Data & Analytics
A green dbt test can mean zero tests ran after inline show
A reported dbt v2 partial-parse state leak can make selected tests run nothing while CI exits successfully.
November 17, 2026
The dangerous dbt failure is a green check with no tests behind it. In dbt issue #17162, a reporter reproduces dbt v2 partial-parse state retaining an inline node after dbt show --inline. Later dbt test --select my_model warns that the inline selection matches nothing and exits 0. The report is version-specific and remains in triage.
Inspect what was selected, not just the process status
Capture the selected-node count, test count and warning stream for every CI run. A zero-test result should fail a gate that expected model tests, even if dbt exits successfully. The issue also reports dbt ls, compile and run targeting the stale inline node; compare actual selected unique IDs with the requested selector before trusting subsequent artifacts.
Reproduce with a clean state boundary
In a disposable project, run dbt show --inline with partial parsing on, then dbt ls and dbt test for a known model. Repeat after dbt clean and with partial parsing disabled. dbt's parse documentation explains the cached parse path and the full-parse option. The issue's reporter found deleting only manifest.json insufficient, so record which state files were removed.
Protect the release pipeline
Until your dbt version passes the regression, isolate exploratory inline queries from the CI workspace, or clean parse state before the test job. Add a fixture with a known failing test so the job proves tests actually execute. Keep a version pin and a rollback route; do not apply an unreviewed workaround to every project based on one issue report.
Adovo’s view
A test is evidence only when it ran. Data quality gates should assert their own coverage, just as a deployment gate asserts that it deployed the intended artifact. Exit code alone is too small a contract for production data.
