apache / apache/parquet-java

Pure Java NIO LocalInputFile Implementation

Offen
#3,028 1 Kommentar 3 Reaktionen 0 zugewiesene Personen Auf GitHub ansehen
Type: enhancement
Vorherrschende Sprache
Java
Sterne
3.1k
Forks
1.6k
Ø Merge
3 T. 12 Std.
Gemergte PRs (30 T.)
33

Beschreibung

### Describe the enhancement requested

Hi there!

In an effort to reduce our coupling to Hadoop, and slim down our shaded Jar sizes, we transitioned all our Parquet reading / writing to using the non-Hadoop public interfaces. We leveraged Java NIO interfaces for reading / writing I/O, as detailed in this blog post here: https://blakesmith.me/2024/10/05/how-to-use-parquet-java-without-hadoop.html

The problem we hit: The current `LocalInputFile` implementation [falls back to Path#toFile and RandomAccessFile](https://github.com/apache/parquet-java/blob/master/parquet-common/src/main/java/org/apache/parquet/io/LocalInputFile.java#L42) calls for I/O operations. Once we plug in other `java.nio.file.FileSystem` implementations ([Such as using AWS' Java NIO SPI implementation](https://github.com/awslabs/aws-java-nio-spi-for-s3/)), the `LocalInputFile` implementation doesn't work anymore.

We've created our own implementation of `LocalInputFile` called `NioInputFile`, that implements the interface, but exclusively using Java NIO operations, instead of falling back to older Java file APIs. You can see the code in the above linked blog post, and also in this gist here: https://gist.github.com/blakesmith/5c08a2504c08c96f9fb62658ec6c9b28

A few questions, before I start sending patches:

1. Is this alternative implementation something that folks are interested in incorporating upstream into Parquet itself?
2. If so, would we prefer to update `LocalInputFile`, or add `NioInputFile` as another implementation of `LocalInputFile`? Note that in our case, we're doing buffered I/O to S3 with this implementation, so it's not technically a "local" file, either way, just one that talks to the Java NIO abstractions.

Of note is that `LocalOutputFile` is working just fine with Java NIO operations, even when we're doing remote S3 I/O.

Thanks you so much! Parquet has been transformative for us, and we're excited to help drive more adoption however we can.

### Component(s)

Core

Beitragsleitfaden

Für dieses Repository ist kein Beitragsleitfaden indexiert

Rechercherichtung

Read parquet-common/src/main/java/org/apache/parquet/io/LocalInputFile.java and compare it with the NioInputFile implementation linked in the issue. Check how LocalOutputFile uses Java NIO, then clarify whether completion means updating LocalInputFile or adding a separate implementation that works with alternate FileSystem providers.

Vom Indexierungsmodell aus dem Issue-Text verfasst.

Bewertung

Tech-Stack
java
Bereich
backend
Issue-Typ
Feature
Schwierigkeit
4/5
Geschätzter Aufwand
3-5 Tage
Aktivitätsstatus
Veraltet
Klarheit
Größtenteils klar
Anfängerfreundlichkeit
35/100

Neue Issues direkt in Ihr Postfach

Eine kurze Übersicht über anfängerfreundliche GitHub-Issues.