divviup / divviup-api / 26540494363

69%

Build Type

push

github

Committed by web-flow

Commit Message

Migrate from Trillium [part 10]: OTel 0.32 upgrade, OTLP metrics, HTTP metrics middleware (#2278)

Upgrade the OpenTelemetry ecosystem from 0.27 to 0.32 and switch from
Prometheus scrape-based metrics export to OTLP push. Prometheus 3.x in
prod supports native OTLP ingestion, so the prometheus and
opentelemetry-prometheus crates are removed entirely.

Changes:

- Upgrade opentelemetry, opentelemetry_sdk, and opentelemetry-otlp to
  0.32. Use the new Resource::builder() and SdkMeterProvider APIs.

- Replace the Prometheus exporter with an OTLP HTTP MetricExporter
  using periodic push (endpoint via OTEL_EXPORTER_OTLP_ENDPOINT).

- Remove the /metrics endpoint from the monitoring server; it now only
  serves /traceconfig.

- Add a custom HTTP metrics middleware (src/handler/http_metrics.rs)
  that records http.server.request.duration and
  http.server.active_requests following OTel HTTP semantic conventions,
  including url.scheme, http.request.method (normalized to known
  methods or _OTHER), http.route (omitted for unmatched paths), and
  http.response.status_code.

- Propagate client IP into trace spans by parsing the rightmost
  X-Forwarded-For value in a custom TraceLayer make_span_with.

- Return a TelemetryProviders struct from install_telemetry() so both
  the meter provider and the OTLP tracer provider (when otlp-trace is
  enabled) are shut down gracefully on exit.

- Remove stale RUSTSEC advisory ignores from deny.toml.

- Use tracing::field::Empty for client.address when XFF header absent
- Add OTel semconv advisory histogram buckets to request duration
- Comment on HttpMetrics::new() to prevent per-request re-registration

Coverage Stats

78 of 116 new or added lines in 4 files covered. (67.24%)

2 existing lines in 2 files now uncovered.

4500 of 6502 relevant lines covered (69.21%)

71.64 hits per line