Uploaded image for project: 'Spark'
  1. Spark
  2. SPARK-19256 Hive bucketing write support
  3. SPARK-33298

Introduce new API to FileCommitProtocol allow flexible file naming

    XMLWordPrintableJSON

Details

    • Sub-task
    • Status: Resolved
    • Minor
    • Resolution: Fixed
    • 3.1.0
    • 3.2.0
    • Spark Core
    • None

    Description

      This Jira is to propose a new version for `FileCommitProtocol` (https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/internal/io/FileCommitProtocol.scala ), e.g. `FileCommitProtocolV2`.

      The motivation is currently we have two requirements to change the API for FileCommitProtocol:

      (1).Support write Hive ORC/Parquet bucketed table (https://github.com/apache/spark/pull/30003 ): need to add new parameter `prefix` into method `newTaskTempFile` and `newTaskTempFileAbsPath`, to allow spark writes hive/presto-compatible bucketed files.

      (2).Fix commit collision in dynamic partition overwrite mode (https://github.com/apache/spark/pull/29000 ): need to add new method `getStagingDir` to allow customize dynamic partition staging directory to avoid commit collision.

       

      The reason to propose FileCommitProtocolV2 instead of changing `FileCommitProtocol` directly, is that the API for FileCommitProtocolV2 is kind of public where we allow customized commit protocol subclass to use during run-time (https://github.com/apache/spark/blob/master/core/src/main/scala/org/apache/spark/internal/io/FileCommitProtocol.scala#L146 ). So if we change the API (e.g. adding method, or changing existing method signature), it will break external subclass for the commit protocol. And we are aware of some of external subclasses for better support of object store, according to cloud_fan .

       

      One proposal for `FileCommitProtocolV2` can be:

      abstract class FileCommitProtocolV2 {
        // `options` to replace `ext`, where we can put more string-string parameters
        def newTaskTempFile(taskContext: TaskAttemptContext, dir: Option[String],
      options: Map[String, String]): String
       
        // `options` to replace `ext`, where we can put more string-string parameters
        def newTaskTempFileAbsPath(
          taskContext: TaskAttemptContext, absoluteDir: String, options: Map[String, String]): String
      
        // other new methods, e.g. getStagingDir
        def getStagingDir(path: String, jobId: String): Path
      
        // rest of FileCommitProtocol methods
        ...
      }
      

       

      FileCommitProtocolV2.instantiate() logic will first try to find a subclass for `FileCommitProtocolV2`, if not will find a subclass for `FileCommitProtocol`, so the current version of `FileCommitProtocol` is still supported.

      Attachments

        Activity

          People

            chengsu Cheng Su
            chengsu Cheng Su
            Votes:
            1 Vote for this issue
            Watchers:
            10 Start watching this issue

            Dates

              Created:
              Updated:
              Resolved: