Singleton Pattern

1. What Is It?

The Singleton pattern ensures a class has exactly one instance throughout the lifetime of an application, and provides a single global access point to that instance. Without it, client code could accidentally create multiple independent instances of what should be a shared resource — imagine two separate logging systems each writing to the same file, or two cache managers each holding a divergent view of the data.

The pattern solves both problems in tandem: it prevents duplicate instantiation and centralizes access. The instance is typically created lazily (on first access) to avoid paying the initialization cost until it's actually needed.

2. How It Works

Step-by-step mechanics:

1. Make the constructor private (Java) or override the creation mechanism (Python).
2. Store the single instance in a private static field on the class itself.
3. Expose a public static factory method (getInstance() / get_instance()) that creates the instance on first call and returns the cached one on every subsequent call.
4. In multi-threaded environments, guard the first-creation check with a lock to prevent two threads from each seeing "no instance yet" simultaneously.

Python

from __future__ import annotations
import threading


class SingletonMeta(type):
    """Thread-safe metaclass that enforces single-instance creation."""

    _instances: dict[type, "SingletonMeta"] = {}
    _lock: threading.Lock = threading.Lock()

    def __call__(cls, *args, **kwargs):
        with cls._lock:
            if cls not in cls._instances:
                instance = super().__call__(*args, **kwargs)
                cls._instances[cls] = instance
        return cls._instances[cls]


class DatabaseConnection(metaclass=SingletonMeta):
    """A database connection pool — exactly one should exist per process."""

    def __init__(self, url: str) -> None:
        self.url = url
        self._connected = False

    def connect(self) -> None:
        if not self._connected:
            print(f"Connecting to {self.url}")
            self._connected = True

    def query(self, sql: str) -> list:
        print(f"Executing: {sql}")
        return []


# Usage
db1 = DatabaseConnection("postgres://localhost/mydb")
db2 = DatabaseConnection("postgres://localhost/mydb")
assert db1 is db2  # True — same object

Java

public final class DatabaseConnection {

    // volatile ensures the write to _instance is visible across threads immediately
    private static volatile DatabaseConnection instance;
    private final String url;
    private boolean connected;

    private DatabaseConnection(String url) {
        this.url = url;
    }

    public static DatabaseConnection getInstance(String url) {
        if (instance == null) {                        // first check (no lock)
            synchronized (DatabaseConnection.class) {
                if (instance == null) {                // second check (with lock)
                    instance = new DatabaseConnection(url);
                }
            }
        }
        return instance;
    }

    public void connect() {
        if (!connected) {
            System.out.println("Connecting to " + url);
            connected = true;
        }
    }

    public void query(String sql) {
        System.out.println("Executing: " + sql);
    }
}

// Preferred modern Java form — enum-based Singleton
public enum AppConfig {
    INSTANCE;

    private final String databaseUrl = System.getenv("DB_URL");

    public String getDatabaseUrl() { return databaseUrl; }
}

// Usage
AppConfig config = AppConfig.INSTANCE;

The Java enum form is thread-safe by JVM specification, serialisation-safe without extra code, and immune to reflection-based instantiation attacks.

3. Variants & Comparisons

4. When to Use It (and When NOT To)

Use Singleton when:

• Exactly one shared resource must coordinate access across the system: logging, configuration, connection pools, device drivers, caches.
• You need a controlled access point, not just a convenient global.

Do NOT use Singleton when:

• You're reaching for it just to avoid passing a dependency — use dependency injection instead.
• You need testability — Singletons are hard to replace with mocks; they carry global state between tests.
• The "one instance" constraint is an assumption, not a hard requirement — requirements change (multi-tenant systems, test isolation).

Anti-patterns:

• Singleton as a service locator — hiding all dependencies behind a Singleton registry couples everything together invisibly.
• Mutable global state — a Singleton holding mutable data creates hidden coupling; concurrent mutations require careful locking.
• Overuse — using Singleton for every utility class turns object-oriented code into disguised global state.

Decision triggers:

• "This resource must be initialized once and shared everywhere" → consider Singleton
• "I just want easy access to this object" → use dependency injection instead
• "I need one instance per thread or per request" → use thread-local storage or scoped DI containers

5. Real-World Usage

Java java.lang.Runtime Runtime.getRuntime() returns the single Runtime instance representing the JVM process. There is exactly one runtime per JVM — the Singleton pattern enforces this constraint at the API level.

Spring Framework ApplicationContext Spring beans are Singleton-scoped by default (@Scope("singleton")). The ApplicationContext acts as a managed Singleton registry — it creates each bean once and returns the same instance on every injection. Spring takes the management burden off the developer while preserving the one-instance guarantee.

Python logging module logging.getLogger("myapp") always returns the same Logger instance for the same name. The logging module uses a module-level dictionary to cache loggers — effectively a Singleton registry — so that all code sharing a logger name sees the same configuration and handlers.

6. Interview Cheat Sheet

Key sentences to demonstrate depth:

1. "Singleton combines two responsibilities — controlling instantiation and providing global access — which technically violates SRP, so I'd use it only when both constraints are genuinely required."
2. "In Java, I prefer the enum form because the JVM guarantees a single enum constant, making it serialization-safe and reflection-safe without any extra code."
3. "In Python, a metaclass override is cleaner than overriding __new__ because it doesn't interfere with the class's own __init__ semantics."
4. "The biggest cost of Singleton is testability — global state bleeds between tests unless you add a reset mechanism or inject the singleton as a dependency."

Common follow-up questions:

• "How do you make a Singleton thread-safe in Java?" → Use double-checked locking with volatile, or the enum form which is JVM-guaranteed thread-safe.
• "What's the difference between Singleton and a static class?" → A Singleton can implement interfaces, be subclassed, and passed as a dependency; a static class cannot. Singletons can be replaced with polymorphic alternatives; static classes cannot.
• "How do you test code that depends on a Singleton?" → Either extract an interface the Singleton implements and inject a mock, or add a package-private resetInstance() method for tests.

Connections to other concepts:

• Monostate pattern — alternative where multiple instances share the same state via static fields; avoids the enforcement mechanism but achieves the same effect.
• Registry pattern — a generalization: one instance per key, not one instance total. Often built using the same caching technique.
• Dependency Inversion — pairing Singleton with an interface allows the one instance to be replaced by a test double, restoring testability.