内核 一个 Doris 干掉 ES+ClickHouse+HBase
- 一句话
- 同一引擎同时扛起亚秒 OLAP 分析和高并发点查/检索服务——从"ES 管搜、CK 管分析、HBase 管 KV"三件套收敛到一套系统。One engine handles both sub-second OLAP analytics and high-concurrency point lookup/search serving — collapsing the "Elasticsearch for search, ClickHouse for analytics, HBase for KV" trio into one system.
- 窄场景
- 用户行为/广告/音乐标签分析:既要复杂 OLAP(漏斗、留存、多维下钻),又要面向用户的毫秒级点查/关键词检索;希望"少运维一套系统"的中厂数据团队。User-behavior/advertising/music-tag analytics that need both complex OLAP (funnels, retention, multi-dimensional drill-down) and user-facing millisecond point lookups/keyword search; mid-size data teams that want "one fewer system to operate".
- 机制
- "列存 MPP + 倒排索引 + 主键点查优化"的缝合体:列式存储 + 向量化执行做 OLAP;inverted index 做关键词/标签过滤(替代 ES 的倒排);Unique Key 模型 + 点查短路径(绕开 MPP 调度)做高并发 KV 式读取。Kwai 淘汰 ClickHouse 的直接原因是"ClickHouse 不支持 unique key 更新"。A hybrid of "columnar MPP + inverted index + primary-key point-lookup optimization": columnar storage + vectorized execution for OLAP; inverted indexes for keyword/tag filtering (replacing Elasticsearch's inverted indexes); Unique Key model + point-query short paths (bypassing MPP scheduling) for high-concurrency KV-style reads. Kwai eliminated ClickHouse for one direct reason: "ClickHouse does not support unique key updates."
- 生产验证
- Kwai(快手,4 亿 DAU):ClickHouse+Elasticsearch→Doris,延迟降 64%-90%,写入吞吐 3 倍,单表实时写入峰值 300 万行/秒/节点(https://medium.com/@VeloDB_poweredby_ApacheDoris/from-clickhouse-elasticsearch-to-apache-doris-how-kwai-unified-trillion-scale-ad-analytics-31528f41513d);NetEase Games:Elasticsearch+HBase+ClickHouse→统一 Doris lakehouse,"查询性能与 CK 相当,但显著更易运维"(https://medium.com/@VeloDB_poweredby_ApacheDoris/netease-games-from-elasticsearch-hbase-and-clickhouse-to-a-unified-apache-doris-lakehouse-686362fa1bc1);Tencent Music(8 亿 MAU):ES→Doris 做统一检索引擎,存储成本降 80%(https://github.com/apache/doris-website/blob/HEAD/blog/tencent-music-migrate-elasticsearch-to-doris.md)。诚实备注:案例多经 VeloDB/社区官方渠道发布,已降档;数字未经独立复现;缺独立英文第三方复盘。Kwai (400M DAU): ClickHouse+Elasticsearch→Doris — latency down 64%-90%, 3x write throughput, single-table realtime peak 3M rows/sec/node (https://medium.com/@VeloDB_poweredby_ApacheDoris/from-clickhouse-elasticsearch-to-apache-doris-how-kwai-unified-trillion-scale-ad-analytics-31528f41513d); NetEase Games: Elasticsearch+HBase+ClickHouse→unified Doris lakehouse — "query performance comparable to ClickHouse, but significantly easier to operate" (https://medium.com/@VeloDB_poweredby_ApacheDoris/netease-games-from-elasticsearch-hbase-and-clickhouse-to-a-unified-apache-doris-lakehouse-686362fa1bc1); Tencent Music (800M MAU): Elasticsearch→Doris as unified search engine — storage cost down 80% (https://github.com/apache/doris-website/blob/HEAD/blog/tencent-music-migrate-elasticsearch-to-doris.md). Honest note: cases mostly arrived via VeloDB/community-official channels — downgraded; figures not independently reproduced; no independent English-language third-party postmortems found.
- 竞品差距
- ClickHouse 无主键行级更新、运维门槛高(但纯 OLAP 扫描仍略强);StarRocks 定位最接近(同样 MySQL 协议、主键更新),差异在生态侧——Doris 的 ES 替代故事(倒排索引+点查)在社区案例中更成体系。"替代 ES+CK 双栈"的完整证据链,31 款中只有 Doris 拿得出来。ClickHouse has no primary-key row-level updates and higher ops burden (though still slightly stronger on pure scan); StarRocks is the closest rival (also MySQL protocol, primary-key updates) — the difference is ecosystem-side: Doris's Elasticsearch-replacement story (inverted index + point lookup) is the more complete case chain. The full "replacing the ES+ClickHouse dual stack" evidence chain belongs to Doris alone among the 31.
- 证据等级
- 社区官方/厂商生态渠道(已降档标注)+ 机制层面的竞品淘汰原因;缺少独立英文第三方复盘,数字未经独立复现Community-official/vendor-ecosystem channels (downgraded, as noted) + mechanistic elimination reasons; no independent English third-party postmortems; figures not independently reproduced
- 最后核验
- 2026-10-012026-10-01