Adovo

Data & Analytics

dbt seed fails on a legacy BigQuery domain project

Isolate a seed-only BigQuery project-ID failure during a dbt Fusion migration.

October 31, 2026

A dbt migration can pass compilation and model runs yet fail when it loads a small reference CSV. A dbt issue reports that Fusion 2.0.6 produces invalid table reference for dbt seed against a legacy BigQuery project ID shaped like domain.com:project-id. In the same environment, models, tests, snapshots, dbt debug, and dbt compile reportedly succeed. The reporter could not reproduce the issue internally because new domain-scoped projects cannot be created; the existing customer environment is the evidence.

Check the failing path

Capture the target project ID, adapter and dbt versions, seed name, fully qualified relation, and exact warehouse error. Run only the affected seed in a nonproduction target and compare it with a simple model against the same dataset. The dbt seed guide describes seeds as small, version-controlled CSVs loaded into warehouse tables. Their load path can differ from a model's SQL compilation path, so a successful dbt compile does not prove that the seed can be created.

Keep reference data available

List the downstream models that depend on the seed. If the migration window is near, keep the known-working dbt runtime for that target or stage an approved alternative way to load the reference table. Do not rename a legacy GCP project or edit every model to work around one unconfirmed seed path. Protect the data contract first, then test a specific fix or supported release.

Ask for a narrow reproduction

When raising the issue with a vendor, include the project-ID format, a minimal CSV, seed configuration, generated relation name, and comparison to a working model. Mark that the behavior is reported for Fusion 2.0.6 and BigQuery. A test on a modern non-domain project is useful as a control, but it cannot disprove the legacy case.

Adovo’s view

Migration parity is measured at the smallest path that still carries business meaning. A three-row seed can stop a large build if the runtime cannot name its target. Test the odd legacy names before the cutover date, while the old path is still available.

Continue exploring

Data & Analytics

dbtMicrobatch

A dbt microbatch rebuild fails before it can repair history

A reported dbt v2 parser error can block a microbatch full refresh. Protect the recovery path before changing a production model.

Read more

Data & Analytics

dbtCI

Before running dbt clean in CI, audit paths outside the project

Check clean-targets and version-specific safety behavior before a CI cleanup step removes or skips files outside the dbt project.

Read more

Data & Analytics

dbtDatabricks

A Databricks materialized view breaks after an upstream column appears

Spot query-schema drift in dbt materialized views before a refresh failure surprises downstream users.

Read more
Explore data and analytics services →