Skip to content

Releases: tchandrap/dbt-airflow-lineage

Release list

v0.1.1

Choose a tag to compare

@github-actions github-actions released this 13 Oct 14:59

Changelog

0.1.1 — 2026-10-13

  • No longer marked as a preview: the Marketplace "Preview" badge is gone.
  • README: animated demo of the extension in action (docs/demo.gif).
  • GCS sync copies only .py files and skips tests/ and __pycache__/ (configurable with dbtAirflow.dagsSyncExclude), so test data in a bucket is never downloaded.
  • GCS sync falls back to gsutil when gcloud itself crashes, and explains the crash.
  • The cache marker now lives outside the mirrored folder, so pruning can no longer delete it.
  • The graph area and dialogs follow the real height of the top bar, so Fit no longer slides the first row of nodes under it in narrow panels.

0.1.0 — first public release

  • Interactive lineage graph from target/manifest.json (drag, pan, zoom, animated links for the selected node).
  • Built-in dbt selector engine: names, tag:, path:, source:, package:, config.*, + / N+ / @, union, intersection, --exclude, named selectors from selectors.yml.
  • Floating selector search, filter popover (select / exclude / types), find-in-graph, legend.
  • Optional Airflow pipelines: which DAG runs a model (dbt selectors) and which DAG ingests a table (BigQuery dataset + table in DAG code).
  • Airflow source setup wizard: local folder or Google Cloud Storage (bucket browser, access test, sign-in recovery).
  • Model details dialog that edits the model's YAML in place: description, tags, config, meta, tests, columns.

v0.1.0

Choose a tag to compare

@github-actions github-actions released this 13 Oct 12:48

Changelog

0.1.0 — first public release

  • Interactive lineage graph from target/manifest.json (drag, pan, zoom, animated links for the selected node).
  • Built-in dbt selector engine: names, tag:, path:, source:, package:, config.*, + / N+ / @, union, intersection, --exclude, named selectors from selectors.yml.
  • Floating selector search, filter popover (select / exclude / types), find-in-graph, legend.
  • Optional Airflow pipelines: which DAG runs a model (dbt selectors) and which DAG ingests a table (BigQuery dataset + table in DAG code).
  • Airflow source setup wizard: local folder or Google Cloud Storage (bucket browser, access test, sign-in recovery).
  • Model details dialog that edits the model's YAML in place: description, tags, config, meta, tests, columns.