wr.athena.to_iceberg - Insert query has mismatched column types

[Mroq93]

Describe the bug

I try to save several Data Frames to Iceberg table using wr.athena.to_iceberg.
A few incremental savings go without any issues, but after some iteration I am getting error:
TYPE_MISMATCH: Insert query has mismatched column types: Table: [varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, timestamp(6), varchar, varchar, varchar, varchar, varchar, varchar, varchar], Query: [varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, varchar, timestamp(3), varchar, varchar, varchar, varchar, varchar, varchar]. If a data manifest file was generated at 's3://bucket-temp/athena/results/c0c18807-2773-4afc-b95c-580034d960ed-manifest.csv', you may need to manually clean the data from locations specified in the manifest. Athena will not delete data in your account.

I see 2 differences between schemas:

• In the table, there is timestamp(6) but the next saving to iceberg recognizes column as timestamp(3)
• the number of types is different by one

Before saving, I cast Dataframe timestamp columns to the same format to be sure that every timestamp is aligned.

        dataframe[name] = pd.to_datetime(
            dataframe[name], format='ISO8601'
        ) 

When I print this timestamp, column for every Dataframe column format is the same.

In dtype in wr.athena.to_iceberg , for timestamp column I provide type as timestamp, I can not provide precision - it is not supported.

I am not sure if the matter of different number of columns should be an issue. I guess it was resolved here:
#2616

PS.
The order of columns does matter?

How to Reproduce

                wr.athena.to_iceberg(
                    df=df,
                    database=database,
                    table=tbl_name,
                    table_location=s3_target_path,
                    temp_path=f"{s3_target_path}temp/{ini_time}",
                    partition_cols={"bookmark_date_str"},
                    workgroup=work_group_name,
                    schema_evolution=True,
                    dtype=columns_for_iceberg,
                )

Expected behavior

No issues with timestamp precision
No issues when saving DataFrames with different schema(missing or additional columns)

Your project

No response

Screenshots

No response

OS

AWS

Python version

3.9

AWS SDK for pandas version

3.5.2

Additional context

No response

[kukushking]

Hi @Mroq93 thanks for opening this - looking into it.

[GalVishi]

Hello @kukushking, do you have any news or updates regarding the bug that we discussed earlier?

[GalVishi]

Just need to change this line:

aws-sdk-pandas/awswrangler/athena/_write_iceberg.py

Line 489 in da4ba40

sql_statement = f'INSERT INTO "{database}"."{table}" SELECT * FROM "{database}"."{temp_table}"'

sql_statement = f'INSERT INTO "{database}"."{table}" SELECT {', '.join([f'"{x}"' for x in df.columns])} FROM "{database}"."{temp_table}"'

[Mroq93]

Hi @jaidisido @GalvFionic ,
I tried 3.7.1 awswrangler version, but still I am facing the issue.
Is the fix available in the latest version?