Plan 03: Adopt before_all in DB-Heavy Test Files

Objective

Convert eligible DB-heavy test files from per-test setup to setup_once/before_all for shared fixture creation. The setup_once helper (alias for before_all) is already wired up in test/test_prof.rb but only used in 1 of 54 eligible files. This saves ~3-5s by eliminating redundant database INSERT/DELETE cycles.

Context

• @ test/test_prof.rb -- setup_once (alias for before_all) already configured and included in ActiveSupport::TestCase
• @ test/lib/source_monitor/logs/query_test.rb -- only existing user of setup_once (reference pattern)
• @ test/lib/source_monitor/analytics/sources_index_metrics_test.rb -- 17 tests, shared read-only fixtures. PRIMARY candidate: creates 3 sources + 3 items in setup, all tests only query this data.
• @ test/lib/source_monitor/analytics/source_activity_rates_test.rb -- 1 test, uses clean_source_monitor_tables!
• @ test/lib/source_monitor/analytics/source_fetch_interval_distribution_test.rb -- 1 test, uses clean_source_monitor_tables!
• @ test/lib/source_monitor/dashboard/upcoming_fetch_schedule_test.rb -- 1 test, uses clean_source_monitor_tables!

Safety analysis performed:

• sources_index_metrics_test.rb: SAFE. All 17 tests construct SourcesIndexMetrics.new(...) and call read-only query methods. No test creates, updates, or deletes records.
• source_activity_rates_test.rb: SAFE but minimal benefit (1 test, setup runs once either way).
• source_fetch_interval_distribution_test.rb: SAFE but minimal benefit (1 test).
• upcoming_fetch_schedule_test.rb: SAFE but minimal benefit (1 test).
• dashboard/queries_test.rb: NOT SAFE. Each test creates its own sources and checks specific counts. Shared state would cause pollution.
• health/source_health_monitor_test.rb: NOT SAFE. Tests mutate @source via SourceHealthMonitor.call.
• items/item_creator_test.rb: NOT SAFE. Tests create items on shared source and check counts.

Rationale: before_all wraps fixture creation in a SAVEPOINT, shared across all tests in the class. After all tests run, the savepoint rolls back. This only works when tests are read-only on the shared data. The sources_index_metrics_test.rb is the highest-value candidate with 17 read-only tests sharing the same 3 sources + 3 items.

Tasks

Task 1: Convert sources_index_metrics_test.rb to setup_once (PRIMARY)

This is the highest-impact conversion. Convert test/lib/source_monitor/analytics/sources_index_metrics_test.rb:

Replace:

setup do
  clean_source_monitor_tables!
  travel_to Time.current.change(usec: 0)
  @fast_source = create_source!(name: "Fast", fetch_interval_minutes: 30)
  # ... fixture creation
end

With:

setup_once do
  clean_source_monitor_tables!
  @fast_source = create_source!(name: "Fast", fetch_interval_minutes: 30)
  # ... same fixture creation, but now runs once for all 17 tests
end

Important: The travel_to call must stay in a regular setup block because travel_to affects the thread-local time for each test independently:

setup_once do
  clean_source_monitor_tables!
  # fixture creation here
end

setup do
  travel_to Time.current.change(usec: 0)
end

teardown do
  travel_back
end

Wait -- travel_to inside setup_once would freeze time for the SAVEPOINT transaction but tests need consistent time for assertions. Actually, the fixtures are created with relative timestamps (1.day.ago, 2.days.ago) which depend on Time.current. If travel_to is in setup_once, the timestamps are fixed at creation time, which is fine since tests read them as-is. But travel_back in teardown would only run once after all tests, and the travel_to in setup_once persists through all tests.

Safest approach: Move travel_to into setup_once and remove the teardown's travel_back (before_all handles cleanup). Add a regular setup with travel_to at the same frozen time to ensure each test sees consistent time.

Actually, the simplest safe approach: keep travel_to and travel_back in regular setup/teardown, and only put the DB operations in setup_once. The fixtures use relative timestamps (1.day.ago) which will be slightly different each test, but since the tests only compare relative values (bucket labels, activity rates), this is fine.

Task 2: Convert single-test analytics files to setup_once

Convert these 3 files for consistency (minimal performance benefit but establishes the pattern):

1. test/lib/source_monitor/analytics/source_activity_rates_test.rb -- Replace setup { clean_source_monitor_tables! } with setup_once { clean_source_monitor_tables! }
2. test/lib/source_monitor/analytics/source_fetch_interval_distribution_test.rb -- Same pattern
3. test/lib/source_monitor/dashboard/upcoming_fetch_schedule_test.rb -- Same pattern

For single-test classes, setup and setup_once are functionally identical, so this is a no-op in terms of performance but normalizes the codebase to use the setup_once pattern for table cleaning.

Task 3: Verify all converted files individually

Run each converted file with PARALLEL_WORKERS=1 to confirm no regressions:

PARALLEL_WORKERS=1 bin/rails test test/lib/source_monitor/analytics/sources_index_metrics_test.rb
PARALLEL_WORKERS=1 bin/rails test test/lib/source_monitor/analytics/source_activity_rates_test.rb
PARALLEL_WORKERS=1 bin/rails test test/lib/source_monitor/analytics/source_fetch_interval_distribution_test.rb
PARALLEL_WORKERS=1 bin/rails test test/lib/source_monitor/dashboard/upcoming_fetch_schedule_test.rb

If any file fails due to test isolation issues, revert it to per-test setup and document why.

Task 4: Full suite verification and lint

# Full suite (all 1031+ tests pass)
bin/rails test

# Lint all modified files
bin/rubocop test/lib/source_monitor/analytics/ test/lib/source_monitor/dashboard/upcoming_fetch_schedule_test.rb

Ensure total test count remains 1031+ and no failures occur.

Files

Verification

# Individual file runs
PARALLEL_WORKERS=1 bin/rails test test/lib/source_monitor/analytics/sources_index_metrics_test.rb
PARALLEL_WORKERS=1 bin/rails test test/lib/source_monitor/dashboard/upcoming_fetch_schedule_test.rb

# Full suite (all 1031+ tests pass)
bin/rails test

# Lint
bin/rubocop test/lib/source_monitor/analytics/ test/lib/source_monitor/dashboard/upcoming_fetch_schedule_test.rb

Success Criteria

• grep -r "setup_once" test/lib/source_monitor/ shows 5+ files (up from 1)
• sources_index_metrics_test.rb uses setup_once for fixture creation
• All 1031+ tests pass in full suite
• No test isolation regressions in parallel runs
• Each converted file passes individually with PARALLEL_WORKERS=1