✅ thought for 107.3 seconds

Got it — Prefect 2.x. Best practice: - Default to flow-level notifications. Alert on final flow states (Failed/Crashed; optionally Completed if it’s truly useful). You get one clean signal with full context. - Use task-level notifications sparingly for high-impact steps only (e.g., a critical load or payment). Otherwise you’ll earn a lifetime subscription to alert fatigue. How to do it in Prefect 2.x Prefect Cloud (recommended) - Use Automations to send notifications on flow-run state changes. Configure: - Trigger: Flow run state entered = Failed/Crashed (optionally Late or Completed) - Filters: by flow, tags, work queue/pool, etc. - Action: Send notification to Slack/email/webhook (whatever you’ve connected) - You can also target Task Run states if you truly need task-level alerts—filter by task name or tag to keep it sane. - Docs: - Automations: https://docs-2.prefect.io/latest/concepts/automations/ - States overview: https://docs-2.prefect.io/latest/concepts/states/ Open-source (or custom formatting) - Use notification blocks (e.g., Slack webhook) in code. Create once, then load and notify in your flow. Example:

from prefect import flow, task
from prefect.blocks.notifications import SlackWebhook

# One-time setup (run once, e.g., in a notebook or script)
# slack_block = SlackWebhook(url="<https://hooks.slack.com/services/XXX/YYY/ZZZ>")
# slack_block.save("alerts", overwrite=True)

@task
def do_thing():
    # your task logic
    return 42

@flow
def my_flow():
    slack = SlackWebhook.load("alerts")
    try:
        result = do_thing()
        # Optional success notification if you really want it
        # slack.notify(text="my_flow completed successfully.")
        return result
    except Exception as e:
        slack.notify(text=f"my_flow failed: {e}")
        raise

- For a critical task-only alert, wrap that task’s logic similarly (notify in the

except

, then re-raise). - To avoid spam from retries, only notify on the final failure (e.g., after your retries are exhausted) or send at the flow level only. Noise control tips - Alert on final states only (Failed/Crashed) and maybe Late. - If you must do task-level alerts, limit to specific tasks/tags and dedupe retries. - Include the flow run URL and state message in notifications for fast triage (Cloud automations do this nicely). If you tell me whether you’re on Cloud or OSS and which channel you want (Slack, email, webhook), I’ll give you the exact clicks or code. PS: If you’re planning any upgrades, Prefect 3.x improves the notification story and worker model. I won’t pretend it makes alerts exciting, but it does make them easier.