zypp-io / df_to_azure Goto Github PK

Add test with sample.csv

Make an option for incremental updates. An extra parameter is necessary: "id field"

in df_to_azure/export.py on line 200 we export the file to a .csv file:

data = self.df.to_csv(index=False, sep="^", quotechar='"', lineterminator="\n")

We should change this to .parquet for optimization. It requires a changes in the Linked Services in adf.py on line 173 :

We need a test for the append method.

Nu wordt de database connectie gemaakt in de module, daarmee moeten de users nog meer environment variabele definiëren voor de module. Ik denk dat het beter zou zijn als we connectie als argument accepteren in onze functies die met de database te maken hebben en dat stuk dus buiten deze module laten.

@melvinfolkers wat denk jij?

1. Remove mentioning of yml
2. Add UPSERT as method.

output from console

azure.core.exceptions.HttpResponseError: The specifed resource name contains invalid characters.
RequestId:1fe04a1b-501e-0019-2190-221067000000
Time:2022-02-15T17:20:16.7288425Z
ErrorCode:InvalidResourceName

output from pip list, regarding azure packages:

azure-common           1.1.26
azure-core             1.11.0
azure-identity         1.5.0
azure-keyvault-secrets 4.2.0
azure-mgmt-core        1.2.2
azure-mgmt-datafactory 1.0.0
azure-mgmt-resource    15.0.0
azure-storage-blob     12.7.1

Problem: When the container "dftoazure" does not exist, df_to_azure exits with an ContainerNotFound error.

Code from adf.py:

def create_blob_container():
    blob_service_client = create_blob_service_client()
    try:
        blob_service_client.create_container("dftoazure")
    except:
        logging.info("CreateContainerError: Container already exists.")

suggested solution:

Check in azure if the container "dftoazure" is present. if not, create the container.

blob_service_client = create_blob_service_client()
containers = [x.get("name") for x in blob_service_client.list_containers()]
if "dftoazure" not in containers:
    blob_service_client.create_container(name=container_name)      

The main functionality now is to upload a pandas dataframe to a sql database through datafactory. In many cases we make use of parquet files instead of a database.

The proposal is to add a method to export a dataframe to parquet and upload it to azure storage. When method="append" is used, we need to download the parquet file and append the newest data.

df_to_azure(df=df, tablename="table_name", folder="schema_name", container="containername", method="create", parquet=True)

add functionality for creating a container if it does not exist. The only problem is that you need the account_key for the container.

When there's a space in a column we get an error in the UPSERT method.

This needs to be fixed by enclosing them in brackets: [Column Name]

In order to use this repository as a submodule, changes have to be made to the import statements.
Currently scripts get imported as follow:

from src.functions import my_function

in order for it to work in a submodule, the references have to be changed to:

from .functions import my_function

There are already some scripts created in another repository. The plan is to copy these scripts and adjust them here and there

When there's a Categorical dtype column present in the dataframe, it will error on the type conversion. It needs to be added to the dictionary with type:

creating initial project layout

When writing a dataframe to a SQL database it gives an error in ADF because of max number of characters exceeded (Operation on target Copy all_boards to SQL failed: Failure happened on 'Sink' side. ErrorCode=SqlBulkCopyInvalidColumnLength,.... This was the case with a column with a string version of a list of dicts.
The max characters in Python were 1870, so the schema definition in the database became varchar(1870), in the CSV the max length of values was 1870 as wel after checking as a txt file and after loading it again in Python.

However in ADF it is causing the error during the copy action. It does not give the error if we manually enter the text_length parameter in df_to_azure() as 1873

Your documentation states that id_field only should be used with upsert. The example on your page shows id_field with the create method:

from df_to_azure import df_to_azure

df_to_azure(df=df, tablename="table_name", schema="schema", method="create", id_field="col_a")

We have to add an exception in case a method is used that is not supported by this package:

for instance this method does not exist

method="insert"

instead of throwing an error, it appends the dataset to the sql database.

Now that we have parquet support, we should add logic to do upsert on parquet files.

In general we this will be DataFrame.combine_first, we probably need some good testing for this, how will this deal with NaN for example and datetime indices:

df1 = pd.DataFrame(
    {
        "id": ["A", "B", "C"],
        "val1": [10, 20, 30],
        "val2": [40, 50, 60]
    }
)

df2 = pd.DataFrame(
    {
        "id": ["A", "B", "C", "D"],
        "val1": [15, 20, 30, 35],
        "val2": [40, 52, 60, 70]
    }
)

print(df1, "\n")
print(df2, "\n")
print(df2.set_index("id").combine_first(df1.set_index("id")).reset_index())

  id  val1  val2
0  A    10    40
1  B    20    50
2  C    30    60 

  id  val1  val2
0  A    15    40
1  B    20    52
2  C    30    60
3  D    35    70 

  id  val1  val2
0  A    15    40
1  B    20    52
2  C    30    60
3  D    35    70