​
Overview

This version of a COPY FROM statement, allows you to specify the imported file format and currently three types are supported.
With each file type there will be differences in performance and behavior
Here’s a list of supported formats:
  • CSV (comma-separated values): simple columnar text format
  • ORC (optimized row columnar): columnar storage format developed by Apache
  • Parquet (Apache Parquet): columnar data storage used in Apache Hadoop ecosystem
Each query without a specified file format assumes to be importing a CSV file (There is no format detection in place)

​
Syntax

In order to sepcify the file format using the COPY FROM statement you can use the following syntax: 
COPY tablename FROM 'file_path' (FORMAT format_name);
Format name is case insensitive

​
Examples

When copying from the CSV, ORC or Parquet formats, as the first step you need to create a destination table: 
CREATE TABLE cab_types (id bigint, cab_type text);
Once that is done, you can copy the file content into the table by following one of the examples below.

​
ORC

Copying a dataset from an ORC format file is only supported if the ORC file resides in an object storage solution, such as an S3 bucket. 
copy cab_types from 's3://oxla-testdata/cab_types.orc' (format orc, HEADER);
  • Use HEADER if your ORC file includes column headers

​
Parquet

Copying a dataset from a parquet format file is only supported if the parquet file resides in an object storage solution, such as an S3 bucket. 
copy cab_types from 's3://oxla-testdata/cab_types.parquet' (format parquet);
Copying from Parquet files is memory consuming. Files bigger than a few gigabytes might result in out of memory error

​
CSV

When it comes to CSV files, we have a few cases here:
  • CSV
copy cab_types from 's3://oxla-testdata/cab_types.csv' (format csv);
  • CSV with Specified Delimiter
copy cab_types from 's3://oxla-testdata/cab_types.csv' (format csv, delimiter ':');
  • CSV skipping invalid rows
copy cab_types from 's3://oxla-testdata/cab_types.csv' (format csv, on_error stop|ignore);
By default invalid rows stops data ingestion and Oxla returns an error. Setting on_error action to ignore enforce further processing after skipping invalid rows.

​
Differences in Behavior

  • Ignored Options
    • HEADER, DELIMITER, NULL, ON_ERROR options are ignored not affecting the execution of the queries for formats different than CSV
  • Null Values Handling
    • All ORC files have nullable columns. In order to import a nullable column to an Oxla column, which are described as NOT NULL, the column in the ORC file cannot contain a null value, otherwise the request will be terminated.
    • For Parquet files, inserting a nullable column to a non nullable column is allowed as long as there are no null values in the source column.
    • When using ON_ERROR option for CSV files, null constraint violation causes row skipping
  • Column Matching
    • ORC and Parquet files are only being matched based on column index, while CSVs can be matched both with names or indexes. For more information regarding that, please refer to our COPY FROM with HEADER doc