Verteilte Sperren mit Redis

Imagine multiple applications or services trying to update the same piece of data, like a user's balance or a unique order ID. Without coordination, they could overwrite each other's changes, leading to data corruption or incorrect states.

• Race conditions: When operations depend on specific timing, leading to unpredictable results.
• Data integrity: Ensuring shared resources remain consistent.
• Concurrency control: Managing access when multiple processes run simultaneously.

Distributed locks are like a single key to a shared room. Only one process can hold the key (lock) at a time, ensuring exclusive access.

Redis is an excellent choice for implementing distributed locks due to its speed, atomic operations, and single-threaded nature.

• Atomicity: Redis commands are executed atomically, meaning they complete entirely or not at all, preventing partial updates.
• Performance: In-memory operations ensure very low latency for lock acquisition and release.
• Simplicity: Basic string commands can be leveraged effectively for locking.

These features allow Redis to manage lock states reliably and efficiently across many different application instances.

The core of a Redis-based lock lies in the SET command with specific options:

• NX (Not eXists): Ensures the key is only set if it doesn't already exist. This is how we acquire the lock exclusively.
• EX (EXpire): Sets an expiration time for the key in seconds. This is crucial to prevent deadlocks if a client crashes before releasing the lock.
• value: A unique identifier for the lock holder (e.g., a UUID). This allows only the original lock holder to release it.

The command looks like this: SET mylock:resource <unique_id> NX EX 30

Here's a simplified example of how a client might attempt to acquire a lock using Python. The SET command returns True (or 'OK') if the lock was acquired, and False (or None) otherwise.

Simply deleting the lock key (e.g., DEL mylock:resource) is NOT safe. Consider this scenario:

1. Client A acquires the lock with an expiration of 10 seconds.
2. Client A's operation takes longer than 10 seconds, so the lock expires automatically.
3. Client B acquires the lock for the same resource.
4. Client A's operation finally finishes and attempts to delete the lock. It unknowingly deletes Client B's lock!

This leads to two clients thinking they have the lock simultaneously, defeating its purpose. We need an atomic way to check ownership AND delete.

Redis allows executing Lua scripts atomically. This means the entire script runs as a single, uninterruptible operation on the Redis server, solving our lock release problem.

The Lua script for releasing a lock typically does two things:

1. It checks if the current lock value matches the unique ID provided by the client attempting to release the lock.
2. If they match, it deletes the lock key.

This guarantees that only the client who originally acquired the lock (and whose unique ID is still stored) can release it, even if the lock has expired and been re-acquired by another client.

Here's how you'd execute a Lua script to safely release a lock. The script ensures that you only delete the lock if you are its rightful owner.

The EX (expiration) option is vital. It acts as a safety net, ensuring locks are eventually released even if the client crashes or fails to release it. Without expiration, a crashed client could hold a lock indefinitely, causing a permanent deadlock.

• Too short: If the expiration is too short, long-running tasks might lose their lock prematurely.
• Too long: If it's too long, deadlocks might persist for an unacceptable duration.

For tasks that might exceed the initial expiration, clients can implement a 'lock renewal' mechanism. This involves periodically checking if they still hold the lock and, if so, extending its expiration time using EXPIRE or PEXPIRE commands.

While a single Redis instance with SET NX EX works for many cases, it has a single point of failure. If that Redis instance goes down, all locks are lost or become unavailable.

The Redlock algorithm addresses this by requiring clients to acquire locks on a majority of independent Redis master instances (e.g., 3 out of 5). This provides higher guarantees:

• Fault tolerance: If one Redis instance fails, the system can still function.
• Stronger safety: Reduces the chance of multiple clients acquiring the same lock.

Redlock is more complex to implement but offers a robust solution for critical distributed systems.

Which of the following are essential properties for a reliable distributed lock using Redis, assuming a single Redis instance?

In this lesson, we explored how to implement reliable distributed locks using Redis.

• We learned that SET key value NX EX seconds is the foundation for acquiring a lock atomically.
• A unique value identifies the lock holder, preventing others from releasing it.
• Expiration (EX) is vital for preventing permanent deadlocks if a client crashes.
• Safely releasing a lock requires an atomic check-and-delete operation, typically achieved with Lua scripting.
• For higher reliability and fault tolerance in multi-instance setups, the Redlock algorithm can be employed.

Understanding these principles is key to building robust and concurrent distributed applications.