HOW TO LOG IN RUN FUNCTION OF SCORING SCRIPT ON AML

@Nguyễn Thanh Tú Welcome to Microsoft Q&A Forum, Thank you for posting your query here!

.

It seems like you're on the right track with your logging setup! To ensure logging works in the run function, you might need to retrieve the current run context within the run function.

Sample code is available here.

Here's an updated version of your code with the necessary changes:

import os
import sys
import json
import logging
from azureml.core import Run
import mlflow
import time

connection_string = os.getenv("CONNECTION_STRING")
APPICATIONINSIGHTS_CONNECTION_STRING = connection_string
# Set up custom JSON formatter for logging
fmt = logging.Formatter("%(levelname)%(asctime)%(message)%(filename)%(funcName)%(lineno)")
 
# Set up a stream handler to output logs to stdout
sh = logging.StreamHandler(sys.stdout)
sh.setFormatter(fmt)
 
# Configure the root logger
logging.basicConfig(handlers=[sh], level=logging.INFO)
 
# Create a logger object for the current module
logger = logging.getLogger(__name__)
logger.setLevel(logging.INFO)
 
# Add the stream handler to the logger
logger.addHandler(sh)
 
# Set up an Azure Log Handler to send logs to Azure Application Insights
azure_handler = AzureLogHandler(connection_string=APPICATIONINSIGHTS_CONNECTION_STRING)
logger.addHandler(azure_handler)
 
workflow_name = os.getenv("WORKFLOW_NAME")
custom_dimensions = {"workflow_name": workflow_name}
 
def init() -> None:
    logger.info(f"connection string: {connection_string}", extra={"custom_dimensions": custom_dimensions})
    """Initialize the scoring process"""
    global model  # noqa: PLW0603
    global input_schema  # noqa: PLW0603
    model_path = Path(os.getenv("AZUREML_MODEL_DIR") or "") / "model"
    model = mlflow.pyfunc.load_model(model_path)
    input_schema = model.metadata.get_input_schema()
 
def run(raw_data: str | bytes) -> str:
    """Process incoming data and return predictions"""
    run = Run.get_context()
    logger.info(f"raw data: {raw_data}", extra={"custom_dimensions": custom_dimensions})
    json_data = json.loads(raw_data)
    if "input_data" not in json_data:
        msg = "Request must contain a top level key named 'input_data'"
        raise Exception(msg)  # noqa: TRY002
 
    serving_input = {"dataframe_split": json_data["input_data"]}
    data = infer_and_parse_data(serving_input, input_schema)
    predictions = model.predict(data)
 
    result = StringIO()
    predictions_to_json(predictions, result)
    return result.getvalue()

.

In this updated code, I've added Run.get_context() to retrieve the current run context within the run function. This should allow you to log information correctly during online endpoint deployment.

.

Hope this helps. If you have any follow-up questions, please let me know. I would be happy to help.

Hi @navba-MSFT

I've tried your suggestion but I still cannot log the raw data in the run function, only the connection string (in the init function) can be logged

@Nguyễn Thanh Tú Thanks for getting back. Could you please paste the log in the text format here ? For some reason. the logging screenshot is not loading at my end clearly.