Sticky sessions and use cases

Sticky sessions, also known as session persistence or session affinity, are a method used by load balancers to ensure that all requests from a specific client are directed to the same backend server for the duration of a session. This technique is particularly useful in web applications where the user's state is stored locally on the server, such as in a user's session object.

How Sticky Sessions Work

When a client first makes a request to a web application using a load balancer that supports sticky sessions, the load balancer selects a backend server based on its load balancing algorithm (e.g., round-robin, least connections). Once the server is selected, the load balancer creates a unique identifier to mark the session. This identifier is then sent to the client, usually as a cookie. For subsequent requests, the client sends this identifier back to the load balancer, allowing it to route the requests to the same server as the first request.

Why Use Sticky Sessions?

Sticky sessions can be important for applications that do not store session data in a centralized database or in-memory data store accessible by all servers in the pool. Instead, session data is stored locally on one server. Without sticky sessions, if a client's requests were routed to different servers, any stored session data (like login status or shopping cart contents) would not be accessible, leading to inconsistencies and errors in the application.

Implementation Considerations

1. Session Identifier: Most implementations use cookies to track sessions. However, some systems might use other methods, such as appending session information to URLs.

2. Timeouts: Sessions usually have a timeout period after which the session is considered expired. This helps release server resources tied to old sessions.

3. Persistence Across Failures: If the server handling a session fails, the session might be lost unless there's a mechanism in place to replicate session data across servers or back them up in a central store.

4. Load Distribution: Sticky sessions can lead to uneven load distribution, as new users might be evenly distributed, but returning users are always sent to the same server, regardless of its current load.

Alternatives to Sticky Sessions

To avoid the potential downsides of sticky sessions, such as the risk of uneven load distribution and the dependency on a single server for session data, many modern applications use centralized session stores. These can be in-memory data stores like Redis or Memcached, where session data is accessible to all servers in the pool. This approach allows the application to remain stateless, improving scalability and reliability while still supporting session data.