1.2 How to Implement Distributed Locks

1. What Is a Distributed Lock?

A lock in a distributed environment (across processes or machines).

It satisfies the following conditions:

• Atomicity

Locking and unlocking operations must be atomic.

• Mutual exclusion

At any moment, only one client can hold the lock.

• No deadlock

Even if a client crashes while holding the lock without actively unlocking it, other clients must still be able to acquire the lock afterward.

• Locking and unlocking must be performed by the same client.

Locking and unlocking must be performed by the same client. A client cannot release a lock acquired by someone else.

For example, in this scenario, A acquires the lock with a lock-holding duration of 10s, but A actually runs for 15s, so the lock is automatically released. Then B successfully acquires the lock. At the 15th second, A finishes execution and releases the lock. Here, A has released the lock acquired by B.

2. Why Are Distributed Locks Needed?

Built-in locks in various languages, such as Java’s synchronized and Go’s mutex, can only guarantee lock properties within a single process and cannot span processes or machines.

3. How to Implement Distributed Locks

3.1. ZooKeeper

ZooKeeper Distributed Lock

3.2. Redis

Redis Distributed Lock

3.3. ZooKeeper vs Redis

	Redis	ZooKeeper
Lock acquisition fails	Needs continuous active polling	Register a listener and wait for a callback
Client holding the lock crashes	Must wait until the timeout expires before acquiring the lock again	The ephemeral node automatically releases the lock
Performance	High	Low

4. References

• Several Ways to Implement Distributed Locks - HollisChuang’s Blog