客户端请求 → 写入Primary Shard → 同步Replica Shard → 写Translog(持久化)→ 写入内存Buffer → Refresh生成Segment(内存)→ Flush将Segment刷到磁盘+清空Translog

单集群理论峰值写入(doc/s) ≈ 
   (节点数 × 单节点极限 ≈120k) × refresh 关闭倍数(3~8倍) × bulk 优化系数(1.2~1.5)

要批量写入 1000~5000 条,数据量控制在

Elasticsearch 写入没有硬上限,但实际生产中:

资料

https://cloud.tencent.com/document/product/845/55213

https://help.aliyun.com/zh/es/product-overview/4-core-16-gib-data-node-instance-performance-test?spm=a2c4g.11186623.help-menu-57736.d_0_0_9_0.40a871862F9LF1&scm=20140722.H_2630215._.OR_help-T_cn~zh-V_1