add support for converting traditional hive tables to iceberg/delta/hudi

[djouallah]

Feature Request / Improvement

there are a lot of systems that produce parquet files only, it will be useful if xtable can convert those parquet to modern tables formats without rewriting data just by adding metadata continuously.

Delta do that already but it is a one off operation and can't accept new files

Are you willing to submit PR?

• Yes I am willing to submit a PR!

Code of Conduct

• I agree to follow this project's Code of Conduct

[JDLongPLMR]

This seems like a pretty easy lift. There are a number of use cases where simply adding parquet files to the table would be handy.

[vinishjail97]

Yes this can be done, we need to implement a parquet source class which needs to do two things - retrieve snapshot and retrieve change log since lastSyncTime.

Using List files

1. List all parquet files in ObjectStorage or HDFS root path to retrieve the snapshot. This would be a simple list call.
2. Fetch the parquet files that have been added since last syncInstant to retrieve the change log. We can this via the same list call and filtering files based on their creationTime is the simplest way but it's expensive.

Using cloud notifications queue

1. The efficient way of doing this for object stores would be to setup the notifications queue for the bucket, consume and insert the file location, creationTime etc. to a key-value store or hudi/delta/iceberg table (let's call this events table). To handle duplicate notifications in the queue, the events table would have the parquet versioned file location as the primary key. If you are using hudi for events table, we don't need to write new code here, this step would be setting up cloud queue via terraform and starting a job which consumes from the queue. Ref: S3EventsSource GcsEventsSource SQS PubSub
2. XTable parquet source class would trigger an incremental query to the events table to get the new files that have been added since the lastSyncTime and generate hudi, iceberg and delta metadata for them.

The design is similar to what hudi does for ingesting large number of files, steps 7 and 8 in the architecture would become XTable sync.
https://hudi.apache.org/blog/2021/08/23/s3-events-source/

If you are using HDFS or object stores which don't support a queue based system for file notifications, we need to build/re-use existing queue implementation for file notifications.

[vinishjail97]

@djouallah @JDLongPLMR Let me know what you think of the two approaches, we can write this as utility tool in xtable-utilities similar to RunSync

https://github.com/apache/incubator-xtable/blob/main/xtable-utilities/src/main/java/org/apache/xtable/utilities/RunSync.java

[alberttwong]

I believe you can covert parquet to hudi files via hudi bootstraping (https://hudi.apache.org/docs/migration_guide). Once it's in hudi, you can apache xtable to other formats. Onehouse can do this automatically.

[djouallah]

@djouallah @JDLongPLMR Let me know what you think of the two approaches, we can write this as utility tool in xtable-utilities similar to RunSync

https://github.com/apache/incubator-xtable/blob/main/xtable-utilities/src/main/java/org/apache/xtable/utilities/RunSync.java

using listing files seems good for my use case

[vinishjail97]

@djouallah Yes listing will be sufficient for a small number of files, do you plan to submit a PR for this ? Let me know if you need any help regarding the PR.

[djouallah]

@vinishjail97 nah, I am just an end user of xtable :)

[vinishjail97]

okay, I will start a thread in dev mailing list to see if someone is interested to work on this feature.

[JDLongPLMR]

thanks all. Seems like a helpful addition

[sudharshanraja-db]

Hi @vinishjail97 if it is not assigned to anyone yet would like to explore and take up this feature