apache / apache/parquet-java

Avoid per-write virtual dispatch in `DictionaryValuesWriter.shouldFallBack()` by caching the size-exceeded check

Abierto
#3,501 0 comentarios 0 reacciones 0 asignados Ver en GitHub
Type: enhancement
Lenguaje dominante
Java
Estrellas
3.1k
Forks
1.6k
Merge medio
3 d 12 h
PR fusionados (30 d)
33

Descripción

### Describe the enhancement requested

`DictionaryValuesWriter.shouldFallBack()` is called by `FallbackValuesWriter.checkFallback()` after every single value write. The current implementation dispatches a virtual call to `getDictionarySize()` on every invocation:
```java
public boolean shouldFallBack() {
return dictionaryByteSize > maxDictionaryByteSize || getDictionarySize() > MAX_DICTIONARY_ENTRIES;
}
```
`getDictionarySize()` is an abstract method overridden in each typed subclass (Binary, Long, Double, Integer, Float) to return the backing map's `.size()`. Since `shouldFallBack()` is polled after every write, including writes of duplicate values that do not grow the dictionary, the virtual dispatch and map-size query are redundant work for the common case where most values are already in the dictionary.
Both `dictionaryByteSize` and the dictionary entry count can only increase when a new entry is added (inside the `if (id == -1)` branch of each subclass's write method). Therefore the size-exceeded condition can only transition from `false` to `true` at that exact point.

### Proposal
Replace the per-write check with a cached boolean `dictionarySizeExceeded` flag. Introduce a `checkDictionarySizeLimit(int newDictionarySize)` method that subclass write methods call only when a new dictionary entry is actually added. `shouldFallBack()` then returns the cached flag directly, a simple field read with no virtual dispatch.

### Component(s)

Core

Guía de contribución

No hay ninguna guía de contribución indexada para este repositorio

Línea de trabajo

Start with DictionaryValuesWriter.shouldFallBack() and FallbackValuesWriter.checkFallback(), then inspect the Binary, Long, Double, Integer, and Float typed subclass write methods, especially their id == -1 branches. Confirm that the size-exceeded state is updated only when a new dictionary entry is added and that fallback behavior remains unchanged for duplicate writes.

Escrito por el modelo de indexación a partir del texto del issue.

Evaluación

Stack tecnológico
java
Área
data-engineering
Tipo de issue
Refactorización
Dificultad
3/5
Tiempo estimado
1-2 días
Estado de actividad
Tranquilo
Claridad
Bien especificado
Aptitud para principiantes
76/100

Recibe los nuevos issues en tu correo

Un resumen breve de issues de GitHub para principiantes.