[SPARK-34720][SQL] MERGE ... UPDATE/INSERT * should do by-name resolution #32192
Conversation
|
Test build #752552040 for PR 32192 at commit |
| UpdateAction( | ||
| updateCondition.map(resolveExpressionByPlanChildren(_, m)), | ||
| resolveAssignments(assignments = None, m, resolveValuesWithSourceOnly = false)) | ||
| // For UPDATE *, the value must from source table. | ||
| resolveAssignments(assignments, m, resolveValuesWithSourceOnly = true)) |
There was a problem hiding this comment.
To avoid resolving UpdateAction again with resolveValuesWithSourceOnly = false in the next analysis round, I added resolveMergeExprOrFail to fail earlier if an attribute can't be resolved.
|
cc @aokolnychyi and @RussellSpitzer |
|
I think this is a great idea, I see folks hitting this extremely frequently. It does feel like it would be a rather large change to the current behavior though. |
|
Kubernetes integration test starting |
|
Kubernetes integration test status failure |
|
Test build #137424 has finished for PR 32192 at commit
|
|
Kubernetes integration test starting |
|
Kubernetes integration test status failure |
|
Test build #138256 has finished for PR 32192 at commit
|
There was a problem hiding this comment.
Although this is not ANSI, there exists some references which the users might expect. It would be great if you add a comparison section to compare your proposal with them in your PR description officially for a record. At least, I guess we can find the followings.
- Snowflake: https://docs.snowflake.com/en/sql-reference/sql/merge.html
- Oracle: https://docs.oracle.com/cd/B19306_01/server.102/b14200/statements_9016.htm
- SQL Server: https://docs.microsoft.com/en-us/sql/t-sql/statements/merge-transact-sql?view=sql-server-ver15
- Databricks Delta: https://docs.databricks.com/spark/latest/spark-sql/language-manual/delta-merge-into.html
|
I've made it clear at the beginning of the PR description: INSERT/UPDATE * is an extension and I can't find it in other mainstream databases. If you open the docs you posted, none of them (except for Delta lake) supports INSERT/UPDATE *. Actually the original Spark PR to add MERGE SQL syntax was following https://docs.databricks.com/spark/latest/spark-sql/language-manual/delta-merge-into.html . It's a mistake that the INSERT/UPDATE * behavior is different and the current behavior is super confusing that we'd better fix it. |
|
Got it for the further explanation. |
|
thanks for the review, merging to master! |
…tion In Spark, we have an extension in the MERGE syntax: INSERT/UPDATE *. This is not from ANSI standard or any other mainstream databases, so we need to define the behaviors by our own. The behavior today is very weird: assume the source table has `n1` columns, target table has `n2` columns. We generate the assignments by taking the first `min(n1, n2)` columns from source & target tables and pairing them by ordinal. This PR proposes a more reasonable behavior: take all the columns from target table as keys, and find the corresponding columns from source table by name as values. Fix the MEREG INSERT/UPDATE * to be more user-friendly and easy to do schema evolution. Yes, but MERGE is only supported by very few data sources. new tests Closes apache#32192 from cloud-fan/merge. Authored-by: Wenchen Fan <wenchen@databricks.com> Signed-off-by: Wenchen Fan <wenchen@databricks.com>
What changes were proposed in this pull request?
In Spark, we have an extension in the MERGE syntax: INSERT/UPDATE *. This is not from ANSI standard or any other mainstream databases, so we need to define the behaviors by our own.
The behavior today is very weird: assume the source table has
n1columns, target table hasn2columns. We generate the assignments by taking the firstmin(n1, n2)columns from source & target tables and pairing them by ordinal.This PR proposes a more reasonable behavior: take all the columns from target table as keys, and find the corresponding columns from source table by name as values.
Why are the changes needed?
Fix the MEREG INSERT/UPDATE * to be more user-friendly and easy to do schema evolution.
Does this PR introduce any user-facing change?
Yes, but MERGE is only supported by very few data sources.
How was this patch tested?
new tests