Introduce a centralized OTel layer (app/telemetry/otel_setup.py) that reuses
the existing OTLP endpoint and token, exports traces and metrics over OTLP/HTTP,
and instruments Flask plus SQLAlchemy. Manual OTLP log export remains for base
and product analytics; log records now include trace_id, span_id, and
event_category where tracing is active.
Business spans and product metrics cover invoices, timers, reports, auth,
webhook delivery, and scheduled jobs. RED-style HTTP metrics are recorded in
after_request alongside existing Prometheus counters. ENABLE_TRACING and
ENABLE_METRICS default on when credentials exist; graceful no-op when they do not.
Privacy is preserved: user_id appears on traces only when detailed analytics is
opted in; metrics never carry user_id; _remove_pii behavior for analytics is
unchanged. Responses inject traceparent when tracing is enabled for future
browser correlation.
Tests: test_otel_integration.py and per-test reset_for_testing() in conftest
so each app factory can reinitialize OTel.
Read telemetry/sentry status in the admin dashboard from build-injected analytics defaults with env fallback, so release images correctly show OTLP configuration. Include both install_id and telemetry_fingerprint in base and detailed telemetry payloads to reliably differentiate installations, and add tests for the new fingerprint fields.
Switch product and installation telemetry to OTLP/Grafana across runtime config, CI injection, docs, and tests to unify telemetry transport and simplify privacy-focused opt-in behavior.
Dries Peeters