Hello,
We use Spark jobs running on EMR to create and update tables which we manage in Glue (they are Parquet files in S3).
We have recently switched to Athena engine v3 and are working to implement our tables as delta native tables.
The goal is to create a new native delta table on top of our S3 address and to populate that table properly. We are running the following SQL commands in Spark (from EMR):
DROP TABLE IF EXISTS <aws_ID>/orders<tablename>;
CREATE EXTERNAL TABLE <aws_ID>/orders.<tablename>;
(<schema>)
PARTITIONED BY ( year STRING, month STRING, day STRING, dt STRING)
LOCATION 's3://REDACTED'
ROW FORMAT SERDE 'org.apache.hadoop.hive.ql.io.parquet.serde.ParquetHiveSerDe'
STORED AS INPUTFORMAT 'org.apache.hadoop.hive.ql.io.parquet.MapredParquetInputFormat'
OUTPUTFORMAT 'org.apache.hadoop.hive.ql.io.parquet.MapredParquetOutputFormat'
TBLPROPERTIES('table_type' = 'DELTA');
At this point the table is created with the proper schema and zero rows. We run the following command to populate it:
MSCK REPAIR TABLE <tablename>
The table contains many rows that are exact duplicates on every field. How can we avoid this? Should we approach it differently?
AWS 官方已更新 1 年前
Thanks Gonzalo! Makes sense. The DDL you linked works from within Athena, but not from within Spark.
Do you know how to create a native delta table for Athena engine v3 in PySpark?
I think it's doable, in Spark write the data with the delta library and then create the table in the catalog using CREATE EXTERNAL TABLE specifying the columns, location and 'table_type'='DELTA'