Avoid Flaky Tests Java | Fresher Interview

Direct Answer

Flaky tests can pass or fail with the same code because an input is uncontrolled: time, scheduling, shared state, randomness, ports, files, databases, or remote services. Control the relevant boundary: inject a time/random source, isolate and reset state, wait on an observable condition for asynchronous work, and use a stable fake or controlled integration environment. Awaitility is one optional third-party waiting library, not part of JUnit.

Code Example

import java.time.*;
import org.junit.jupiter.api.Test;
import static org.junit.jupiter.api.Assertions.*;

// FIXED: inject Clock instead of calling now() directly
public class PromotionService {
    private final Clock clock;
    public PromotionService(Clock clock) { this.clock = clock; }

    public boolean isPromotionActive(LocalDateTime start, LocalDateTime end) {
        var now = LocalDateTime.now(clock); // controllable in tests
        return now.isAfter(start) && now.isBefore(end);
    }
}

class PromotionServiceTest {
    @Test
    void isPromotionActive_whenWithinWindow_thenReturnsTrue() {
        var promoTime = LocalDateTime.of(2024, 12, 25, 12, 0);
        var clock = Clock.fixed(promoTime.toInstant(ZoneOffset.UTC), ZoneOffset.UTC);
        var service = new PromotionService(clock);

        boolean result = service.isPromotionActive(
            promoTime.minusHours(1), promoTime.plusHours(1));

        assertTrue(result); // always deterministic
    }
}

Common Mistakes & Pitfalls

• Fixed sleep for async work — a longer sleep is still a timing guess. Prefer a completion signal or a bounded wait on an observable condition.
• Uncontrolled wall-clock time — inject Clock when current time is an input, and make the relevant zone explicit.
• Shared mutable state — isolate data per test or reset it safely; @BeforeEach alone does not protect a static from parallel tests.
• Assuming the test alone is wrong — investigate whether the failure reveals a real race, timeout, cleanup, or dependency problem.
• Rerunning without evidence — preserve logs, random seeds, order, and environment so the cause can be reproduced.