| 1 | package telemetry |
| 2 | |
| 3 | // EmitFromTrace updates relevant Prometheus metrics from a completed |
| 4 | // SelectionTrace. It is safe to call concurrently. |
| 5 | // |
| 6 | // Metrics updated: |
| 7 | // - relay_selected_total{relay, reason} — one increment per OutputURL. |
| 8 | // - selection_duration_seconds — one observation for the whole invocation. |
| 9 | // - congestion_mode — set according to Congested + NonLinear. |
| 10 | // - selection_skipped_total{reason} — one increment per suppressed URL that |
| 11 | // has a reason entry. |
| 12 | // - rtt_seconds{relay} — one observation per Ranked entry with non-zero RTT. |
| 13 | // |
| 14 | // Metrics NOT updated here (wired by later phases / other code paths): |
| 15 | // - relay_pool_size — set by RelaySet pool management. |
| 16 | // - active_tunnels_per_relay — incremented/decremented at tunnel accept/close. |
| 17 | // - failures_total — incremented on discovery/active failure events. |
| 18 | func EmitFromTrace(t SelectionTrace) { |
| 19 | // --- relay_selected_total --- |
| 20 | reason := selectionReason(t) |
| 21 | for _, url := range t.OutputURLs { |
| 22 | RelaySelectedTotal.WithLabelValues(BoundedRelay(url), reason).Inc() |
| 23 | } |
| 24 | |
| 25 | // --- selection_duration_seconds --- |
| 26 | SelectionDurationSeconds.Observe(t.SelectionTook.Seconds()) |
| 27 | |
| 28 | // --- congestion_mode --- |
| 29 | CongestionMode.Set(congestionModeValue(t.Congested, t.NonLinear)) |
| 30 | |
| 31 | // --- selection_skipped_total --- |
| 32 | // Build suppressed set for O(1) lookup. |
| 33 | suppressedSet := make(map[string]struct{}, len(t.Suppressed)) |
| 34 | for _, url := range t.Suppressed { |
| 35 | suppressedSet[url] = struct{}{} |
| 36 | } |
| 37 | for url, reason := range t.Reasons { |
| 38 | if _, ok := suppressedSet[url]; ok { |
| 39 | SelectionSkippedTotal.WithLabelValues(reason).Inc() |
| 40 | } |
| 41 | } |
| 42 | |
| 43 | // --- rtt_seconds --- |
| 44 | for _, entry := range t.Ranked { |
| 45 | if entry.RTT != 0 { |
| 46 | RTTSeconds.WithLabelValues(BoundedRelay(entry.URL)).Observe(entry.RTT.Seconds()) |
| 47 | } |
| 48 | } |
| 49 | } |
| 50 | |
| 51 | // selectionReason derives the reason label for relay_selected_total from the |
| 52 | // trace flags. Explicit/fallback semantics are wired by later phases; this |
| 53 | // function defaults to "auto" for uninstrumented call sites. |
| 54 | func selectionReason(t SelectionTrace) string { |
| 55 | switch { |
| 56 | case t.NonLinear: |
| 57 | return "variant-grid" |
| 58 | case t.Congested: |
| 59 | return "congestion-promoted" |
| 60 | default: |
| 61 | return "auto" |
| 62 | } |
| 63 | } |
| 64 | |
| 65 | // congestionModeValue maps the Congested + NonLinear pair to the metric value. |
| 66 | // 0 = normal, 1 = congested without variant-grid, 2 = variant-grid active. |
| 67 | func congestionModeValue(congested, nonLinear bool) float64 { |
| 68 | switch { |
| 69 | case nonLinear: |
| 70 | return 2 |
| 71 | case congested: |
| 72 | return 1 |
| 73 | default: |
| 74 | return 0 |
| 75 | } |
| 76 | } |