apache/hudi

Persist the HoodieIndex type in the hoodie.properties file #409

Open

#14,425 创建于 2025年11月30日

在 GitHub 查看
 (1 评论) (0 反应) (0 负责人)Java (2,431 fork)batch import
area:storagefrom-jiragood first issuehelp-wantedpriority:highstatus:pr-availabletype:bug

仓库指标

Star
 (4,823 star)
PR 合并指标
 (平均合并 16天 21小时) (30 天内合并 127 个 PR)

描述

https://github.com/uber/hudi/issues/409

JIRA info


Comments

07/Aug/19 11:18;taherk77;[~vinoth] Can you please give some background on this? What is required to be done here?;;;


07/Aug/19 11:34;vinoth;We write some one time configs into hoodie.properties during dataset creation time. Index type is not one of these atm.

This issue's goal is to first understand if it makes sense for the index to be not changeable after dataset creation. If yes, then make changes to enforce that. ;;;


07/Aug/19 11:35;vinoth;You can imagine for e.g, if you used BloomIndex all along to write data and suddenly switched to HBaseIndex in a subsequent write, it will introduce duplicates, since HBase will be empty and thus treat everything as inserts (even though looking up through BloomIndex would have revealed the updates) ;;;


07/Aug/19 12:45;taherk77;In log file that we write can we also write the index used in the footer? So that in such scenarios we can just match the 2 index and throw an exception. What do you think?;;;

贡献者指南