Testing Event-Sourced Systems — Event Sourcing & CQRS

Event sourcing makes behavior tests expressive: Given past events, When a command is handled, Then expect new events (or an error). Projection tests: given events, expect read-model rows. These tests are fast (no database for aggregates), document business rules, and serve as regression tests for upcasters (old events → expected current shape).

Test shapes

Test type	Given	When	Then
Aggregate behavior	Past events	Command	New events / error
Projection	Events	Projector runs	Read-model rows
Upcaster	Old event JSON	Upcast	Current schema
Process manager	Events	Policy reacts	Commands issued

def test_cannot_withdraw_more_than_balance():
    given = [{"type": "AccountOpened", "account_id": "a1"}, {"type": "MoneyDeposited", "amount": 100}]
    with pytest.raises(DomainError, match="Insufficient funds"):
        decide_withdraw(rehydrate(given), amount=150)

def test_withdrawal_emits_event():
    given = [{"type": "AccountOpened", "account_id": "a1"}, {"type": "MoneyDeposited", "amount": 100}]
    assert decide_withdraw(rehydrate(given), amount=40) == [{"type": "MoneyWithdrawn", "amount": 40}]

it("closes an account only when balance is zero", () => {
  const given: Event[] = [{ type: "AccountOpened", accountId: "a1" }, { type: "MoneyDeposited", amount: 10 }];
  expect(() => decideClose(rehydrate(given))).toThrow("Balance must be zero");
  const settled = [...given, { type: "MoneyWithdrawn", amount: 10 } as Event];
  expect(decideClose(rehydrate(settled))).toEqual([{ type: "AccountClosed" }]);
});

Common pitfalls

Testing internal state instead of emitted events.
No tests for old event versions.
Projection tests that depend on a shared database state.

Key points

Given-when-then on events
Test upcasters with real old payloads
Fast, database-free aggregate tests

Further reading: Martin Fowler: Given When Then

Practice questions

Why are given-when-then tests natural for event-sourced aggregates?

Answer

Aggregate state is defined by past events (given), behavior by commands (when) and outcomes by emitted events (then), so tests read like business rules.

How do you test upcasters?

Answer

Store real examples of old event payloads as fixtures and assert they upcast to the expected current schema.