The two Parquet files (legacy_nested.parquet and modern_nested.parquet) were generated
using the kite script located here:
testdata/src/main/java/org/apache/impala/datagenerator/JsonToParquetConverter.java

The Parquet files can be regenerated by running the following commands in the testdata
directory:

mvn package

mvn exec:java \
  -Dexec.mainClass="org.apache.impala.datagenerator.JsonToParquetConverter" \
  -Dexec.args="--legacy_collection_format
  data/schemas/nested/nested.avsc
  data/schemas/nested/nested.json
  data/schemas/nested/legacy_nested.parquet"

mvn exec:java \
  -Dexec.mainClass="org.apache.impala.datagenerator.JsonToParquetConverter" \
  -Dexec.args="
  data/schemas/nested/nested.avsc
  data/schemas/nested/nested.json
  data/schemas/nested/modern_nested.parquet"

The script takes an Avro schema and a JSON file with data and creates a Parquet file.
The --legacy_collection_format flag makes the script output a Parquet file that uses the
legacy two-level format for nested types, rather than the modern three-level format.

More information about the Parquet nested types format can be found here:
https://github.com/apache/parquet-format/blob/master/LogicalTypes.md