Dependency Injection in Java — Circular Refs That Crash

Every Java application beyond 'Hello World' has objects that depend on other objects. A UserService needs a UserRepository. A PaymentProcessor needs a NotificationClient. The way you wire those relationships together determines how testable, maintainable, and scalable your codebase will be — arguably more than any other single design decision. Get it wrong and you end up with a tightly-coupled monolith where changing one class breaks five others and mocking anything in a unit test requires heroic effort.

Dependency Injection (DI) solves the coupling problem by inverting control: instead of a class creating or locating its own dependencies, an external mechanism provides them. This is the practical application of the Dependency Inversion Principle (the D in SOLID). The result is code where each class declares what it needs without caring how those needs are fulfilled — making it trivially easy to swap implementations, inject mocks in tests, and reason about each class in isolation.

By the end of this article you'll understand the three injection styles and when each one is appropriate, how an IoC container actually resolves a dependency graph at runtime, how Spring implements DI under the hood, and the real-world gotchas that trip up experienced engineers — circular dependencies, prototype beans inside singletons, and the performance cost of reflection-based injection. You'll leave with patterns you can apply tomorrow morning.

What Is Dependency Injection?

Dependency Injection (DI) is a technique where an object receives other objects it depends on from an external source, rather than creating them itself. This external source is called an Inversion of Control (IoC) container. The term 'inversion' refers to the shifted responsibility: your class no longer controls dependency creation — the container does.

In Java, DI is implemented via three primary injection styles: constructor injection (dependencies passed via the constructor), setter injection (dependencies set via setter methods), and field injection (dependencies injected directly onto private fields via reflection).

Constructor injection is the most reliable — it ensures the object is fully initialized upon creation and enables immutability. Field injection, while convenient, introduces testability problems because you can't easily provide mocks without the container. Setter injection sits in the middle: useful for optional dependencies but requires the object to tolerate a partially constructed state.

package io.thecodeforge.di;

// Constructor injection: dependencies are explicit and mandatory
public class PaymentService {
    private final NotificationClient notificationClient;
    private final TransactionRepository transactionRepository;

    public PaymentService(NotificationClient notificationClient,
                          TransactionRepository transactionRepository) {
        this.notificationClient = notificationClient;
        this.transactionRepository = transactionRepository;
    }

    public void process(Order order) {
        transactionRepository.save(order);
        notificationClient.send(order.customerEmail(), "Payment processed");
    }
}

// Field injection: fragile for testing
@Component
public class FragileService {
    @Autowired
    private NotificationClient notificationClient;

    public void notify(String message) {
        notificationClient.send(message); // NPE if client not injected
    }
}

How an IoC Container Resolves the Dependency Graph

When you annotate a class with @Component, @Service, @Repository, or @Controller, Spring's IoC container picks it up during component scanning. It then builds a dependency graph by inspecting each bean's constructors, fields, and setters (depending on the injection strategy).

The container uses a process called 'bean post-processing' to determine the order of instantiation. It first creates beans with no dependencies, then progressively creates those that depend on already-created beans. This is essentially a topological sort of the dependency graph.

If a circular dependency is detected — bean A needs bean B which needs bean A — Spring will throw a BeanCurrentlyInCreationException at startup, unless one of the dependencies uses @Lazy (which breaks the cycle by deferring the creation of the lazy bean until it's actually accessed). Under the hood, Spring uses a 'singleton currently in creation' set to track beans during construction.

package io.thecodeforge.container;

import java.util.*;

// A simplified illustration of how an IoC container resolves dependencies
public class ContainerSimulation {
    private final Map<Class<?>, Object> singletons = new HashMap<>();
    private final Set<Class<?>> currentlyInCreation = new HashSet<>();

    @SuppressWarnings("unchecked")
    public <T> T getBean(Class<T> beanClass) {
        if (singletons.containsKey(beanClass)) {
            return (T) singletons.get(beanClass);
        }
        if (currentlyInCreation.contains(beanClass)) {
            throw new RuntimeException("Circular dependency detected for: " + beanClass.getName());
        }
        currentlyInCreation.add(beanClass);
        // In a real container, this would inspect constructors and resolve dependencies
        T instance = createInstance(beanClass);
        currentlyInCreation.remove(beanClass);
        singletons.put(beanClass, instance);
        return instance;
    }

    private <T> T createInstance(Class<T> beanClass) {
        // Placeholder: real container uses reflection to call constructor with resolved args
        try {
            return beanClass.getDeclaredConstructor().newInstance();
        } catch (Exception e) {
            throw new RuntimeException("Cannot instantiate: " + beanClass.getName(), e);
        }
    }
}

Spring's Autowiring and Qualifiers

Spring's autowiring resolves dependencies by type first, then by qualifier if multiple beans of the same type exist. When a constructor parameter type matches exactly one bean, Spring injects it. If there are multiple beans, Spring tries to match the parameter name to the bean name — and if the name doesn't match, you must use @Qualifier.

