Handling Stale Reads in Weak Consistency

In this lesson, you will learn how to prevent your application from accidentally serving stale data by using versioning and sequence numbers within your read-write workflows.

When working with read-replicas in a weakly consistent system, there is often a time lag between a write on the leader and the replication of that data to followers. If a user performs an update and immediately refreshes their page, a load balancer might route them to a replica that hasn't received the update yet, resulting in a "stale read."

To combat this, we implement Version Checks or Sequence Numbers (often called "log sequence numbers" or LSNs).

How it works

1. The Write: When the application writes to the leader, the database returns a unique version identifier or a transaction sequence number associated with that specific commit.
2. The Metadata: The application stores this version identifier (often in a session cookie or a cache).
3. The Read: When the application queries a replica, it sends this version number along with the request.
4. The Validation: The replica compares the requested version with its own current state. If the replica’s LSN is behind the requested version, it has two choices: wait until it catches up, or redirect the request back to the leader to ensure the user sees the most recent data.

Implementation Illustration

Consider an application that updates a user profile:

-- Application receives a version stamp from the write
-- SET @last_seen_version = 105;

-- When reading from a replica, we check consistency:
SELECT data 
FROM users 
WHERE id = 123 
AND version_id >= 105; 

If the replica returns empty (because its version_id is only 104), the application knows it is looking at stale data. It can then choose to either inform the user that a sync is in progress or force a fetch from the leader. By attaching these sequence markers to our read operations, we decouple the application's correctness from the inherent replication lag of the underlying cluster.

Key Takeaway: By tracking version numbers or sequence stamps, applications can detect and reject stale data from replicas, ensuring that users only interact with information that meets their required level of consistency.