What is a replica in Elasticsearch?

In Elasticsearch, replicas are copies of index shards, primarily used to enhance system reliability and query performance.

Replicas' Roles

1. Fault Tolerance: If a node fails, replicas ensure data remains available. Since data is replicated across multiple nodes, Elasticsearch can recover from replicas when a node goes down.

2. Load Balancing: For read requests (such as searches or data retrieval), replicas distribute the load across different nodes, improving query response times. Write operations (like updates or adding documents) are still performed only on primary shards, but are later synchronized to replica shards.

Types of Replicas

• Primary Shard: The original shard of data, responsible for handling write operations.

• Replica Shard: An exact copy of the primary shard, used for handling read requests and providing data redundancy.

Example

Suppose there is an Elasticsearch index containing a large number of frequently queried documents. If this index is configured with only one primary shard and no replicas, when many users query, all read requests concentrate on this single shard, potentially slowing query speeds and affecting system stability.

To address this, multiple replica shards can be configured for the index. For example, setting two replica shards means each primary shard has two corresponding replicas, allowing read requests to be load-balanced across the primary and both replicas. This not only significantly improves query speed but also enhances data reliability, as data can be recovered from replicas even if a primary shard's node fails.

In summary, replicas are a key mechanism for ensuring high availability and high performance in the Elasticsearch system.