function_app.py

"""
Copyright (c) AtLongLast Analytics LLC

Licensed under the Apache License, Version 2.0

Project: https://github.com/AtLongLastAnalytics/vigil
Author: Robert Long
Date: 2026-03
Version: 0.1.0

File: function_app.py
Description: Azure Function App entry point; defines the timer-triggered function that
    orchestrates monitoring and email reporting.
"""

# import standard library
import logging
import os
from datetime import datetime, timezone

# import third-party libraries
import azure.functions as func

# import project modules
from archive_service import archive_runs
from config import load_config
from constants import PipelineStatus
from email_service import generate_email_body, send_email
from monitor import get_credential, get_recent_pipeline_runs

# configure logging
logging.basicConfig(level=logging.INFO)

# initialize logger
logger = logging.getLogger(__name__)

app = func.FunctionApp()

# schedule defaults to 06:00 and 08:00 UTC daily — override with MONITOR_SCHEDULE app setting
_MONITOR_SCHEDULE = os.environ.get("MONITOR_SCHEDULE", "0 0 6,8 * * *")


@app.timer_trigger(schedule=_MONITOR_SCHEDULE, arg_name="myTimer", run_on_startup=False)
def vigil_monitor(myTimer: func.TimerRequest) -> None:
    """
    Monitor Synapse pipelines and send daily report.

    Schedule is controlled by the MONITOR_SCHEDULE environment variable
    (default: "0 0 6,8 * * *" — 06:00 and 08:00 UTC daily).
    Fetches pipeline runs from the past N hours and sends an email report.

    Args:
        myTimer (func.TimerRequest): timer trigger request object provided by the Azure Functions host
    """
    logger.info("=== Vigil Starting ===")

    try:
        # load and validate configuration
        logger.info("Loading configuration...")
        config = load_config()

        # get Azure credential
        logger.debug("Obtaining Azure credential...")
        credential = get_credential(config)

        # fetch recent pipeline runs
        logger.info(f"Fetching pipeline runs for the past {config.hours_back} hours...")
        runs = get_recent_pipeline_runs(credential, config)

        if not runs:
            logger.info("No pipeline runs found in the specified time range")
            return

        logger.info(f"Retrieved {len(runs)} pipeline runs")

        now_utc = datetime.now(timezone.utc)

        # structured summary — custom_dimensions are forwarded to Application Insights
        failed_count = sum(1 for r in runs if r['status'] == PipelineStatus.FAILED)
        succeeded_count = sum(1 for r in runs if r['status'] == PipelineStatus.SUCCEEDED)
        in_progress_count = sum(1 for r in runs if r['status'] == PipelineStatus.IN_PROGRESS)
        success_rate = (succeeded_count / len(runs) * 100) if runs else 0.0

        logger.info(
            f"Pipeline run summary: total={len(runs)}, succeeded={succeeded_count}, "
            f"failed={failed_count}, in_progress={in_progress_count}",
            extra={"custom_dimensions": {
                "total_runs": len(runs),
                "succeeded": succeeded_count,
                "failed": failed_count,
                "in_progress": in_progress_count,
                "success_rate": f"{success_rate:.1f}",
                "hours_back": config.hours_back,
            }},
        )

        logger.debug("Generating email body...")
        html_body = generate_email_body(runs, now_utc)

        # create subject line
        date_str = now_utc.strftime('%Y-%m-%d')
        subject = (
            f"Vigil - Pipeline Alert: {failed_count} Failed - {date_str}"
            if failed_count > 0
            else f"Vigil - All Clear - {date_str}"
        )

        # send email
        logger.info(f"Sending email: {subject}")
        send_email(
            subject,
            html_body,
            config.acs_endpoint,
            credential,
            config.sender_address,
            config.recipient_list
        )

        # archive pipeline run data
        logger.debug("Archiving pipeline run data...")
        archive_runs(runs, config, credential, now_utc)

        logger.info("=== Vigil Completed Successfully ===")

    except ValueError as e:
        logger.error(f"Configuration error: {e}")
        raise

    except Exception as e:
        logger.error(f"Unexpected error in pipeline monitor: {e}", exc_info=True)
        raise