---
title: "Airflow DAGs with no retries | Metacenta"
description: "A dropped connection should not cost a day of stale data. How we spot Airflow, Dagster and Prefect jobs with no retry policy, and when zero retries is right."
url: "https://metacenta.com/rules/dag-no-retries"
updated: "2026-10-14"
tokens: 616
---

Orchestration & reliability · rule dag-no-retries

# Airflow DAGs with no retries

A Metacenta review checks this under the rule **A scheduled job retries a transient failure**. Everything below applies whether or not you ever commission one.

## What this rule checks

This rule flags each job file that declares no retry count, or sets it to zero. A count above zero passes, as a `default_args` key, a `retries=` keyword, `max_retries=` or a Dagster `RetryPolicy`. We rate a job with ten or more tasks higher.

## Why it matters

Warehouses drop connections, credentials refresh and APIs rate-limit. None of that means the pipeline is wrong. Without a retry, each one becomes a failed run that somebody reruns by hand the next morning, and the data is stale until then.

## How to fix it

**Retry transient failures in scheduled jobs.** Set `retries` and `retry_delay` in `default_args`. Two or three retries with exponential backoff is the usual answer. The point is a decision rather than an omission.

Before:

```python
default_args = {
    "owner": "finance-data",
}
```

After:

```python
default_args = {
    "owner": "finance-data",
    "retries": 3,
    "retry_delay": timedelta(minutes=5),
    "retry_exponential_backoff": True,
}
```

## When it is fine to leave

A task that is not safe to run twice should not retry. An append without a unique key, or a call that sends email or charges a card, belongs here. Make those idempotent where you can, and set `retries=0` on that task alone.

## What we need to check it

The job files you name to us one at a time. We read the count only when it is a number; a value imported from a shared module reads as undeclared. A file that only wires jobs declared elsewhere is reported as not examined, not as missing retries.

[Ask about a review](https://metacenta.com/enquire?src=rule-dag-no-retries&via=md)

## Related pages

- [Airflow DAG owner: how to set it and what it does](https://metacenta.com/rules/dag-unowned)
- [Airflow DAGs with no failure alert](https://metacenta.com/rules/dag-no-failure-alert)
- [Airflow catchup: what it does and how to control it](https://metacenta.com/rules/dag-catchup-unbounded)
- [Passwords and API keys hardcoded in DAG files](https://metacenta.com/rules/dag-credential-literal)
- [A sample review, as a client reads it](https://metacenta.com/sample)
- [All published rules](https://metacenta.com/rules)
- [How a review is assessed](https://metacenta.com/standard)