opensearch-project/OpenSearch

[Remote Store] Evaluate safely skipping segment_n file uploads after flush for remote store backed indices

Open

#18,239 创建于 2025年5月8日

在 GitHub 查看
 (2 评论) (0 反应) (0 负责人)Java (8,123 star) (1,505 fork)batch import
Storage:Remoteenhancementgood first issuelucene

描述

Is your feature request related to a problem? Please describe

In remote-backed storage indices, OpenSearch provides refresh-level durability by uploading segment metadata on each refresh. This metadata contains information about segments and userdata (such as local checkpoint and max seq no) for that refresh. Currently, we also upload the segment_n file to remote storage when Lucene creates it during each flush

During recovery and replication, we don't use the segment_n file information Reference: IndexShard.java#L5174

Instead, we use the infobytes from remote segment metadata to create another commit on replica. Reference: IndexShard.java#L5202.

Describe the solution you'd like

Proposed Change

Evaluate if we can safely skip uploading the segment_n file to remote storage, as it appears to be unused in recovery and replication scenarios.

Investigation Points

  1. Identify potential risks or edge cases where segment_n file might be needed
  2. Analyze impact on:
    • Recovery scenarios
    • Replication processes

Related component

No response

Describe alternatives you've considered

No response

Additional context

No response

贡献者指南

技术栈
java
领域
databasebackend
议题类型
feature
难度面向新贡献者的预计实现难度，1 表示很小改动，5 表示专家级工作。
4
预计时间有经验贡献者完成调查、实现、测试并准备 pull request 的粗略时间范围。
3-5 days
活动状态议题当前的可参与程度：新鲜、活跃、陈旧、阻塞或等待维护者输入。
fresh
清晰度议题是否清楚说明期望改动、验收标准和下一步。
mostly clear
前置要求
OpenSearch remote storeJavaLucene segmentsIndexShard recovery
新手友好度1-100 的估计分数，表示该议题对首次贡献者的友好程度。
35
研究方向
Investigate if the segment n file can be safely skipped during upload after flush. Focus on files like IndexShard.java (lines 5174 and 5202) to understand how segment metadata is used in recovery and replication. Check if any other components rely on segment n, such as during snapshot or rollback. Review the remote segment metadata upload process and identify potential edge cases (e.g., full cluster restart, primary relocation). Also examine whether skipping the upload would break any existing tests or behavior. Maintainers have not clarified scope; ask about expected test coverage and performance impact.