test: share the aws_accounts and aws_connect_sessions fixture tables

[GoddeyUwamari]

Description

Test-only change. No production code, migration, schema or CI workflow is touched.

The race. Six backend suites each created aws_accounts (and two of them aws_connect_sessions) when the table was missing, then dropped whatever they had created in afterAll. Jest runs suites in parallel workers against one database, so a suite finishing first could drop a table another suite was still using. This is the same pattern #171 removed for api_keys and sso_configurations.

The change. Both tables move into backend/src/routes/__tests__/shared-fixture-tables.ts and follow its existing pattern: CREATE TABLE IF NOT EXISTS inside a transaction holding pg_advisory_xact_lock, client released in finally. No suite drops either table any more. Each suite deletes only rows for the organizations it created.

Definitions. Before moving anything, every suite's CREATE TABLE was compared after whitespace normalisation:

Table	Suites defining it	Distinct definitions
aws_accounts	6	1
aws_connect_sessions	2	1

The helper carries those definitions unchanged (same columns, types, defaults, primary keys and the two named unique constraints on aws_accounts). .github/scripts/ci-bootstrap-schema.js defines neither table, so there was nothing to reconcile against it.

Affected suites

• routes/__tests__/aws-connection-funnel-event.test.ts
• routes/__tests__/aws-connection-api-key-authorization.test.ts
• services/__tests__/system-intelligence.service.cost-score.test.ts
• services/__tests__/aws-cost-freshness.test.ts
• services/__tests__/cloudwatch.service.fleet-pagination.test.ts
• services/__tests__/cloudwatch.service.concurrency.test.ts

One behavioural note: aws-connection-funnel-event.test.ts previously relied on the table drop to clear its rows. It now deletes its own aws_accounts / aws_connect_sessions rows by organization id.

Verification

All runs used a fresh local database built with .github/scripts/ci-bootstrap-schema.js (48 migrations; both tables absent afterwards). Local Postgres is 15; CI uses 14.

• Each suite alone, tables dropped first: 6/6 suites pass (68 tests).
• All six concurrently, --maxWorkers=6, 8 rounds, both tables dropped before every round: 8/8 rounds pass, 6 suites / 68 tests each. No relation ... does not exist, no duplicate-key or constraint errors, and 0 rows left in either table after every round.
• Helper under concurrent invocation: 24 separate sessions calling the helper for both tables at once, 20 rounds with the tables dropped before each: 960 calls, 0 rejected, all clients returned to their pools. The same load with a bare CREATE TABLE IF NOT EXISTS and no lock rejected 460 of 960 calls (pg_type_typname_nsp_index), which is the failure the lock prevents.
• tsc --noEmit (backend): clean.
• Full backend suite, --maxWorkers=4: 186 of 187 suites pass, 3014 of 3016 tests. The two failures are both in services/__tests__/migrate-deploy.test.ts ("Case F" symlink cases). That suite runs a shell script and uses no database; it fails the same way when run on its own and is not touched here.

Type of Change

• Bug fix (test reliability)

Checklist

• Self-review completed
• Only the shared fixture helper and the six suites above changed

🤖 Generated with Claude Code

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
devcontrol	Ready	Preview	Oct 5, 2026 1:22pm UTC