WRITE_NOS allows you to extract selected or all columns from a database table or from derived results and write to external object storage, such as Amazon S3, Azure Blob storage, Azure Data Lake Storage Gen2, and Google Cloud Storage.
WRITE_NOS stores data in Parquet format.
Data written to external object storage can be queried using a foreign table and READ_NOS.
You can use WRITE_NOS to transform CSV and JSON data without having to write it locally to a relational table. Read the CSV or JSON data using CREATE FOREIGN TABLE or READ_NOS, then use WRITE_NOS to write the data in Parquet format.
WRITE_NOS can write relational data to external object store, which you can partition in an optimal way for reading back based on expected query usage.
The maximum input data for both READ_NOS and WRITE_NOS is 16,776,192 bytes (including LOB and LOB-based UDT columns).
It is the responsibility of the user to clean up object store files on interrupted write operations. Write operations can be interrupted on transaction aborts or system resets, among other reasons.
Concurrent WRITE_NOS requests to the same location are likely to cause an error to be returned. If a request gets an error, any objects written by the request are not deleted from external storage and must be manually cleaned up using tools provided by the external object store vendor, such as s3cmd. To avoid this, Teradata recommends specifying a unique location for concurrent requests.
An error is reported if you attempt to write an object with same path name in the same location.