Skip to content

OpenDBAPI 0.2.0

Choose a tag to compare

@viethqb viethqb released this 17 Nov 15:21
· 12 commits to main since this release
855530e

Configuration as code, API orchestration, and more ways to run OpenDBAPI in production.

Run it

docker pull ghcr.io/egret-data/opendbapi:0.2.0
helm install opendbapi deploy/helm/opendbapi --set database.url=jdbc:postgresql://… --set existingSecret=…

Upgrading from 0.1.0: Flyway migrates the schema on start (V13–V17).

Configuration as code

  • Plan / apply a workspace's YAML (POST /config:plan, /config:apply): plan shows creates, updates (with the differing fields), deletes (with prune); apply writes everything in one transaction, then publishes endpoints marked live only after their tests pass. An export always plans as unchanged.
  • Environments map the file's datasource names to each workspace's datasources, so one file serves dev, staging and prod.
  • opendbapi command line in the image: docker run ghcr.io/egret-data/opendbapi:0.2.0 plan|apply|export ….
  • API tokens (user menu → API tokens): Authorization: Bearer odb_…, optionally limited to one workspace and given an expiry; the CLI reads OPENDBAPI_TOKEN. The README has a GitHub Actions example.
  • Lock a workspace to configuration files: endpoints, groups, service accounts and data policies then change through apply only (opendbapi.ui.read-only locks every workspace).
  • Major versions and deprecation travel in the files too.
  • UI: Settings → Import & export → Apply YAML, and Environments.

Endpoints

  • Major versions side by side: New major version… copies an endpoint to v2/…; deprecated endpoints answer with Deprecation, Sunset and a link to the newest live major; validation warns about breaking changes.
  • Groovy full mode: scripts run their own queries with db.one / list / value / update and return any answer; every query still goes through the static checks, row filter, masks and limits.
  • Workflows (DBAPI-style orchestration) on a drag-and-drop canvas: SQL and script nodes, Call endpoint nodes that reuse other endpoints, Decision nodes with Yes/No branches, End nodes per branch; independent nodes run in parallel; called endpoints on the same datasource join the workflow's transaction.
  • GET endpoints never change data, neither through scripts nor through the endpoints they call.

Secrets

  • HashiCorp Vault references vault:PATH#KEY (token, AppRole or Kubernetes auth), each workspace confined to its own paths.

Running several instances

  • Rate limits can be shared across pods (OPENDBAPI_RATE_LIMIT_SHARED, on in the Helm chart).

Fixes

  • Migration V14 now backfills endpoint lineages on databases that already have endpoints (it failed there before; nothing was half-applied). If an empty database ran the earlier V14 from a main image, run flyway repair.
  • The image build retries when Maven Central throttles.