Enable APM for 30% of requests, add cleanup cron, rename admin URL, and send user ID

[cycomachead]

General Info

• Pivotal Tracker Story
• Breaking?

Changes

Configures Faultline per the ticket: renames the dashboard URL, enables sampled APM, adds cleanup for retained data, and ensures a basic user ID is sent with every error.

• Dashboard URL: mounted at /admin/faultline (was /admin/errors); ignore lists and route helpers updated accordingly.
• APM enabled at 30% sample rate with 30-day trace retention. APM ignore paths now skip the load balancer health check (/status/health_check) and the dashboard itself.
• FaultlineCleanupJob (renamed from an APM-only job): Faultline records retention_days/apm_retention_days but never enforces them itself. This job now nightly:
  • Deletes APM traces and attached CPU profiles past apm_retention_days.
  • Deletes error occurrences (and their context rows) past retention_days, recomputes the occurrences_count counter cache on affected groups, and removes groups left empty — except groups marked ignored, so a recurrence doesn't silently open a fresh alerting group.
  • Runs via the existing GoodJob cron (config/application.rb, key faultline_cleanup) nightly at 3:30 AM Pacific.
• User ID: added a custom_context that attaches the Canvas uid from the session to every occurrence, in addition to Faultline's existing current_user tracking, so a user is identifiable even when no controller instance is available.
• Bug fix found along the way: unhandled request exceptions were being recorded twice — once by Faultline's middleware (with request/user context) and again by Rails' executor re-reporting to the error subscriber (source application.action_dispatch, no user/URL context). Added a before_track hook to drop that duplicate report. Job errors and explicit Rails.error calls are unaffected.

Testing

• spec/config/faultline_spec.rb: pins mount path, ignore lists, custom context, dedupe hook, APM settings and retention values.
• spec/config/good_job_cron_spec.rb: verifies the faultline_cleanup cron schedule.
• spec/jobs/faultline_cleanup_job_spec.rb: covers stale vs. fresh traces/profiles, occurrence + context deletion, counter cache recompute, empty-group removal, the ignored-group exception, and retention_days: nil (keep forever).
• spec/requests/faultline_user_identification_spec.rb: signs in through the real OmniAuth flow, forces a controller exception, and asserts exactly one occurrence is recorded with the correct user ID, email, and Canvas uid — plus the anonymous case.
• Full RSpec suite and rubocop pass; also smoke-tested the new /admin/faultline URL (401 for anonymous, both error and performance pages render for admins).

Documentation

Updated docs/developers.md: added faultline_cleanup to the cron table and documented what the job enforces and why (Faultline stores retention settings but doesn't act on them).

Checklist

• Name of branch corresponds to story

---

Superconductor Ticket Implementation | App Preview | Guided Review