This is a common source of head-scratching bugs: you add a new implementation of an interface, and suddenly startup fails with 'NoUniqueBeanDefinitionException' because Spring doesn't know which one to use. The fix is to mark one bean as @Primary or to use @Qualifier on the injection point.

For collections, Spring supports injection of all beans of a given type into a List<Interface>, which is incredibly useful for chain-of-responsibility patterns or multi-algorithm strategies.

package io.thecodeforge.injection;

import org.springframework.context.annotation.*;

@Configuration
public class NotificationConfig {
    @Bean
    @Primary
    public NotificationClient emailClient() {
        return new EmailNotificationClient();
    }

    @Bean
    public NotificationClient smsClient() {
        return new SmsNotificationClient();
    }
}

// In service:
@Service
public class NotificationService {
    private final NotificationClient primaryClient; // gets emailClient due to @Primary
    private final List<NotificationClient> allClients; // gets both

    public NotificationService(NotificationClient primaryClient,
                               List<NotificationClient> allClients) {
        this.primaryClient = primaryClient;
        this.allClients = allClients;
    }

    public void broadcast(String message) {
        allClients.forEach(c -> c.send(message));
    }
}

Circular Dependencies: The Silent Startup Killer

A circular dependency occurs when Bean A depends on Bean B, and Bean B depends directly or indirectly on Bean A. Spring detects this during container initialization and throws BeanCurrentlyInCreationException. The fix is never to 'fix' the annotation; fix the design.

Three proven strategies to break circular dependencies: 1. Extract the shared logic into a third bean that both depend on (Mediator pattern). 2. Use setter injection with @Lazy on one side — this defers the creation of the lazy bean until it's actually needed, but beware: if you call a method on the lazy bean before its dependencies are resolved, you get a NullPointerException. 3. Redesign the architecture: circular dependencies almost always violate the Single Responsibility Principle. Maybe both beans should be merged, or an event-driven approach (ApplicationEventPublisher) would be cleaner.

In large legacy codebases, you'll often encounter indirect cycles (A → B → C → A). These are harder to spot. Use the startup debug log: 'spring.beaninfo.ignore=true' and check for 'Currently in creation' lines.

package io.thecodeforge.di;

import org.springframework.context.annotation.Lazy;
import org.springframework.stereotype.Service;

// BAD: circular constructor injection (will fail)
// @Service
// public class OrderService {
//     private final CustomerService customerService;  // CustomerService needs OrderService
// }

// GOOD: break the cycle with a third service
@Service
public class OrderService {
    private final LoyaltyService loyaltyService;

    public OrderService(LoyaltyService loyaltyService) {
        this.loyaltyService = loyaltyService;
    }
}

@Service
public class CustomerService {
    private final LoyaltyService loyaltyService;

    public CustomerService(LoyaltyService loyaltyService) {
        this.loyaltyService = loyaltyService;
    }
}

// LOOSE COUPLING via events: ApplicationEventPublisher avoids any direct dependency

Performance Cost of Reflection-Based Injection

Spring's DI relies heavily on Java Reflection: scanning classes, inspecting constructors, fields, and annotations, and invoking methods reflectively. For startup, this is acceptable — the overhead is in the order of tens of milliseconds for a typical microservice. However, for request-scoped beans (prototype or request scope) that are created on every request, reflection-based field injection adds measurable latency.

Consider a prototype bean with 10 fields injected via @Autowired. Each field injection requires: field lookup by name, setting accessibility, and field.set() operation. That's ~200μs per prototype bean. Under 5000 requests/sec, that's an extra second per second of CPU time solely for injection.

Constructor injection avoids this because the container calls the constructor with resolved arguments using Constructor.newInstance() — a single reflective call for the entire bean. The difference is an order of magnitude for prototype-scoped beans.

Spring 6 and Spring Boot 3 have improved reflection caching, but the principle remains: constructor injection is not just cleaner design — it's measurably faster under load.

package io.thecodeforge.performance;

// Conceptual benchmark comparison
public class InjectionBenchmark {
    // Field injection: ~2μs per field
    // 10 fields -> ~20μs per instance of a prototype bean

    // Constructor injection: ~5μs for the entire bean (one reflective call)
    // 10 fields -> still ~5μs

    // At 5000 req/s with prototype beans:
    // Field injection: 5000 * 20μs = 100ms/s overhead
    // Constructor injection: 5000 * 5μs = 25ms/s overhead
    // That's 4x less CPU spent on injection.
}

