test(testing-drivers): isolate pre-aggregations schema per run to fix shared-cloud races

[paveltiunov]

Parallel CI legs (use_tesseract_sql_planner false/true) for the same
shared-cloud database pointed at the same hardcoded pre-aggregations
schema, so they raced building, refreshing, and dropping the same
pre-agg tables non-deterministically (confirmed on BigQuery; athena and
databricks-jdbc have the identical latent race).

Make the schema overridable via docker-compose interpolation, defaulting
to each fixture's existing value so local/other runs are unaffected. The
companion cube-runtime workflow sets
DRIVERS_TESTS_CUBEJS_PRE_AGGREGATIONS_SCHEMA per matrix leg so each
concurrent leg gets its own pre-aggregations schema.

Co-Authored-By: Claude Opus 4.8 noreply@anthropic.com
Claude-Session: https://claude.ai/code/session_015MhpNg5eCmC34941RWYvnw

[claude]

Claude finished @paveltiunov's task in 25s —— View job

---

Code Review

Verdict: LGTM ✅ — Targeted, minimal fix. 0 high, 0 medium, 1 low.

Details

Summary

The change swaps a hardcoded CUBEJS_PRE_AGGREGATIONS_SCHEMA value for a ${DRIVERS_TESTS_CUBEJS_PRE_AGGREGATIONS_SCHEMA:-<old default>} interpolation in three fixtures (athena, bigquery, databricks-jdbc). docker-compose ${VAR:-default} syntax is supported, so when the env var is unset behavior is identical to before — preserves local/other runs. The companion workflow can now set a unique schema per matrix leg to eliminate the shared-cloud race.

Correctness

• BigQuery default preserved as dev_pre_aggregations; athena/databricks-jdbc preserved as drivers_tests_preaggs. ✅
• No quoting issues — the entire value is a single JSON string consumed by compose interpolation.
• Three fixtures touched are exactly the shared-cloud DBs susceptible to the race; local engines (postgres/mysql/clickhouse/etc.) intentionally not changed since each spawns its own container.

Low-severity observations

• Cleanup responsibility unchanged. If a leg crashes before teardown, the per-run schema may leak in the shared cloud account. Worth a follow-up to periodically reap stale drivers_tests_preaggs_* schemas (e.g. older than 24h) on BigQuery/Athena/Databricks. Not a blocker for this PR.
• The fix relies on the companion cube-runtime workflow setting DRIVERS_TESTS_CUBEJS_PRE_AGGREGATIONS_SCHEMA uniquely per leg. Worth double-checking that PR ensures uniqueness across concurrent legs and across concurrent PRs (e.g. include run id + matrix index, not just leg name).

Testing

No unit-test changes needed — this is CI infra. Validation is the green/non-flaky CI matrix itself.

• [Branch](https://github.com/cube-js/cube/tree/claude/driver-test-preagg-schema-rd6gyd)

[claude]

Claude finished @paveltiunov's task in 28s —— View job

---

Code Review

Verdict: LGTM ✅ — Minimal, targeted CI fix. 0 high, 0 medium, 1 low.

Details

• Read PR diff
• Review correctness, security, performance
• Post review

Summary

This iteration extends the original fix (athena/bigquery/databricks-jdbc) to redshift, which was previously inheriting whatever default CUBEJS_PRE_AGGREGATIONS_SCHEMA Cube uses and was equally vulnerable to the same shared-cloud race between parallel use_tesseract_sql_planner legs. Now redshift gets an explicit ${DRIVERS_TESTS_CUBEJS_PRE_AGGREGATIONS_SCHEMA:-dev_pre_aggregations} entry consistent with the other three fixtures.

Correctness

• ${VAR:-default} interpolation is valid docker-compose syntax; when the env var is unset, behavior matches prior runs. ✅
• Defaults preserved per fixture: athena/databricks drivers_tests_preaggs, bigquery/redshift dev_pre_aggregations. ✅
• Only the four shared-cloud DBs touched; local-container DBs intentionally untouched.
• Redshift entry is correctly inserted into the environment block alongside the other CUBEJS_DB_* vars (packages/cubejs-testing-drivers/fixtures/redshift.json:25).

Security / Performance

• No secret-handling, query path, or runtime behavior change. N/A.

Low

• Schema-leak risk in shared accounts. If a leg crashes before teardown, per-run schemas leak in BigQuery/Athena/Databricks/Redshift. Not a regression introduced here, but the per-run schemes increase cardinality of leaked schemas. Worth a follow-up reaper job (e.g. drop drivers_tests_preaggs_* / dev_pre_aggregations_* older than 24h).
• Effectiveness still depends on the companion workflow setting DRIVERS_TESTS_CUBEJS_PRE_AGGREGATIONS_SCHEMA uniquely across concurrent PRs as well as legs — recommend including run id + matrix index.

Testing

CI-infra change; validation is a green non-flaky matrix.

• [Branch](https://github.com/cube-js/cube/tree/claude/driver-test-preagg-schema-rd6gyd)

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 58.80%. Comparing base (6a640a6) to head (9a83fb6).

❗ There is a different number of reports uploaded between BASE (6a640a6) and HEAD (9a83fb6). Click for more details.

HEAD has 1 upload less than BASE

Flag	BASE (6a640a6)	HEAD (9a83fb6)
cubesql	1	0

Additional details and impacted files

@@             Coverage Diff             @@
##           master   #11151       +/-   ##
===========================================
- Coverage   79.18%   58.80%   -20.38%     
===========================================
  Files         474      217      -257     
  Lines       96895    17511    -79384     
  Branches     3563     3563               
===========================================
- Hits        76727    10298    -66429     
+ Misses      19650     6695    -12955     
  Partials      518      518               

Flag	Coverage Δ
cube-backend	58.80% <ø> (ø)
cubesql	?

