ElasticSearch

• ElasticSearch

  • Use cases

    • Offload read requests (not requiring realtime) from MySQL

      • Resiliency

      • Latency

  • Comparison - Lucene vs Solr vs ElasticSearch

ElasticSearch

Use cases

Offload read requests (not requiring realtime) from MySQL

• Why?

  *

• Approach: Option 1 vs Option 2 below

  • Option 2 is adopted because Option 1 needs to introduce additional components and does not improve much on latency

┌───────────────┐                                                                  ┌──────────────┐
│               │         Option1: ES monitor MySQL's binlog and put binlog        │              │
│               │                          data inside ES                          │              │
│     MySQL     ├─────────────────────────────────────────────────────────────────▶│ElasticSearch │
│               │              Option2: Write data to ES using ES APIs             │              │
│               │                                                                  │              │
└───────────────┘                                                                  └──────────────┘

Resiliency

• What if there is an error using Option 2: Add another task inside SQL server. There will be a worker task regularly scanning through these tasks. Use this type of compensation mechanism to make ES and MySQL data eventual consistent.

Latency

• There will be a delay when data comes into ES's index buffer but not into file system cache. So for requests requiring real time data, better rely on MySQL.

ElasticSearch

• Reference:

  • In Chinese: ES在京东订单中心的应用

Comparison - Lucene vs Solr vs ElasticSearch

• Reference [In Chinese: https://zhuanlan.zhihu.com/p/161645496]