Uploaded image for project: 'Beam'
  1. Beam
  2. BEAM-11231

Better support for Hadoop native libs and compression codecs in Java Dataflow batch worker

    XMLWordPrintableJSON

Details

    Description

      Current(as of Beam SDK 2.24.0) Java Dataflow workers does not ship with Hadoop native library and popular native compression codecs(e.g libsnappy) .

      If one try to read files using IO classes that assume native library support(e.g, org.apache.hadoop.hbase.mapreduce.TableSnapshotInputFormat) on GCP Dataflow, the following error will be throw if the file is compressed by snappy:

       

      
      java.io.IOException: Failed to start reading from source: org.apache.beam.sdk.io.hadoop.format.HadoopFormatIO$HadoopInputFor
       matBoundedSource@311fa603
      
      ...
      
      Caused by: org.apache.hadoop.hbase.DoNotRetryIOException: java.lang.RuntimeException: native snappy library not available: this version of libhadoop was built without snappy support.
        
      

      There is very little one can do from within pipeline job code unless rewriting the IO class, so it might make sense to include these native libraries for Hadoop IOs

      Attachments

        Activity

          People

            Unassigned Unassigned
            lobatt Cheng Li
            Votes:
            0 Vote for this issue
            Watchers:
            3 Start watching this issue

            Dates

              Created:
              Updated: