Parquet files keep data in contiguous chunks by column, appending new records to a dataset requires rewriting substantial portions of existing a file or buffering records to create a new file.
The JIRA proposes to add the possibility to import an individual table from a RDBMS into HDFS as a set of Parquet files. We will also provide a command-line interface with a new argument --as-parquetfile
sqoop import --connect JDBC_URI --table TABLE --as-parquetfile --target-dir /path/to/files
The major items are listed as follows:
- Implement ParquetImportMapper
- Hook up the ParquetOutputFormat and ParquetImportMapper in the import job.
- Be able to support import from scratch or in append mode
Note that as Parquet is a columnar storage format, it doesn't make sense to write to it directly from record-based tools. So we'd consider to use Kite SDK to simplify the handling of Parquet specific things.