Uploaded image for project: 'Apache Drill'
  1. Apache Drill
  2. DRILL-4977

Reading parquet metadata cache from S3 with fadvise=random and Hadoop 3 generates a large number of requests

    XMLWordPrintableJSON

    Details

    • Type: Improvement
    • Status: Open
    • Priority: Major
    • Resolution: Unresolved
    • Affects Version/s: 1.8.0
    • Fix Version/s: None
    • Component/s: Storage - Parquet
    • Labels:
      None
    • Environment:

      Hadoop 3.0

      Description

      When using the new fs.s3a.experimental.input.fadvise=random mode for accessing Parquet files stored in S3, we see a significant improvement for the query performance but a slowdown on query planning. This is due to the way the metadata file is read (each chunk of 8000 bytes generates a new GET request to S3). Indicating with FSDataInputStream.setReadahead(metadata-filesize) that we will read the whole file, this behaviour is circumvented.

        Attachments

          Issue Links

            Activity

              People

              • Assignee:
                Unassigned
                Reporter:
                uwe Uwe Korn
              • Votes:
                0 Vote for this issue
                Watchers:
                1 Start watching this issue

                Dates

                • Created:
                  Updated: