Replace parquet-tools with parquet-avro

[sekikn]

The newer version of parquet-tools is not available via Maven Central as it was deprecated in Parquet 1.12.0 and removed from source in 1.12.3 (PARQUET-2020). This PR replaces it with another module so that we can catch up with the recent version of Parquet later.

[civitaspo]

I'm so sorry for the sooooooo late response. I understand the background to change the 'parquet-tools' but I cannot understand the 'parquet-avro' dependency is required. Please explain more.

[sekikn]

Thank you for the comment.

I cannot understand the 'parquet-avro' dependency is required. Please explain more.

Parquet itself doesn't provide native in-memory object model, except for the org.apache.parquet.example.data and org.apache.parquet.tools.read packages. The former is not for production use as its name suggests, and the latter is deprecated since v1.12.0 as described above.

So when handling Parquet records as Java objects programmatically, we have the following three options generally (of course, we can also define new object model ourselves, but I'd like to avoid it due to its maintenance effort):

• parquet-avro: mapping Parquet record to Avro's GenericRecord
• parquet-protobuf: mapping Parquet record to Protobuf's Message
• parquet-thrift: mapping Parquet record to Thrift's TBase

I chose parquet-avro here simply because I'm more familiar with it than others. But if you prefer Protobuf or Thrift, I can revise the PR to use either of them instead of Avro.

[civitaspo]

I see. Thanks for the explanation. I prefer protobuf, but the current usecase is for tests, so I don't care.

Thank you for your contribution.