apache / apache/arrow-java

VectorAppender fails with OversizedAllocationException when appending to empty UnionVector

未关闭
#812 0 条评论 3 个 reaction 已指派 0 人 在 GitHub 查看
Type: bug
主要语言
Java
星标
94
派生
152
平均合并
3 天 16 小时
30 天内合并 PR
11

描述

### Describe the bug, including details regarding any error messages, version, and platform.

**Description:**
When appending a `UnionVector` to an empty target `UnionVector`, [this](https://github.com/apache/arrow-java/blob/main/vector/src/main/java/org/apache/arrow/vector/util/VectorAppender.java#L556) while loop runs infinitely until the exception occurs:
```
org.apache.arrow.vector.util.OversizedAllocationException: Unable to expand the buffer
```
It seems to be a issue with the `UnionVector.getValueCapacity()` method. Logging the capacity before and after calling `reAlloc()` shows that it's always `0`.
Debugging further, it seems like the value capacity of the `internalStruct` in the `UnionVector` always returns `0` because it has no children so [size](https://github.com/apache/arrow-java/blob/main/vector/src/main/java/org/apache/arrow/vector/complex/NonNullableStructVector.java#L373) is `0`.
```
@Override
public int getValueCapacity() {
return Math.min(getTypeBufferValueCapacity(), internalStruct.getValueCapacity());
}
```

**Steps to reproduce**
```
@Test
public void testAppendEmptyTargetUnionVector() {
final int length = 5;

try (final UnionVector target = UnionVector.empty("target", allocator);
final UnionVector delta = UnionVector.empty("delta", allocator)) {

// populate the delta vector
delta.setType(0, Types.MinorType.FLOAT4);
delta.setType(1, Types.MinorType.FLOAT4);
delta.setType(2, Types.MinorType.FLOAT4);
delta.setType(3, Types.MinorType.FLOAT4);
delta.setType(4, Types.MinorType.FLOAT4);

Float4Vector deltaFloatVector = delta.getFloat4Vector();
deltaFloatVector.allocateNew();
ValueVectorDataPopulator.setVector(deltaFloatVector, 1f, 2f, 3f, 4f, 5f);
assertEquals(length, deltaFloatVector.getValueCount());
delta.setValueCount(length);

VectorAppender appender = new VectorAppender(target);
delta.accept(appender, null);

assertEquals(length, target.getValueCount());

for (int i = 0; i < length; i++) {
Object floatObj = target.getObject(i);
assertTrue(floatObj instanceof Float);
assertEquals(i+1, ((Float) floatObj).intValue());
}
}
}
```
This issue does not occur if any value is set in the target vector before appending.

贡献指南

打开贡献指南

调研方向

从 VectorAppender.java 第556行附近的循环开始,检查 UnionVector.getValueCapacity(),包括 NonNullableStructVector.java 第373行附近的代码。运行提供的 testAppendEmptyTargetUnionVector 复现,并比较 reAlloc() 前后的容量。当向空的 UnionVector 追加数据时能够在没有 OversizedAllocationException 的情况下完成,并保留全部五个 float 值,即表示完成。

由索引模型根据 Issue 内容生成。

评估

技术栈
java
领域
data
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
55/100

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。