In this lesson, you will learn how to navigate the trade-off between strict data consistency and system performance by selecting an appropriate replication mode.
When architecting distributed systems, you must decide how the primary node communicates data changes to its replicas. This decision defines your system's "write path" and dictates your resilience against data loss.
Synchronous Replication: Safety Over Speed
In synchronous replication, the primary node waits for a confirmation (acknowledgment) from one or more replicas before committing a transaction to the client. This guarantees that if the primary fails, the replicas have an identical copy of the data.
The trade-off: Write latency increases because the total time for a write is T_primary + T_network + T_replica. If a replica is slow or partitioned, the primary stalls, which can significantly degrade throughput.
-- Conceptual flow of a synchronous write:
-- 1. Primary receives UPDATE
-- 2. Primary sends data to Replica
-- 3. Replica confirms receipt
-- 4. Primary sends "Success" to Client
Asynchronous Replication: Speed Over Safety
In asynchronous replication, the primary node applies the write locally and acknowledges it to the client immediately, then propagates the change to replicas in the background.
The trade-off: Write latency is minimized, as the client doesn't wait for the replica network round-trip. However, this introduces the risk of data loss: if the primary crashes before the background sync completes, the most recent writes are lost (the "durability gap").
Choosing the Right Mode
Selecting the mode depends on your business requirements:
- Choose Synchronous when: Data integrity is non-negotiable. Financial systems or inventory management systems often require this to ensure no transaction is ever "lost" during a failover.
- Choose Asynchronous when: The system requires high availability and low latency. Many social media platforms or real-time analytics dashboards prioritize the user experience of a fast write, accepting that a rare, minor lag (replication lag) or the loss of a millisecond of data is preferable to a blocked application.
By understanding the latency penalty of synchronous coordination versus the durability risk of asynchronous propagation, you can tune your distributed database to match the specific needs of your application's workload.
Key Takeaway: Synchronous replication prioritizes zero data loss at the cost of higher latency, while asynchronous replication prioritizes high throughput and low latency at the risk of losing recent updates during a primary node failure.