읽기 설정: 읽기 부하 분산하기

A read preference controls which replica set member the driver sends read operations to. By default, all reads go to the primary to ensure you always see the latest data. However, redirecting reads to secondaries can reduce load on the primary and bring reads geographically closer to users — at the cost of potentially reading slightly stale data.

primary (the default) routes every read to the primary. This guarantees strong consistency — you will always read your own writes and see the most up-to-date data. The downside is that the primary handles all read and write traffic. Use this mode whenever stale data would be unacceptable, such as in financial applications or user-facing dashboards.

primaryPreferred reads from the primary when it is available but falls back to a secondary if the primary is unreachable. This provides a balance: you normally get fresh data, but the application degrades gracefully during primary elections. Be aware that during the fallback, you may read data that is a few seconds behind.

secondary sends all reads to available secondaries, never touching the primary. This is ideal for analytics queries, batch jobs, or reporting that can tolerate eventual consistency. It offloads significant read pressure from the primary. If no secondary is available, the operation fails — it will NOT fall back to the primary.

secondaryPreferred prefers secondaries but falls back to the primary if no secondary is available. This is the most commonly used non-default mode — it spreads read load across secondaries under normal conditions while remaining available when a secondary goes down. The caveat is the same as secondary: reads may be slightly stale.

nearest sends reads to the member with the lowest measured network round-trip time, regardless of whether it is primary or secondary. This is perfect for globally distributed applications where users in different regions should read from the closest data center. Note that different users may read from different members, so they may temporarily see different data states.

Tag sets let you label replica set members (e.g., { region: 'us-east' }, { purpose: 'analytics' }) and then write read preferences that target only matching members. This enables sophisticated routing: send application reads to the nearest region and analytics queries to a dedicated secondary, all within the same replica set.

When reading from secondaries, you may encounter stale reads — the secondary hasn't replicated the latest writes from the primary yet. The staleness depends on replication lag. MongoDB 3.6 introduced maxStalenessSeconds: the driver refuses to use a secondary lagging more than the specified number of seconds, preventing very stale reads.

You can set the default read preference for all operations in the connection string using the readPreference parameter. This is the easiest way to configure the default without changing every call site. Individual operations can still override it by passing a readPreference option explicitly.

Multi-document transactions must always use primary read preference. Operations inside a transaction that specify a different read preference will throw an error. This ensures that all reads within the same transaction see a consistent snapshot as of the transaction's start time on the primary.

Use this guide: Real-time user data (cart, profile, notifications) → primary. Product catalog, content → secondaryPreferred. Reports and analytics → secondary with a dedicated analytics member. Global CDN-like reads → nearest. Always measure replication lag before routing critical reads to secondaries.

Test your understanding of MongoDB & NoSQL Databases concepts from this lesson.

In this lesson you learned: read preferences control which replica set member handles read operations, secondary and secondaryPreferred offload read traffic at the cost of eventual consistency, and nearest minimises latency in geo-distributed deployments. Next up we explore sharding — MongoDB's strategy for scaling beyond a single replica set.