Dagster Partitions and Automation¶

Partitioned asset graphs are often where Dagster becomes more valuable than a generic scheduler. Truthound does not redefine partition semantics; it relies on Dagster to decide what should run and then executes checks with shared runtime semantics.

Who This Is For¶

teams validating daily, hourly, or tenant-scoped partitions
operators using Dagster automation rules
asset owners who need partition-aware failure isolation

When To Use It¶

Use this page when:

asset checks should execute per partition
only new partitions should be validated
failures need to stay local to the partition that broke

Prerequisites¶

a Dagster asset graph with partition definitions
Truthound asset helpers or resource-backed assets
a clear ownership model for partitions

Minimal Quickstart¶

Use quality_checked_asset on partitioned assets so the quality contract stays next to the asset:

from dagster import DailyPartitionsDefinition
from truthound_dagster import quality_checked_asset

@quality_checked_asset(
    partitions_def=DailyPartitionsDefinition(start_date="2026-01-01"),
    rules=[{"column": "id", "check": "not_null"}],
)
def users_partitioned(context):
    return load_partition(context.partition_key)

Production Pattern¶

Recommended automation model:

Need	Recommended Pattern
validate every fresh partition	asset checks tied to the asset definition
rerun only failed partitions	use Dagster re-execution on the failed partition keys
separate expensive quality work	move profile/learn paths into distinct jobs

Operational advice:

keep partition-level rules deterministic
use partition metadata to help operators locate the failing slice
avoid shared mutable state between partition checks

Failure Modes and Troubleshooting¶

Symptom	Likely Cause	What To Do
one bad partition blocks everything	validation is modeled at the whole-asset level	move quality logic to partition-aware assets or jobs
partition reruns are expensive	profile or learn logic is mixed into every partition	split exploratory work from steady-state checks
operators cannot tell which slice failed	run metadata omits partition context	include partition keys in alert and metadata output