start-stop-mwaa-environment - mwaa_import_data.py - task_instance.csv - fails for field next_kwargs is export contains a json object

[gamerf]

If a task_instance export has a json object on next_kwargs field, the mwaa_export.py writes the object as a string with single quotes (i.e. "{'var': {}, 'type': 'dict'}")

The mwaa_import.py COPY command, for json objects is expecting the field in the following format: {"var": {}, "type": "dict"}

as result, the import fails with the following error:

[2024-05-17T11:48:18.260+0000] {{taskinstance.py:1937}} ERROR - Task failed with exception
Traceback (most recent call last):
  File "/usr/local/airflow/.local/lib/python3.11/site-packages/airflow/operators/python.py", line 192, in execute
    return_value = self.execute_callable()
                   ^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/airflow/.local/lib/python3.11/site-packages/airflow/operators/python.py", line 209, in execute_callable
    return self.python_callable(*self.op_args, **self.op_kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/usr/local/airflow/dags/mwaa_import.py", line 192, in load_data
    cursor.copy_expert(query, f)
psycopg2.errors.InvalidTextRepresentation: invalid input syntax for type json
DETAIL:  Token "'" is invalid.
CONTEXT:  JSON data, line 1: {'...
COPY task_instance, line 4811, column next_kwargs: "{'__var': {}, '__type': 'dict'}"

[gamerf]

BTW,

The way I resolved this for the customer was by doing a SED on the exported files before importing the content

from sedpy import sedpy

def read_s3(context, filename):
    resource = boto3.resource('s3')
    bucket = resource.Bucket(get_s3_bucket(context))
    tempfile = f"/tmp/{filename}"
    try:
        bucket.download_file(S3_KEY + filename, tempfile)
    except botocore.exceptions.ClientError as e:
        if e.response['Error']['Code'] == "404":
            return None
        else:
            raise
    else:
        sedpy.replace("'", "\"\"", tempfile)
        return tempfile

There might be a more elegant way to fix this problem

[crupakheti]

Thank you, @gamerf for reporting the issue and providing a workaround. We will investigate this further and include a fix soon.