Why Injecting Interfaces is the Only Safe Choice in Production

Most tutorials show you how to inject a concrete class and call it DI. That's not DI. That's fancy instantiation. Real dependency injection only works when you program to interfaces. Why? Because concrete classes chain you to their implementation details. Swap a JDBC connection pool? Now you're rewriting every injection point. Stub a service for integration tests? Can't, if the constructor expects a concrete MySqlPaymentGateway. The interface is your contract. The implementation is just a detail you can replace without touching consumers. In production, your DI container resolves the interface to a concrete bean based on qualifiers, profiles, or even runtime conditions. That's the whole point of “Don’t call us, we’ll call you.” Your code should never, ever instantiate its own dependencies. If I see new PaymentGateway() inside a service class during a code review, that PR gets rejected on the spot. Every class that needs a collaborator should declare that need via an interface in its constructor or setter. That keeps your system decoupled, testable, and deployable without a rewrite every time a vendor changes their API.

// io.thecodeforge — java tutorial

import java.math.BigDecimal;

// The contract — no implementation details, just capability
interface PaymentGateway {
    boolean charge(BigDecimal amount, String currency);
}

// Production implementation
class StripeGateway implements PaymentGateway {
    @Override
    public boolean charge(BigDecimal amount, String currency) {
        System.out.println("Stripe: charging " + amount + " " + currency);
        return true;
    }
}

// Consumer depends on the interface, not the concrete class
class OrderService {
    private final PaymentGateway gateway;

    // Injection point — container provides the implementation
    public OrderService(PaymentGateway gateway) {
        this.gateway = gateway;
    }

    public void checkout(BigDecimal total) {
        if (!gateway.charge(total, "USD")) {
            System.out.println("Payment failed");
        }
        System.out.println("Checkout complete");
    }
}

public class PaymentProcessorWithInterface {
    public static void main(String[] args) {
        // Simulate DI container wiring
        PaymentGateway gateway = new StripeGateway();
        OrderService service = new OrderService(gateway);
        service.checkout(new BigDecimal("49.99"));
    }
}

Constructor Injection is Production-Ready; Field Injection is Technical Debt

You see it everywhere: @Autowired slapped on a private field. Clean, concise, wrong. Field injection hides dependencies. Your class looks like it has none until Spring calls new through reflection, leaving the field null until the proxy is fully constructed. That means you can’t test it without the container. Constructor injection makes dependencies explicit. Every constructor parameter is a dependency your class admits it needs. No surprises. No null pointer exceptions because the container missed a field. The class is immutable after construction and always in a valid state. In production, your build tool (SpotBugs, Checkstyle) can verify that every field is final and set in the constructor. Field injection defeats static analysis. Use it in throwaway prototypes if you must. In production code, you’re asking for runtime failures that compile-time checks would have caught. Spring’s own docs recommend constructor injection. When you see a colleague using field injection, send them here. It’s not style. It’s correctness.

// io.thecodeforge — java tutorial

import org.springframework.beans.factory.annotation.Autowired;
import org.springframework.stereotype.Component;

// BAD: Field injection — dependencies are invisible and mutation-prone
@Component
class ReportServiceFieldInjected {
    @Autowired
    private DataRepository repository;

    public void generate() {
        // Can NPE if Spring didn't wire repository
        System.out.println(repository.fetch());
    }
}

// GOOD: Constructor injection — explicit, final, testable
@Component
class ReportServiceConstructorInjected {
    private final DataRepository repository;

    // Spring calls this constructor with the resolved dependency
    public ReportServiceConstructorInjected(DataRepository repository) {
        this.repository = repository;
    }

    public void generate() {
        System.out.println(repository.fetch());
    }
}

// Stub for demonstration
class DataRepository {
    public String fetch() {
        return "production data";
    }
}

public class ConstructorVsFieldInjection {
    public static void main(String[] args) {
        // Manual wiring — works for constructor injection
        DataRepository repo = new DataRepository();
        ReportServiceConstructorInjected service = new ReportServiceConstructorInjected(repo);
        service.generate();

        // Field-injected version cannot be tested this way — requires Spring Context
        // ReportServiceFieldInjected bad = new ReportServiceFieldInjected();
        // bad.generate(); // NPE!
    }
}

Scoping: How Singletons, Prototypes, and Request Scopes Actually Behave Under Load

Every DI framework defaults to singleton scoping. That means one bean instance lives for the entire application context. It’s fast and thread-safe if the bean is stateless. But throw a stateful bean into a singleton scope and you’re debugging race conditions at 3 AM. Prototype scope creates a new instance every time you request a bean. Period. Useful for objects that hold request-specific state. But if a singleton bean injects a prototype bean, guess what? That prototype is instantiated once when the singleton is created, not when you call a method. The container only creates new prototypes when you call applicationContext.getBean() or use @Lookup or javax.inject.Provider. In a web app, request and session scopes keep beans alive for the duration of an HTTP request or user session. These rely on proxy mode (scoped-proxy="target-class" in XML or @Scope(proxyMode = ScopedProxyMode.TARGET_CLASS)). Without the proxy, a singleton that injects a request-scoped bean will hold a stale instance. I’ve seen that bug sink a production deployment. Know your scope lifetimes. Over-scope to singleton, you share state. Under-scope to prototype, you leak memory. Pick the right scope for the job.

