feat(httpapi): harden control plane and speed up prefix refresh

Очередь job получила lease и fencing, телеметрия и прогресс deploy сохраняются, ревизия пишет manifest входов. Обновление префиксов не переписывает неизменённый снимок, большие наборы идут через COPY, а устаревшая ревизия не попадает в deploy. Бандл ноды кэшируется, prune и постановка job больше не держат горячий путь.

Co-authored-by: Cursor <cursoragent@cursor.com>

internal/httpapi/telemetry.go

@@ -0,0 +1,154 @@
+package httpapi
+
+import (
+	"context"
+	"net/http"
+	"time"
+
+	"evobgp/internal/nodedispatch"
+	"evobgp/internal/store"
+	"evobgp/internal/telemetry"
+)
+
+func (s *Server) loadTelemetry(ctx context.Context, tenant string, speakers []*store.Speaker) ([]speakerBGPLive, map[string]map[string]any, error) {
+	ctx, cancel := context.WithTimeout(ctx, 2*time.Second)
+	defer cancel()
+	snapshots, err := s.store.ListTelemetry(ctx, tenant)
+	if err != nil {
+		return nil, nil, err
+	}
+	views, statuses := hydrateTelemetry(speakers, snapshots, time.Now(), telemetry.ConfigFromEnv().StaleAfter)
+	return views, statuses, nil
+}
+
+// snapshotObservation uses control-plane receipt time, never the remote clock.
+func snapshotObservation(o store.TelemetryObservation, now time.Time, staleAfter time.Duration, invalid string) liveObservation {
+	v := liveObservation{ObservedAt: liveTime(o.ObservedAt), ReceivedAt: liveTime(o.ReceivedAt), LastSuccessAt: liveTime(o.LastSuccessAt), Source: o.Source, Quality: "unknown"}
+	if !o.ReceivedAt.IsZero() {
+		v.ExpiresAt = liveTime(o.ReceivedAt.Add(staleAfter))
+	}
+	if v.Source == "" {
+		v.Source = "none"
+	}
+	switch {
+	case invalid != "":
+		v.ErrorCode = invalid
+	case o.ReceivedAt.IsZero():
+		v.ErrorCode = "not_polled"
+	case o.ReceivedAt.After(now) || now.Sub(o.ReceivedAt) > staleAfter:
+		v.Quality, v.ErrorCode = "stale", "snapshot_expired"
+	case o.Error != "":
+		v.Quality, v.ErrorCode = "unreachable", liveErrorCode(o.Error)
+		if !o.LastSuccessAt.IsZero() {
+			v.Quality = "stale"
+		}
+	default:
+		v.Quality = "ok"
+	}
+	return v
+}
+
+func hydrateTelemetry(speakers []*store.Speaker, snapshots []store.TelemetrySnapshot, now time.Time, staleAfter time.Duration) ([]speakerBGPLive, map[string]map[string]any) {
+	byID := make(map[string]*store.Speaker, len(speakers))
+	for _, sp := range speakers {
+		if sp != nil {
+			byID[sp.ID] = sp
+		}
+	}
+	grouped := make(map[string][]store.TelemetrySnapshot)
+	for _, snapshot := range snapshots {
+		if snapshot.SpeakerID != "" && byID[snapshot.SpeakerID] == nil {
+			continue
+		}
+		if snapshot.SpeakerID == "" && snapshot.Kind != "local_bird" {
+			continue
+		}
+		grouped[snapshot.SpeakerID] = append(grouped[snapshot.SpeakerID], snapshot)
+	}
+	views := make([]speakerBGPLive, 0, len(speakers)+1)
+	statuses := make(map[string]map[string]any, len(speakers))
+	appendView := func(id string, sp *store.Speaker) {
+		view := speakerBGPLive{SpeakerID: id, Label: speakerDisplayLabel(sp)}
+		if sp == nil {
+			view.Label = "CP (local)"
+		}
+		var health *agentHealthPoll
+		rows := grouped[id]
+		// A restarted local collector may leave an expired authority behind.
+		// Only simultaneously current authorities conflict; never prefer one
+		// of two current BIRD nodes arbitrarily.
+		if len(rows) > 1 {
+			current := make([]store.TelemetrySnapshot, 0, len(rows))
+			for _, row := range rows {
+				if row.Kind == "local_bird" && row.ConfigFingerprint == store.TelemetrySpeakerFingerprint(sp) && !row.BGP.ReceivedAt.IsZero() && !row.BGP.ReceivedAt.After(now) && now.Sub(row.BGP.ReceivedAt) <= staleAfter {
+					current = append(current, row)
+				}
+			}
+			if len(current) > 0 {
+				rows = current
+			}
+		}
+		if len(rows) > 1 {
+			view.Error = "ambiguous telemetry authorities"
+			view.ObservationOverride = &liveObservation{Source: "none", Quality: "unknown", ErrorCode: "ambiguous_authority"}
+		} else if len(rows) == 1 {
+			row := rows[0]
+			invalid := ""
+			if row.ConfigFingerprint != store.TelemetrySpeakerFingerprint(sp) {
+				invalid = "configuration_changed"
+			}
+			bgp := snapshotObservation(row.BGP.TelemetryObservation, now, staleAfter, invalid)
+			view.ObservationOverride = &bgp
+			view.Polled = !row.BGP.ReceivedAt.IsZero()
+			view.ObservedAt, view.ReceivedAt, view.LastSuccessAt, view.Source = row.BGP.ObservedAt, row.BGP.ReceivedAt, row.BGP.LastSuccessAt, row.BGP.Source
+			if bgp.Quality == "ok" {
+				view.Sessions = row.BGP.Sessions
+			} else {
+				view.Error = bgp.ErrorCode
+			}
+			if row.Kind == "remote_agent" {
+				h := snapshotObservation(row.Health.TelemetryObservation, now, staleAfter, invalid)
+				result := nodedispatch.AgentHealthResult{OK: row.Health.OK && h.Quality == "ok", Error: row.Health.Error, LastSyncAt: row.Health.LastSyncAt, LastAppliedRevisionID: row.Health.LastAppliedRevisionID}
+				if h.Quality != "ok" {
+					result.Error = h.ErrorCode
+				}
+				if h.Quality == "ok" && !result.OK {
+					h.Quality, h.ErrorCode = "unreachable", "reported_unhealthy"
+				}
+				health = &agentHealthPoll{result: result, observationOverride: &h}
+			}
+		}
+		views = append(views, view)
+		if sp != nil {
+			statuses[id] = speakerLiveStatusWithPoll(sp, view, health)
+		}
+	}
+	for _, sp := range speakers {
+		if sp != nil {
+			appendView(sp.ID, sp)
+		}
+	}
+	if len(grouped[""]) > 0 {
+		appendView("", nil)
+	}
+	return views, statuses
+}
+
+func (s *Server) handleTelemetryRefresh(w http.ResponseWriter, r *http.Request) {
+	a, ok := authFromContext(r.Context())
+	if !ok {
+		writeProblem(w, http.StatusUnauthorized, "Unauthorized", "missing auth")
+		return
+	}
+	if !s.requirePerm(w, a, "bgp:network:read") {
+		return
+	}
+	ctx, cancel := context.WithTimeout(r.Context(), 2*time.Second)
+	defer cancel()
+	requested, err := s.store.RequestTelemetryRefresh(ctx, a.TenantID, 10*time.Second)
+	if err != nil {
+		writeProblem(w, http.StatusServiceUnavailable, "Service Unavailable", "telemetry refresh unavailable")
+		return
+	}
+	writeJSON(w, http.StatusAccepted, map[string]any{"requested": requested})
+}