[jira] [Created] (SPARK-38904) Low cost DataFrame schema swap util

["Rafal Wojdyla (Jira)" <ji...@apache.org>]

Rafal Wojdyla created SPARK-38904:
-------------------------------------

             Summary: Low cost DataFrame schema swap util
                 Key: SPARK-38904
                 URL: https://issues.apache.org/jira/browse/SPARK-38904
             Project: Spark
          Issue Type: New Feature
          Components: SQL
    Affects Versions: 3.2.1
            Reporter: Rafal Wojdyla


This question is related to [https://stackoverflow.com/a/37090151/1661491]. Let's assume I have a pyspark DataFrame with certain schema, and I would like to overwrite that schema with a new schema that I *{*}know{*}* is compatible, I could do:
{code:python}
df: DataFrame
new_schema = ...

df.rdd.toDF(schema=new_schema)
{code}
Unfortunately this triggers computation as described in the link above. Is there a way to do that at the metadata level (or lazy), without eagerly triggering computation or conversions?

Edit, note:
 * the schema can be arbitrarily complicated (nested etc)
 * new schema includes updates to description, nullability and additional metadata (bonus points for updates to the type)
 * I would like to avoid writing a custom query expression generator, *{*}unless{*}* there's one already built into Spark that can generate query based on the schema/`StructType`

Copied from: [https://stackoverflow.com/questions/71610435/how-to-overwrite-pyspark-dataframe-schema-without-data-scan]

See POC of workaround/util in https://github.com/ravwojdyla/spark-schema-utils

Also posted in [https://lists.apache.org/thread/5ds0f7chzp1s3h10tvjm3r96g769rvpj]



--
This message was sent by Atlassian Jira
(v8.20.1#820001)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org