is the mirror image. Pull every table for the id, then hand the
rows to a factory that rebuilds the aggregate.
public function findById(OrderId $id): ?Order
{
$header = $this->fetchHeader($id->value);
if ($header === null) {
return null;
}
$items = $this->fetchItems($id->value);
$address = $this->fetchAddress($id->value);
return OrderMapper::toDomain($header, $items, $address);
}
private function fetchHeader(string $id): ?array
{
$stmt = $this->pdo->prepare(
'SELECT id, customer_id, status
FROM orders WHERE id = :id'
);
$stmt->execute(['id' => $id]);
$row = $stmt->fetch(PDO::FETCH_ASSOC);
return $row === false ? null : $row;
}
private function fetchItems(string $id): array
{
$stmt = $this->pdo->prepare(
'SELECT sku, qty, price_minor, currency
FROM order_items WHERE order_id = :id'
);
$stmt->execute(['id' => $id]);
return $stmt->fetchAll(PDO::FETCH_ASSOC);
}
Three queries, all eager. The order, its items, its address, fetched
before the aggregate is built. No state in the returned Order is a
promise to query later. That is the difference between this and a lazy
ORM relation, and it is the difference that bites in production.
A lazy collection looks loaded. You read $order->items() inside a
controller and it works, because the entity manager is still open. The
same call inside a queue worker, after the unit of work has closed,
throws a LazyInitializationException or hands you an empty set. The
read path above cannot do that. The items are plain value objects in an
array the moment findById returns.
The mapper keeps the ORM out of the domain
The mapper is the only code that knows column names. It turns rows into
value objects and calls the domain's restore factory.
<?php
declare(strict_types=1);
namespace App\Infrastructure\Persistence\Pdo;
use App\Domain\Order\Address;
use App\Domain\Order\CustomerId;
use App\Domain\Order\LineItem;
use App\Domain\Order\Order;
use App\Domain\Order\OrderId;
use App\Domain\Order\OrderStatus;
use App\Domain\Shared\Money;
final class OrderMapper
{
public static function toDomain(
array $header,
array $itemRows,
array $addressRow,
): Order {
$items = array_map(
static fn (array $r): LineItem => new LineItem(
sku: $r['sku'],
quantity: (int) $r['qty'],
price: new Money(
(int) $r['price_minor'],
$r['currency'],
),
),
$itemRows,
);
return Order::restore(
id: new OrderId($header['id']),
customerId: new CustomerId(
$header['customer_id']
),
items: $items,
shipTo: new Address(
$addressRow['line1'],
$addressRow['city'],
$addressRow['country'],
),
status: OrderStatus::from($header['status']),
);
}
}
restore rebuilds the object without re-running placement invariants.
The rows were valid when they were written; re-checking them on load
hides the bug in whatever wrote a bad row. The domain Order has zero
imports from Infrastructure. Swap PDO for Doctrine's DBAL, or for a
document store that keeps the whole aggregate in one JSON column, and
only this adapter and mapper change.
Where the transaction actually lives
In the PDO adapter above, save owns its transaction, which is fine
when one save is one unit of work. The moment a use case saves two
aggregates that must commit together, the transaction belongs one layer
out, around the use case, behind a port:
interface TransactionRunner
{
/** @template T @param callable():T $work @return T */
public function run(callable $work): mixed;
}
The use case wraps its work in run(...); the repository's save joins
the open transaction instead of starting its own. The same delete-insert
write stays correct; it just stops being the outermost boundary. Either
way, the rule holds: the aggregate lands whole or not at all.
The discipline in one line
One aggregate maps to one repository. The repository owns the spread
across tables on write and the gather across tables on read. Children
load eagerly, save inside a single transaction, and the domain class
never learns a table name. That is the whole pattern, and it is the same
whether the ORM is Doctrine, Eloquent mapped to a side class, or raw
PDO.
If this was useful
This mapping problem — one aggregate, many tables, no leaks — gets a
full treatment in Decoupled PHP, next to the repository, transaction,
and unit-of-work patterns it leans on. The book's whole argument is that
storage is an adapter your domain shouldn't feel, and the order
aggregate here is the running example it builds on.
Available on Kindle, Paperback, and Hardcover. English, German, and Japanese editions out now — Portuguese and Spanish coming soon.