Send Automated Chat Alerts

Many developer teams use real-time chat services like Slack for their day-to-day communication. Members of those teams might also want to have automated notifications be sent to those chat services from within ML pipelines, e.g., to quickly and conveniently get notified about the state of each pipeline, or even to build human-in-the-loop ML systems.
Overview
The alerter component in ZenML allows teams to easily interact with their pipelines via a chat service.
To use alerters in a ML pipeline, ZenML provides two standard steps in zenml.alerter.alerter_step:
alerter_post_step() takes a string, posts it to the desired chat service, and returns True if the operation succeeded, else False.
alerter_ask_step(): does the same as alerter_post_step(), but after sending the message, it waits until someone approves or rejects the operation from within the chat service (e.g., by sending "approve" / "reject" to the bot as response). alerter_ask_step() then only returns True if the operation succeeded and was approved, else False.
To use those steps with your preferred chat service, you only need to register a corresponding alerter component in your ZenML stack. Right now, the SlackAlerter is the only alerter you can use out-of-the-box, but it is straightforward to extend ZenML and build an alerter for other chat services, as shown here.
Send alerts in your pipelines
To send alerts in your pipelines, you first need to register an alerter component using zenml alerter register <MY_ALERTER> and then add it to your stack with zenml stack register ... -al <MY_ALERTER>.
Afterward, you can simply import the standard alerter steps and use them in your pipeline.
Since these steps expect a string message as input (which needs to be the output of another step), you typically also need to define a dedicated formatter step that takes whatever data you want to communicate and generates the string message that the alerter should post.
As an example, adding alerter_ask_step() into your pipeline could look like this:
1
from zenml.alerter.alerter_step import alerter_ask_step
2
from zenml.steps import step
3
from zenml.pipelines import pipeline
4
​
5
​
6
@step
7
def my_formatter_step(artifact_to_be_communicated) -> str:
8
    return f"Here is my artifact {artifact_to_be_communicated}!"
9
​
10
​
11
@pipeline
12
def my_pipeline(..., formatter, alerter):
13
    ...
14
    artifact_to_be_communicated = ...
15
    message = formatter(artifact_to_be_communicated)
16
    approved = alerter(message)
17
    ... # Potentially have different behavior in subsequent steps if `approved`
18
​
19
​
20
my_pipeline(
21
    ...
22
    formatter=my_formatter_step(),
23
    alerter=alerter_ask_step(),
24
).run()
Copied!
For complete code examples of both alerter steps, see the slack_alert example here, where we first send the test accuracy of a model to Slack and then wait with model deployment until a user approves it in Slack.

Advanced Guide - Previous

Track your Experiments

Next - Extending ZenML

Getting Started

Last modified 10d ago

Export as PDF

Copy link

Edit on GitHub

Contents

Overview

Send alerts in your pipelines