Airflow SLA And Callbacks¶

Airflow is a natural place to attach SLA and alerting behavior to quality checks. The first-party package keeps the metric definitions and violation semantics shared, while leaving callback wiring and DAG behavior inside Airflow.

SLAConfig¶

Use SLAConfig to define acceptable latency and failure envelopes for quality tasks or quality-gated DAG steps.

from truthound_airflow import SLAConfig

config = SLAConfig(
    max_failure_rate=0.01,
    max_execution_time_seconds=300.0,
)

Typical questions this answers:

when should a quality task be considered degraded?
when should a callback warn versus fail loudly?
how do we route high-severity quality regressions?

SLAMetrics¶

SLAMetrics is the shared metric payload that the monitor evaluates. It keeps the same core logic available across Airflow and other hosts.

SLAViolation¶

SLAViolation and SLAViolationType are the structured outputs that callbacks and monitors consume.

SLAMonitor¶

SLAMonitor evaluates recorded results against the configured SLA contract.

from truthound_airflow import SLAMonitor, SLAConfig

monitor = SLAMonitor(
    config=SLAConfig(
        max_failure_rate=0.05,
        max_execution_time_seconds=600.0,
    )
)

Callback Types¶

The Airflow package exposes several callback helpers:

DataQualitySLACallback
QualityAlertCallback
CallbackChain

Use them to map structured violation information into Airflow-native alerting behavior.

Registry Support¶

SLARegistry is useful when you want several DAGs to share named SLA definitions instead of re-declaring them inline.

Recommended Airflow Pattern¶

from truthound_airflow import DataQualityCheckOperator, DataQualitySLACallback, SLAConfig

sla_callback = DataQualitySLACallback(
    sla_config=SLAConfig(max_failure_rate=0.02, max_execution_time_seconds=300),
)

check = DataQualityCheckOperator(task_id="quality_check", data_path="users.parquet")

The important design rule is to keep rule execution in operators and policy escalation in callbacks or monitors.