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[claude]

Claude finished @paveltiunov's task in 48s —— View job

---

Code Review

Verdict: LGTM ✅ — 0 high, 0 medium, 2 low.

Details

• Read PR diff (4 fixtures + parser)
• Review new ${VAR:-default} parser in runEnvironment.ts
• Check for dup/stale review threads (none)
• Post review

Summary

New commit 9a83fb6 adds the actual parser support that the fixtures depend on: runEnvironment.ts now recognizes docker-compose-style ${VAR:-default} in fixture env values. Previously only bare ${VAR} was supported, so prior iterations would have failed in any local/forked run that didn't export DRIVERS_TESTS_CUBEJS_PRE_AGGREGATIONS_SCHEMA. Closes that gap.

Correctness — runEnvironment.ts:125-146

• ${VAR} form: unchanged behavior — uses env value, throws if missing. ✅
• ${VAR:-default} form: uses env value if set & non-empty, otherwise the default. ✅
• Empty-string env var falls through to default — matches compose :- semantics (the :- form fires on null OR empty, unlike : which fires only on null). ✅
• Empty default ${VAR:-} yields '' — also matches compose. ✅
• Outer guard val.indexOf('${') === 0 && val.indexOf('}') === length - 1 still safely rejects multi-interpolation or mid-string interpolation since indexOf('}') returns the first match. So ${A}${B} is left untouched (preserving prior behavior, not new functionality). 👍
• Fixture parity: athena/databricks drivers_tests_preaggs, bigquery/redshift dev_pre_aggregations. Defaults preserve pre-change behavior. ✅

Security / Performance

N/A — CI infra only.

Low-severity observations

• name is trimmed but fallback is not (runEnvironment.ts:136-137). Compose itself doesn't trim the default either, so this is consistent — but it does mean a typo like ${VAR:- foo} would silently apply a leading-space default. Not worth fixing unless it bites someone; flagging only because the trimming asymmetry is easy to miss.
• Schema-leak risk in shared accounts (carried over from prior iterations). If a leg crashes before teardown, per-run schemas accumulate in BigQuery/Athena/Databricks/Redshift. Cardinality is now higher with the per-run scheme. Worth a follow-up reaper job (e.g. drop drivers_tests_preaggs_* / dev_pre_aggregations_* older than 24h).

Testing

CI-infra change; validation is a green non-flaky matrix across both use_tesseract_sql_planner legs.

• [Branch](https://github.com/cube-js/cube/tree/claude/driver-test-preagg-schema-rd6gyd)

[ovr]

@claude Question, does anyone setup DRIVERS_TESTS_CUBEJS_PRE_AGGREGATIONS_SCHEMA on CI? Could you point me to exact place where it's done

[claude]

Claude finished @ovr's task in 46s —— View job

---

Answer

Short version: DRIVERS_TESTS_CUBEJS_PRE_AGGREGATIONS_SCHEMA is not set anywhere in this repository. A repo-wide search turns it up in exactly four places, all consumers (the fixtures), and zero producers:

packages/cubejs-testing-drivers/fixtures/athena.json:9
packages/cubejs-testing-drivers/fixtures/bigquery.json:10
packages/cubejs-testing-drivers/fixtures/databricks-jdbc.json:76
packages/cubejs-testing-drivers/fixtures/redshift.json:25

It is not referenced anywhere under .github/ (grep over .github returns no matches).

Where it's supposed to be set

Per this PR's description, the value is set by the companion cube-runtime workflow, per matrix leg — that workflow lives in a separate (private) repo, not in cube-js/cube, which is why you can't find it here.

For comparison, the analogous knob that is wired up in this repo is DRIVERS_TESTS_CUBEJS_TESSERACT_SQL_PLANNER, set per matrix leg in the Run tests step:

.github/workflows/drivers-tests.yml:378

      - name: Run tests
        ...
        env:
          DRIVERS_TESTS_CUBEJS_TRANSPILATION_NATIVE: false
          DRIVERS_TESTS_CUBEJS_TESSERACT_SQL_PLANNER: ${{ (github.event_name == 'workflow_dispatch' && github.event.inputs.use_tesseract_sql_planner) || matrix.use_tesseract_sql_planner }}

DRIVERS_TESTS_CUBEJS_PRE_AGGREGATIONS_SCHEMA is absent from that same env: block.

Practical implications

• In cube-js/cube's own drivers-tests.yml runs, the var is currently unset → the fixtures fall back to their ${VAR:-default} defaults (dev_pre_aggregations for bigquery/redshift, drivers_tests_preaggs for athena/databricks-jdbc). Behavior is identical to before this PR.
• The race this PR targets is therefore only actually fixed once the cube-runtime workflow sets the var uniquely per leg. If that wiring isn't in place (or doesn't guarantee uniqueness across concurrent legs and concurrent PRs — e.g. run id + matrix index), the shared-cloud race remains.

If you'd like, I can add a DRIVERS_TESTS_CUBEJS_PRE_AGGREGATIONS_SCHEMA entry to the Run tests step's env: block in .github/workflows/drivers-tests.yml so it's set per leg directly in this repo — though note my GitHub App permissions don't allow modifying files under .github/workflows, so that change would need to be made by a human.
• Branch