// io.thecodeforge — java tutorial

import org.springframework.beans.factory.annotation.Lookup;
import org.springframework.context.annotation.*;
import org.springframework.stereotype.Component;

// Prototype bean — new instance each time
@Component
@Scope("prototype")
class RequestContext {
    private final String id = java.util.UUID.randomUUID().toString();
    public String getId() { return id; }
}

// Singleton bean — lives once across the app
@Component
@Scope("singleton")
class TrackerService {
    // BAD: Direct injection of prototype into singleton — captured once
    @Autowired
    private RequestContext context;

    public void track() {
        // This prints the same UUID every time! Context is not fresh
        System.out.println("Tracker context ID: " + context.getId());
    }
}

// CORRECT: Use Provider or @Lookup to get a new prototype each call
@Component
@Scope("singleton")
class TrackerServiceCorrect {
    @Lookup
    public RequestContext getContext() { return null; } // Spring overrides this

    public void track() {
        System.out.println("TrackerCorrect context ID: " + getContext().getId());
    }
}

@Configuration
@ComponentScan
public class ScopeTrapExample {
    public static void main(String[] args) throws Exception {
        AnnotationConfigApplicationContext ctx = new AnnotationConfigApplicationContext(ScopeTrapExample.class);

        TrackerService bad = ctx.getBean(TrackerService.class);
        bad.track();
        bad.track();  // Same UUID — broken

        TrackerServiceCorrect good = ctx.getBean(TrackerServiceCorrect.class);
        good.track();
        good.track();  // Different UUID each time

        ctx.close();
    }
}

Stop Wiring by Hand: How to Build a Custom Injector That Won't Embarrass You in Production

Your team is wasting time wiring objects manually or cargo-culting Spring annotations without understanding what happens underneath. That's how you get startups that fail silently at 3 AM. You need to know how to build a minimal DI container from scratch—not to replace Spring, but to understand why Spring works the way it does.

// io.thecodeforge — java tutorial

import java.lang.reflect.*;
import java.util.*;

public class CustomInjector {
    private final Map<Class<?>, Object> singletons = new HashMap<>();
    private final Set<Class<?>> resolving = new HashSet<>(); // cycle guard

    @SuppressWarnings("unchecked")
    public <T> T resolve(Class<T> clazz) {
        if (singletons.containsKey(clazz)) return (T) singletons.get(clazz);
        if (!resolving.add(clazz)) throw new RuntimeException("Cycle: " + clazz);
        try {
            Constructor<?> ctor = clazz.getDeclaredConstructors()[0];
            Object[] args = Arrays.stream(ctor.getParameterTypes())
                    .map(this::resolve).toArray();
            T instance = (T) ctor.newInstance(args);
            singletons.put(clazz, instance);
            return instance;
        } catch (Exception e) {
            throw new RuntimeException("Inject failed: " + clazz, e);
        } finally {
            resolving.remove(clazz);
        }
    }

    public static void main(String[] args) {
        var injector = new CustomInjector();
        System.out.println(injector.resolve(Service.class).work());
    }
}

class Service {
    private final Repo repo;
    public Service(Repo repo) { this.repo = repo; }
    public String work() { return repo.fetch(); }
}

class Repo {
    public String fetch() { return "data"; }
}

// io.thecodeforge — java tutorial

import java.lang.reflect.Constructor;

public class ReflectionBenchmark {
    static class Payload {
        public Payload(String a, int b) {}
    }

    public static void main(String[] args) throws Exception {
        // Warmup
        for (int i = 0; i < 10_000; i++) {
            new Payload("x", 1);
            Constructor<?> c = Payload.class.getConstructors()[0];
            c.newInstance("x", 1);
        }

        long direct = System.nanoTime();
        for (int i = 0; i < 100_000; i++) new Payload("x", 1);
        long directTime = System.nanoTime() - direct;

        long reflect = System.nanoTime();
        Constructor<?> c = Payload.class.getConstructors()[0];
        for (int i = 0; i < 100_000; i++) c.newInstance("x", 1);
        long reflectTime = System.nanoTime() - reflect;

        System.out.println("Direct: " + directTime / 100_000 + " ns");
        System.out.println("Reflect: " + reflectTime / 100_000 + " ns");
        System.out.println("Overhead: " + 
            ((reflectTime - directTime) / 100_000) + " ns per bean");
    }
}