[Umbrella] Collect observation metrics at all stages of Spark SQL.

[wForget]

Code of Conduct

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

Search before asking

• I have searched in the issues and found no similar issues.

Describe the proposal

In the process of upgrading the spark version or introducing new optimizations, in order to ensure data consistency, we usually double-run SQL in multiple environments, and then compare the results to determine inconsistent behavior. But when the results are inconsistent, it is difficult for us to quickly find the stage where the inconsistency first occurred.

Spark provides Observation to allows inserting observers into dataframes to define and obtain observation metrics.

I want to insert CRC checksum metric observers at all stages of SQL so that inconsistent stages can be found quickly.

I made a simple implementation, like:

Task list

• [KYUUBI #6017] Support observe hint #6017
• Get results of SQL observations
• [Subtask] Inject crc checksum observers at all stages of SQL #6024
• [Subtask] Display observation metrics in Spark UI #6031

Are you willing to submit PR?

• Yes. I would be willing to submit a PR with guidance from the Kyuubi community to improve.
• No. I cannot submit a PR at this time.