Measure reliability with Service Level Agreements

​

Prerequisites

• Prefect Client Version 3.1.12 or later
• Prefect Cloud account (SLAs are only available in Prefect Cloud)

​

Service Level Agreements

Service Level Agreements (SLAs) help you set and monitor performance standards for your data stack. By establishing specific thresholds for flow runs on your Deployments, you can automatically detect when your system isn’t meeting expectations. When you set up an SLA, you define specific performance criteria - such as a maximum runtime of 10 minutes for a flow. If a flow run exceeds this threshold, the system generates an alert event. You can then use these events to trigger automated responses, whether that’s sending notifications to your team or initiating other corrective actions through automations.

​

Defining SLAs

To define an SLA you can add it to the deployment through a prefect.yaml file, a .deploy method, or the CLI:

deployments:
  my-deployment:
    sla:
        - name: "time-to-completion"
          duration: 10
          severity: "high"
        - name: "lateness"
          within: 600
          severity: "high"

    from prefect import flow
    from prefect._experimental.sla.objects import TimeToCompletionSla, LatenessSla, FrequencySla

    @flow
    def my_flow():
        pass

    flow.from_source(
        source=source,
        entrypoint="my_file.py:my_flow",
    ).deploy(
        name="my-deployment",
        work_pool_name="my_pool",
        _sla=[
            TimeToCompletionSla(name="my-time-to-completion-sla", duration=30, severity="high"),
            LatenessSla(name="my-lateness-sla", within=timedelta(minutes=10), severity="high"),
            FrequencySla(name="my-frequency-sla", stale_after=timedelta(hours=1), severity="high"),
        ]
    )

prefect deploy --sla '{"name": "my-time-to-completion-sla", "duration": 10, "severity": "high"}'

​

Types of SLAs

​

Time to Completion SLA

A Time to Completion SLA describes how long flow runs should take and the severity of going over that duration. The SLA is triggered when a flow run takes longer than the specified duration to complete.

The Time to Completion SLA requires one unique parameter:

• duration: The maximum allowed duration in seconds before the SLA is violated

This SLA is useful for monitoring long-running flows and ensuring they complete within an expected timeframe. For example, if you set a Time to Completion SLA on your deployment with a duration of 600 seconds (10 minutes) the backend will emit an prefect.sla.violation event if a flow run does not complete within that timeframe.

​

Frequency SLA

A Frequency SLA describes the interval a deployment should run at and the severity of missing that interval.

The Frequency SLA requires one unique parameter:

• stale_after: The maximum allowed time between flow runs (e.g. 1 hour for hourly jobs)

For example, if you have a deployment that should run every hour, you can set up a Frequency SLA with a stale_after of 1 hour. This SLA triggers when more than an hour passes between Completed flow runs.

​

Lateness SLA

A Lateness SLA describes the severity of a flow run failing to start at its scheduled time.

The Lateness SLA requires one unique parameter:

• within: The amount of startup time allotted to a flow run before the SLA is violated.

For example, if a deployment is scheduled to run every day at 10am, and has a Lateness SLA of 5 minutes, the backend will emit an SLA violation event if a flow run does not start by 10:05am.

​

Monitoring SLAs

You can monitor SLAs in the Prefect Cloud UI. On the runs page you can see the SLA status in the top level metrics:

​

Setting up an automation

To set up an automation to notify a team or to take other actions when an SLA is triggered, you can use the automations feature. To create the automation first you’ll need to create a trigger.

1. Choose trigger type ‘Custom’.
2. Choose any event matching: prefect.sla.violation
3. For “From the Following Resources” choose: prefect.flow-run.*

After creating the trigger, you can create an automation to notify a team or to take other actions when an SLA is triggered using automations.