thesuperzapper commented on issue #24780: [SPARK-26388][SQL][WIP]_implement_hive_ddl_alter_table_replace_columns URL: https://github.com/apache/spark/pull/24780#issuecomment-499309529 @emanuelebardelli If we are trying to replicate [Hive's REPLACE COLUMN command](https://cwiki.apache.org/confluence/display/Hive/LanguageManual+DDL#LanguageManualDDL-Add/ReplaceColumns), we need to allow overwriting the entire schema of the table, regardless of if the final table will be readable. (The only exception is for table formats which it doesn't make sense, like ones which determine columns from index rather than name) I am not sure if your current approach will achieve this, can you confirm that: 1. You can fully replace columns (including comments) 1. You can reorder columns 1. You can change the type of columns * There are cases where a new schema is able to read both old and new data without issues. What is allowed, depends on the format, e.g. PARQUET/AVRO. (This is often used for schema evolution) * Example: `STRUCT<`field1`: STRING>` → `STRUCT<`field1`: STRING, `field2`: STRING>` * Note that this is __a column of type STRUCT__, not the full schema of a table Perhaps you could look into using one of the existing methods to overwrite the schema of the table, they are implemented in the various external catalogue classes. For example [HiveExternalCatalog.scala](https://github.com/apache/spark/blob/v2.4.3/sql/hive/src/main/scala/org/apache/spark/sql/hive/HiveExternalCatalog.scala#L647), which calls [HiveClient.scala](https://github.com/apache/spark/blob/v2.4.3/sql/hive/src/main/scala/org/apache/spark/sql/hive/client/HiveClient.scala#L111).
---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
