InternalParquetRecordWriter doesn't immediately limit current row group to threshold
- 主要语言
- Java
- 星标
- 3.1k
- 派生
- 1.6k
- 平均合并
- 3 天 12 小时
- 30 天内合并 PR
- 33
描述
The MemoryManager adjust the row group size threshold of writers when the allocated memory pool fills up.
**Problem**: However InternalParquetRecordWriter only re-adjusts the row group size on the next flush meaning they still use the old size.
This opens up a possibility of getting an OOM error if all writers are started at relatively the same time and progress in tandem(I saw this when investigating failing jobs while writing to disk in Spark)
**Reporter**: [Brian Mwambazi](https://issues.apache.org/jira/secure/ViewProfile.jspa?name=brimzi) / @brimzi
#### PRs and other links:
- [GitHub Pull Request #748](https://github.com/apache/parquet-mr/pull/748)
**Note**: *This issue was originally created as [PARQUET-1767](https://issues.apache.org/jira/browse/PARQUET-1767). Please see the [migration documentation](https://issues.apache.org/jira/browse/PARQUET-2502) for further details.*
贡献指南
这个仓库没有索引到贡献指南
调研方向
首先查看 GitHub Pull Request #748,以及 issue 中提到的入口点 InternalParquetRecordWriter 和 MemoryManager。确认 memory pool 填满时 row-group threshold 如何更新,并验证 writers 会立即停止使用旧 threshold,从而使并行 writers 不会有发生 OOM 的风险。
由索引模型根据 Issue 内容生成。
评估
- 技术栈
- java
- 领域
- data-engineering
- Issue 类型
- 缺陷
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 活跃度
- 停滞
- 描述清晰度
- 基本清楚
- 新手友好度
- 25/100