After reading some pages and slides about Parquet, I get some questions about defining the schema:

1. Do I need to define the schema directly?
2. Should I define the schema directly?
3. Can I define the schema directly?

Parquet documentation did not give definition language or any demos.

Viewed a lot of pages and founding them all get Parquet schema from other format’s schema, such as JSON, AVRO. Here’s a well-known example: Spark SQL load and write parquet

Parquet is oriented from Dremel paper, which described the record shredding and assembly algorithm. Below is my understanding of it, please be skeptical.

Parquet is the format of columnar representation of a nested record, its schema depends on and oriented from the record’s schema. So, the answers to the above questions are all “NO”.