Union All optimisation

[ankitdixit]

Test plan - (Please fill in how you tested your changes)

== RELEASE NOTES ==

General Changes
* Improve performance for queries that union all many constants.

[linux-foundation-easycla]

The committers are authorized under a signed CLA.

• ✅ Ankit Dixit (6a201d8ebcdadb6de48722ae71a9ab97ec13f2a9)

[ajaygeorge]

Hi @ankitdixit Thanks for working on this. Can you please fill in the test plan and look into the travis build failures to see if they are related to your changes.

@rongrong / @rschlussel Would you be able to review this PR?. This can fix the high stage count issue we see which causes the REMOTE_TASK_ERROR issues.

[rschlussel]

fixes #13896

[rongrong]

The test failure seems related. Please take a look. Thanks!

[ankitdixit]

@rongrong I have tried to address the comments.
The failure seems to be unrelated, can you please check?

[ankitdixit]

The test failure seems related. Please take a look. Thanks!

Current set of failures are in presto-verifier and the test cases seem to pass locally.
Please let me know I we need to fix them.

[kaikalur]

I would like to see some test queries with this pattern as well - not just the unit test.

[ankitdixit]

I would like to see some test queries with this pattern as well - not just the unit test.

@kaikalur Do you want me to paste some queries and plan here? I do n]ot understand what test queries mean here, can you please give me an example

[rschlussel]

We don't need all the children to be values nodes. If any are values nodes, we can merge those and leave the other children as they were.

[ankitdixit]

@rschlussel As if now this optimizer replaces a Union node whose all children are valuesnode with a single values node.
What you are asking for is to subsume all valuesNode under union node ino the valuesnode.
I can :

1. Modify this optimizer to convert ot valuesnode or subsume in union depending upon whether all child are valuesnode
2. Implement a separate optimizer
   which is the preffered option?

[rschlussel]

Good question. I think option 1 is best.

[rschlussel]

I would like to see some test queries with this pattern as well - not just the unit test.

@kaikalur Do you want me to paste some queries and plan here? I do n]ot understand what test queries mean here, can you please give me an example

@ankitdixit you can add queries with this pattern to AbstractTestQueries. That will test that the queries work end to end and validates correctness

[kaikalur]

I would like to see some test queries with this pattern as well - not just the unit test.

@kaikalur Do you want me to paste some queries and plan here? I do n]ot understand what test queries mean here, can you please give me an example

I want to see actually something like

SELECT * FROM ( select 1, 2, 3 union all select * from (values (5,6,6),(10,20,30))

and it's output.

[ajaygeorge]

Hi @ankitdixit , can we get this PR merged in a week or so. This would really help us with the Remote task errors that we are seeing.

[stale]

This pull request has been automatically marked as stale because it has not had recent activity. If you'd still like this PR merged, please comment on the task, make sure you've addressed reviewer comments, and rebase on the latest master. Thank you for your contributions!