CometShuffleManager treats Spark's own shuffles differently from SortShuffleManager
维护者通常 2 天内回复
还没有人认领这个 Issue。
评估
- 难度
- 4/5
- 预计耗时
- 3-5 天
- 新手友好度
- 55/100
- Issue 类型
- 缺陷
- 描述清晰度
- 描述清楚
- 活跃度
- 活跃
- 技术栈
- scala, spark
- 领域
- backend, performance
调研方向
Read spark/src/main/scala/org/apache/spark/sql/comet/execution/shuffle/CometShuffleManager.scala and trace how getWriter, unregisterShuffle, and the IndexShuffleBlockResolver share shuffle state. Check the Spark exchange copy behavior described in the issue and compare cleanup with SortShuffleManager. Done when all three reported differences are addressed; the issue gives no named tests or reproduction.
由索引模型根据 Issue 内容生成。
描述
Describe the bug
CometShuffleManager handles shuffles that Comet does not replace differently from Spark's SortShuffleManager. These shuffles are RDD shuffles and Spark exchanges Comet leaves in place. They are not new, but they reach more applications once the plugin sets CometShuffleManager by default on Spark 4.0 and later (#7336). Found in review of that PR.
- Extra row copy in Spark exchanges.
ShuffleExchangeExec.needToCopyObjectsBeforeShufflereturns true for any manager that is not aSortShuffleManager, so Spark's own exchanges copy every row before the shuffle. sortShuffleManager.getWriterentries are never released. For a Spark shuffle dependency,getWriterdelegates to the innersortShuffleManager, which records each map task in its owntaskIdMapsForShuffle.unregisterShuffleclears onlyCometShuffleManager's map and never callssortShuffleManager.unregisterShuffle, so those entries stay on the executor heap until the application ends.- Blocks migrated in during decommissioning are never deleted. The
IndexShuffleBlockResolveris built with a freshConcurrentHashMapinstead oftaskIdMapsForShuffle. The resolver records blocks migrated in during decommissioning in that map, sounregisterShufflenever deletes them.
All three are in spark/src/main/scala/org/apache/spark/sql/comet/execution/shuffle/CometShuffleManager.scala.
Steps to reproduce
Read the code paths above. No reproduction has been run for any of them.
Expected behavior
A shuffle that Comet does not replace behaves as it does under SortShuffleManager: no extra row copy, and no state left behind after unregisterShuffle.
- 主要语言
- Scala
- 星标
- 1.3k
- 派生
- 414
- 平均合并
- 3 天 5 小时
- 30 天内合并 PR
- 210
环境准备
- 没有 Dockerfile 或 Docker Compose 文件
- 有 Pull Request 模板
- 阅读贡献指南
从这里开始
- 先读完整个 Issue，再读项目的贡献指南。
- 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
- Fork 仓库，在一个分支上完成修改。
- 提交 Pull Request，并在描述里引用这个 Issue 编号。
apache/datafusion-comet 的其他 Issue
-
requires-triage
难度 2/5 1-3 小时 新手友好度 72/100
apache/datafusion-comet#7583 ·
维护者通常 2 天内回复
-
bug priority:medium
难度 2/5 1-3 小时 新手友好度 73/100
apache/datafusion-comet#7559 · 1 条评论 ·
维护者通常 2 天内回复
-
area:writer bug priority:low
难度 2/5 1-3 小时 新手友好度 78/100
apache/datafusion-comet#7558 ·
维护者通常 2 天内回复
-
enhancement
难度 2/5 1-3 小时 新手友好度 78/100
apache/datafusion-comet#7557 ·
维护者通常 2 天内回复
-
`rand` and `randn` with an integer seed fall back to Spark without a fallback reason可能已有人在做 @galaxytemple 于 3 天前认领。 未关闭requires-triage
难度 2/5 1-3 小时 新手友好度 78/100
apache/datafusion-comet#7479 · 1 条评论 · 已指派 1 人 ·
维护者通常 2 天内回复
查看 apache/datafusion-comet 的全部 Issue
相似的 Issue
-
A-api C-bug S-ready-for-human
难度 2/5 1-3 小时 新手友好度 70/100
ergoplatform/ergo#2932 ·
维护者通常 3 天内回复
-
难度 2/5 1-3 小时 新手友好度 68/100
lichess-org/lila#22454 ·
维护者通常 1 天内回复
-
难度 2/5 1-3 小时 新手友好度 76/100
chipsalliance/rocket-chip#3845 ·
-
难度 2/5 1-3 小时 新手友好度 76/100
-
难度 1/5 1 小时以内 新手友好度 86/100
维护者通常 13 天内回复