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_test.go

@@ -0,0 +1,102 @@
+package httpapi
+
+import (
+	"evobgp/internal/birdfmt"
+	"evobgp/internal/store"
+	"testing"
+	"time"
+)
+
+func TestTelemetryHydrationFreshnessAndConfiguration(t *testing.T) {
+	now := time.Now().UTC()
+	sp := &store.Speaker{ID: "replica", Role: "replica", Endpoint: "old"}
+	observation := store.TelemetryObservation{ObservedAt: now.Add(-time.Hour), ReceivedAt: now, LastSuccessAt: now, Source: "agent"}
+	base := store.TelemetrySnapshot{SpeakerID: sp.ID, Kind: "remote_agent", ConfigFingerprint: store.TelemetrySpeakerFingerprint(sp), BGP: store.TelemetryBGP{TelemetryObservation: observation, Sessions: []birdfmt.BGPSession{{Name: "peer", State: "Established"}}}, Health: store.TelemetryHealth{TelemetryObservation: observation, OK: true}}
+	cases := []struct {
+		name        string
+		mutate      func(*store.TelemetrySnapshot)
+		bgp, health bool
+		quality     string
+	}{
+		{"fresh receipt despite remote clock", func(*store.TelemetrySnapshot) {}, true, true, "ok"},
+		{"expired BGP independent of health", func(s *store.TelemetrySnapshot) { s.BGP.ReceivedAt = now.Add(-2 * time.Minute) }, false, true, "stale"},
+		{"expired health independent of BGP", func(s *store.TelemetrySnapshot) { s.Health.ReceivedAt = now.Add(-2 * time.Minute) }, true, false, "ok"},
+		{"future receipt is not fresh", func(s *store.TelemetrySnapshot) { s.BGP.ReceivedAt = now.Add(time.Hour) }, false, true, "stale"},
+		{"configuration changed", func(s *store.TelemetrySnapshot) { s.ConfigFingerprint = "old" }, false, false, "unknown"},
+		{"failed BGP", func(s *store.TelemetrySnapshot) { s.BGP.Error = "timeout" }, false, true, "stale"},
+	}
+	for _, tc := range cases {
+		t.Run(tc.name, func(t *testing.T) {
+			snapshot := base
+			tc.mutate(&snapshot)
+			views, statuses := hydrateTelemetry([]*store.Speaker{sp}, []store.TelemetrySnapshot{snapshot}, now, time.Minute)
+			status := statuses[sp.ID]
+			if status["bgp_poll_ok"] != tc.bgp || status["agent_ok"] != tc.health || views[0].observation().Quality != tc.quality {
+				t.Fatalf("status=%+v observation=%+v", status, views[0].observation())
+			}
+			if !tc.bgp && (status["bgp_established"] != nil || len(views[0].Sessions) != 0) {
+				t.Fatal("stale BGP exposed live sessions")
+			}
+		})
+	}
+}
+
+func TestTelemetryHydrationMissingAndAmbiguousLocal(t *testing.T) {
+	now := time.Now()
+	sp := &store.Speaker{ID: "master", Role: "master"}
+	for _, rows := range [][]store.TelemetrySnapshot{nil, {{SpeakerID: sp.ID, Kind: "local_bird", LocalAuthority: "a"}, {SpeakerID: sp.ID, Kind: "local_bird", LocalAuthority: "b"}}} {
+		views, statuses := hydrateTelemetry([]*store.Speaker{sp}, rows, now, time.Minute)
+		if views[0].observation().Quality != "unknown" || statuses[sp.ID]["agent_ok"] == true || statuses[sp.ID]["bgp_poll_ok"] == true {
+			t.Fatalf("unexpected healthy master: %+v", statuses)
+		}
+	}
+}
+
+func TestTelemetryHydrationLocalAuthorityRestart(t *testing.T) {
+	now := time.Now()
+	sp := &store.Speaker{ID: "master", Role: "master"}
+	fresh := store.TelemetrySnapshot{SpeakerID: sp.ID, Kind: "local_bird", LocalAuthority: "new", ConfigFingerprint: store.TelemetrySpeakerFingerprint(sp), BGP: store.TelemetryBGP{TelemetryObservation: store.TelemetryObservation{ReceivedAt: now}}}
+	old := fresh
+	old.LocalAuthority = "old"
+	old.BGP.ReceivedAt = now.Add(-2 * time.Minute)
+	_, status := hydrateTelemetry([]*store.Speaker{sp}, []store.TelemetrySnapshot{old, fresh}, now, time.Minute)
+	if status[sp.ID]["bgp_poll_ok"] != true {
+		t.Fatal("expired authority hides current local snapshot")
+	}
+	old.BGP.ReceivedAt = now
+	_, status = hydrateTelemetry([]*store.Speaker{sp}, []store.TelemetrySnapshot{old, fresh}, now, time.Minute)
+	if status[sp.ID]["bgp_poll_ok"] != false {
+		t.Fatal("two current local authorities are ambiguous")
+	}
+}
+
+func TestTelemetryPeerObservationPreservesUnknownConfiguration(t *testing.T) {
+	now := time.Now()
+	sp := &store.Speaker{ID: "replica", Role: "replica"}
+	row := store.TelemetrySnapshot{SpeakerID: sp.ID, Kind: "remote_agent", ConfigFingerprint: "previous", BGP: store.TelemetryBGP{TelemetryObservation: store.TelemetryObservation{ReceivedAt: now, LastSuccessAt: now}}}
+	views, _ := hydrateTelemetry([]*store.Speaker{sp}, []store.TelemetrySnapshot{row}, now, time.Minute)
+	observation := peerLiveObservation(views)
+	if observation.Quality != "unknown" || observation.ErrorCode != "configuration_changed" {
+		t.Fatalf("changed config is not a failed current poll: %+v", observation)
+	}
+}
+
+func TestTelemetryExpiryUsesConfiguredIntervalAndEarliestSource(t *testing.T) {
+	now := time.Now().UTC()
+	old := now.Add(-30 * time.Second)
+	a := snapshotObservation(store.TelemetryObservation{ReceivedAt: old}, now, 5*time.Minute, "")
+	b := snapshotObservation(store.TelemetryObservation{ReceivedAt: now}, now, 5*time.Minute, "")
+	want := old.Add(5 * time.Minute).Format(time.RFC3339Nano)
+	if a.ExpiresAt == nil || *a.ExpiresAt != want {
+		t.Fatalf("expiry ignores configured interval: %+v", a)
+	}
+	views := []speakerBGPLive{{Polled: true, ObservationOverride: &b}, {Polled: true, ObservationOverride: &a}}
+	combined := peerLiveObservation(views)
+	if combined.ExpiresAt == nil || *combined.ExpiresAt != want {
+		t.Fatalf("aggregate extends oldest source: %+v", combined)
+	}
+	missing := snapshotObservation(store.TelemetryObservation{}, now, time.Minute, "")
+	if missing.ExpiresAt != nil {
+		t.Fatal("unobserved source has expiry")
+	}
+}