Issue #160: da query --remote against query_mode='remote' BQ rows whose
underlying entity is a VIEW or MATERIALIZED_VIEW returned a DuckDB catalog
error because the extractor (with legacy_wrap_views=False default since
the v2 fetch primitives release) skipped master-view creation for those
entity types — but kept inserting the _meta row, leaving operators with a
registered name that resolves to nothing.
Always create a master view for entity types we have proven runtime support
for in this codebase:
BASE TABLE → bq."<dataset>"."<source_table>"
(Storage Read API path; predicate pushdown)
VIEW / MAT_VIEW → bigquery_query('<project>', 'SELECT * FROM `proj.ds.tbl`')
(jobs API path; no pushdown — the upcoming /api/query
cost guardrail bounds the scan; was the legacy
legacy_wrap_views=True branch SQL form, just always-on)
For other entity types (EXTERNAL, SNAPSHOT, CLONE, future), log a warning
and SKIP both the master view AND the _meta row. The registry row remains
intact so /api/v2/scan still works for `da fetch`; we just don't expose a
stale _meta entry that the orchestrator would later strand.
The legacy_wrap_views config knob is still readable in this commit (read
returns the value, which is then ignored). Removal across the rest of
the codebase happens in the follow-up REFACTOR commit.
tests/test_bigquery_extractor.py:
- Add 3 RED tests covering the new always-wrap behavior:
test_view_creates_wrap_view_with_default_config,
test_materialized_view_creates_wrap_view_with_default_config,
test_unsupported_entity_type_skips_meta_and_view.
- Fix pre-existing flakiness in test_main_exits_when_project_missing
by resetting app.instance_config cache before the no-project mock —
the prior test populates the cache with a project, and removing the
legacy_wrap_views get_value() call surfaced this latent ordering bug.
Replaces the BigQuery wrap-view pattern with a discovery + scoped-fetch toolkit driven by the analyst's Claude session. Adds /api/v2/{catalog,schema,sample,scan,scan/estimate}, da catalog/schema/describe/fetch/snapshot/disk-info CLI commands, sqlglot-backed WHERE validator, process-local quota tracker, agent rails skill (cli/skills/agnes-data-querying.md). BREAKING: BQ wrap views off by default — set data_source.bigquery.legacy_wrap_views=true for one cycle. Backward-compat field_validator on primary_key. Catalog cache now matches documented 300s TTL with RBAC fresh per request. Cuts release v0.14.0.
- Add close_system_db() function in src/db.py to cleanly close shared DB connection
- Add lifespan context manager in app/main.py to trigger shutdown on app exit
- Integrate lifespan into FastAPI app initialization
- All API tests pass (77/77)
BigQuery extension handles auth via GOOGLE_APPLICATION_CREDENTIALS env var,
so _remote_attach uses empty token_env. Orchestrator now supports both
token-based (Keboola) and env-based (BigQuery) authentication modes.
