DE//AGENT
Explore
Tutorial Lesson 5 of 10 · The Orchestration Series community beginner

Review a DAG for idempotency and retry safety

Walk the Pipeline Reviewer's four-step pass over a DAG: apply the idempotency test to every task, check retry, timeout and concurrency settings, predict backfill behaviour, and propose missing integration tests — then close the loop with the Airflow DAG Reviewer's checklist so every task is safe to re-run before the DAG merges.

At a glance

When A DAG is about to merge and must be safe to retry; A backfill is being considered and its behaviour must be predicted first
Where Airflow / Orchestration
Why Non-idempotent tasks duplicate rows on every retry, and the reviewer's read-only four-step pass surfaces that before the DAG ships.

Learning objectives

  • Apply the idempotency test to every task in a DAG
  • Justify retry, timeout and concurrency settings per task
  • Predict what a backfill will do before running it

What you need

  • A DAG file or pull request to review
  • The pipeline's schedule and SLA
  • [object Object]

Steps

  1. Apply the idempotency test to every task. (from agent-pipeline-reviewer )
  2. Check retry, timeout and concurrency settings. (from agent-pipeline-reviewer )
  3. Predict backfill behaviour. (from agent-pipeline-reviewer )
  4. Propose missing integration tests. (from agent-pipeline-reviewer )
  5. Set retries with a backoff — a task that retries instantly will hammer a downstream. (from skill-airflow-dag-reviewer )
  6. Cap max_active_runs and pool limits so a backfill cannot starve the daily run. (from skill-airflow-dag-reviewer )

Check yourself

  • On the fixture of 8 DAGs with seeded non-idempotency bugs, the review flags every seeded bug and no healthy task.
  • Reviewing a task that appends without a partition key, you flag it as duplicating rows on retry.

Worked example

The idempotency test — from skill-airflow-dag-reviewer 

"If this task runs twice for the same logical date, is the output the same?"
If no, fix it before anything else — retries will happen.

DAG review checklist — from skill-airflow-dag-reviewer 

1. Every task writes to a partition keyed by the logical date, or uses MERGE.
2. `catchup` is a deliberate choice, not a default.
3. Retries have a backoff; a task that retries instantly will hammer a downstream.
4. `max_active_runs` and pool limits prevent a backfill from starving the daily run.
5. Sensors use `reschedule`, not `poke`, so they don't hold a worker slot.

Install the DAG reviewer — from skill-airflow-dag-reviewer 

npx de-agent-platform add airflow-dag-reviewer

Sourced from

Security scan

Last sweep 2026-10-24: 0 finding(s), 0 marked as examples.

Common gotchas

  • A task that appends without a partition key will duplicate rows on retry.
  • `catchup` is a deliberate choice, not a default.
  • Sensors use `reschedule`, not `poke`, so they don't hold a worker slot.

Production considerations

  • The reviewer is read-only analysis and never triggers a run.
  • Keep the fixture of 8 DAGs with seeded non-idempotency bugs in CI so regressions resurface.

Diagram

dag review pass
   |
   |- idempotency test per task --> flag appends without partition keys
   |
   |- retry/timeout/concurrency check --> backoff + max_active_runs
   |
   |- predict backfill behaviour --> propose missing integration tests

Registry stats

Verification community · Oct 21, 2026
Completeness 81
Trust multiplier ×1.25
Composite 34.15 · adequate
Freshness 3 days since last source check
Difficulty beginner
Version 0.1.0
License CC-BY-4.0
Security flags none