Description
Extend httpRequestsAdaptiveGroups metrics beyond Access-protected hosts to every zone, with cardinality limits.
Acceptance Criteria
- #1 Configurable zone scope with cardinality thresholds
Definition of Done
- #1 just check (fmt-check, lint, vet, test, tidy-check, build, vuln)
- #2 just ci before a change that touches the Dockerfile, goreleaser or the image (adds snapshot + image)
- #3 Every new signal or attribute name declared in internal/semconv and listed in docs/signals.md
Implementation Plan
Add a separate metrics scope that inherits the existing HTTP scope by default, with configurable per-zone host and per-window series limits. Implement all-zone Groups metrics without expanding per-request events, then verify limits and live signal.
Implementation Notes
Prebuild discovery: account zone listing succeeded and a bounded httpRequestsAdaptiveGroups sample returned nonzero count rows. Sample payload is private and excluded from the tracker.
Loop 3 source at 20e9981 adds http.metrics_scope (empty inherits the prior HTTP scope), a 1000-host per-zone cap and 10000-series per-window cap. All-zone metrics remain separate from request-event scope; discovered zones are restricted to the configured account before explicit selectors; cap or incomplete-zone failures leave the checkpoint unchanged. Prechange regressions failed; exact-SHA just check, independent REV and CodeRabbit completed. Read-only local harness queried 23 configured-account zones in [07:34,07:49) UTC, emitted 31 request points totaling 3895 and four optional-duration points; 33 provider -1 duration sentinels were omitted from gauges while requests were retained. No Camden config change or deployed all-zone OTLP/m7kni signal occurred. Resume by releasing/deploying source, explicitly selecting metrics_scope=all in approved Camden config, then verify checkpoint and m7kni request counts without broadening events. Keep task Parked until that live proof.
Loop 4 live proof: Camden v0.4.0 restarted at 14:15:34 UTC with only http.metrics_scope: all added; config remained 0600 65532:65532, healthy, no checkpoint rewind, no cap or config error. After two full five-minute post-restart windows and httpreq.metrics checkpoint 14:34:23Z, Mimir count(count by (cloudflare_http_zone)(max_over_time(cloudflare_http_requests_total[10m]))) at 14:30Z was 11 versus 2 in equal pre-CFG [14:00,14:10) window. Post had 29 active series, below 10000, and zero increase in cf2otel_window_commit_failures_total for httpreq.metrics. Read-only httpRequestsAdaptiveGroups source for post-restart [14:15:34,14:34:23) totaled 2230 requests across 23 queried account zones; Mimir sum(max_over_time(cloudflare_http_requests_total[20m])) at 14:40Z, after delivery, was 2260 (1.35% difference). The earlier increase query at source-window end undercounted sparse newly created series before exporter delivery; it is not the source comparison. No cap-breach log line.
Final Summary
Camden now runs all-zone HTTP metrics on v0.4.0. Live Mimir zone coverage rose from 2 to 11 on equal ten-minute windows, with 29 series, zero commit failures, and a source/counter-level comparison within 1.35% after delivery.