FileIO.Write.relativeFileNaming should not be public
- Dominant language
- Java
- Stars
- 8.7k
- Forks
- 4.7k
- Avg merge
- 1d 20h
- Merged PRs (30d)
- 196
Description
I think the existing FileIO.writeDynamic is a bit easy to misuse, as something like this looks correct, and compiles:
` FileIO.writeDynamic()`
` .by(...)`
` .withNaming(new SerializableFunction[String, FileNaming] {`
` override def apply(str: String): FileNaming =`
` FileIO.Write.relativeFileNaming(`
` "some/directory",`
` new FileNaming {`
` override defFilename(window: BoundedWindow, pane: PaneInfo, numShards: Int, shardIndex: Int, compression: Compression): String = "some_filename.txt"```}
` .via(...)`
` .to("gs://some/bucket")`
However, for dynamic writes, if `outputDirectory` (.to("...")) is set, under the hood, Beam will wrap the provided `fileNamingFn` in `FileIO.Write.relativeFileNaming(...)` as well, so it ends up as a nested `relativeFileNaming` function. ([https://github.com/apache/beam/blob/da9e17288e8473925674a4691d9e86252e67d7d7/sdks/java/core/src/main/java/org/apache/beam/sdk/io/FileIO.java#L1243)](https://github.com/apache/beam/blob/da9e17288e8473925674a4691d9e86252e67d7d7/sdks/java/core/src/main/java/org/apache/beam/sdk/io/FileIO.java#L1243))]
IMO, `relativeFileNaming` should either be made private, so that it's only used internally by FileIO.Write, or a precondition should be added when a dynamic FileIO.Write is expanded, to check that `outputDirectory` can't be set if the provided `fileNamingFn` is relative.
wdyt?
Imported from Jira [BEAM-9316](https://issues.apache.org/jira/browse/BEAM-9316). Original Jira may contain additional context.
Reported by: clairemcginty.
Contributor guide
Assessment
This issue has not been assessed yet.