4paradigm / 4paradigm/OpenMLDB
batch: bug in handle key if is null
- 主要语言
- C++
- 星标
- 1.7k
- 派生
- 331
- 平均合并
- 12 天 12 小时
- 30 天内合并 PR
- 1
描述
https://github.com/4paradigm/OpenMLDB/blob/908e9edbe05a86dccdbf8d1579bad88454480d59/java/openmldb-batch/src/main/scala/com/_4paradigm/openmldb/batch/nodes/WindowAggPlan.scala#L424-L425
We get key by `extractKey`, and then check it by `isValidOrder`.
After #1408 fixed, `isValidOrder` arg is java Long, it may be null or a long.
But `extractKey` returns scala Long, which can't be null, ref
https://github.com/4paradigm/OpenMLDB/blob/8a20b4849c52dfee7dacd02f04a6adede05673c2/java/openmldb-batch/src/main/scala/com/_4paradigm/openmldb/batch/window/WindowComputer.scala#L225-L227
The function `getLongFromIndex` called by `extractKey`, returns java Long.
https://github.com/4paradigm/OpenMLDB/blob/2e7fb76aca6a41a96c43b46b1be2194281c2166f/java/openmldb-batch/src/main/scala/com/_4paradigm/openmldb/batch/utils/SparkRowUtil.scala#L66
We miss null in `extractKey`.
And after I temporarily make `extractKey` return java Long, some batch tests failed. e.g. https://github.com/4paradigm/OpenMLDB/blob/836b34119249ff9b667130b125c88395ef18ee4f/java/openmldb-batch/src/test/scala/com/_4paradigm/openmldb/batch/nulldata/TestWindowWithNullData.scala#L28
reproduce:
1. make `extractKey` return java Long
2. run TestWindowWithNullData
贡献指南
调研方向
The bug is in WindowAggPlan.scala line 424-425 where extractKey returns a Scala Long (non-nullable) but isValidOrder expects a Java Long (nullable). Examine extractKey in WindowComputer.scala and getLongFromIndex in SparkRowUtil.scala. Run TestWindowWithNullData to reproduce the failure after modifying extractKey to return java.lang.Long. The fix must handle nulls correctly without breaking existing batch tests.
由索引模型根据 Issue 内容生成。
评估
- 技术栈
- java, scala
- 领域
- databases, machine-learning
- Issue 类型
- 缺陷
- 难度
- 3/5
- 预计耗时
- 1-2 天
- 活跃度
- 停滞
- 描述清晰度
- 描述清楚
- 新手友好度
- 45/100