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_http_test.go

@@ -0,0 +1,126 @@
+package httpapi
+
+import (
+	"context"
+	"errors"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+
+	"evobgp/internal/nodedispatch"
+	"evobgp/internal/store"
+)
+
+type telemetryHTTPStore struct {
+	store.Backend
+	discoveryWrites int
+	snapshotErr     error
+	refreshCalls    int
+}
+
+func (s *telemetryHTTPStore) UpsertPeerDiscovery(tenant string, in *store.PeerDiscoveryUpsert) (*store.BGPPeerDiscovery, error) {
+	s.discoveryWrites++
+	return s.Backend.UpsertPeerDiscovery(tenant, in)
+}
+func (s *telemetryHTTPStore) ListTelemetry(ctx context.Context, tenant string) ([]store.TelemetrySnapshot, error) {
+	if s.snapshotErr != nil {
+		return nil, s.snapshotErr
+	}
+	return s.Backend.ListTelemetry(ctx, tenant)
+}
+func (s *telemetryHTTPStore) RequestTelemetryRefresh(ctx context.Context, tenant string, cooldown time.Duration) (bool, error) {
+	s.refreshCalls++
+	return s.Backend.RequestTelemetryRefresh(ctx, tenant, cooldown)
+}
+func telemetryRequest(method, path, tenant string) *http.Request {
+	r := httptest.NewRequest(method, path, nil)
+	return r.WithContext(context.WithValue(r.Context(), authCtxKey, Auth{TenantID: tenant, Role: "viewer"}))
+}
+
+func TestTelemetryGETsAndRefreshNeverPollAgents(t *testing.T) {
+	var calls atomic.Int64
+	agent := httptest.NewTLSServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		calls.Add(1)
+		_, _ = w.Write([]byte(`{"ok":true,"sessions":[]}`))
+	}))
+	defer agent.Close()
+	t.Setenv("EVOBGP_NODE_DISPATCH_INSECURE_TLS", "1")
+	mem := store.NewMemory()
+	mem.SeedDemo()
+	tenant, _, _, _, _ := mem.DemoIDs()
+	sp, err := mem.CreateSpeaker(tenant, &store.Speaker{Role: "replica", MetaJSON: store.SpeakerMetaJSON(store.SpeakerMeta{AgentDomain: strings.TrimPrefix(agent.URL, "https://"), AgentSecret: "test"})})
+	if err != nil {
+		t.Fatal(err)
+	}
+	// Positive control: the legacy adapter reaches the fake configured agent.
+	if result := nodedispatch.FetchAgentHealth(context.Background(), sp, nodedispatch.Options{HTTPClient: agent.Client()}); !result.OK {
+		t.Fatal(result)
+	}
+	calls.Store(0)
+	st := &telemetryHTTPStore{Backend: mem}
+	server := &Server{store: st}
+	claims, err := mem.ClaimTelemetry(context.Background(), "collector", 1, time.Minute, "")
+	if err != nil || len(claims) != 1 {
+		t.Fatalf("claim: %v %v", claims, err)
+	}
+	for i := 0; i < 20; i++ {
+		for _, tc := range []struct {
+			path    string
+			handler http.HandlerFunc
+		}{
+			{"/v1/peers", server.handleListPeers}, {"/v1/peers?live=1", server.handleListPeers},
+			{"/v1/speakers?live=1", server.handleListSpeakers}, {"/v1/peers/discovered?status=all", server.handleListPeerDiscoveries},
+		} {
+			w := httptest.NewRecorder()
+			tc.handler(w, telemetryRequest(http.MethodGet, tc.path, tenant))
+			if w.Code != http.StatusOK {
+				t.Fatalf("%s: %d %s", tc.path, w.Code, w.Body.String())
+			}
+		}
+		w := httptest.NewRecorder()
+		server.handleTelemetryRefresh(w, telemetryRequest(http.MethodPost, "/v1/telemetry/refresh", tenant))
+		if w.Code != http.StatusAccepted {
+			t.Fatalf("refresh: %d %s", w.Code, w.Body.String())
+		}
+		want := `"requested":false`
+		if i == 0 {
+			want = `"requested":true`
+		}
+		if !strings.Contains(w.Body.String(), want) {
+			t.Fatalf("refresh coalescing: %s", w.Body.String())
+		}
+	}
+	if calls.Load() != 0 || st.discoveryWrites != 0 {
+		t.Fatalf("GET/refresh caused IO: agent=%d discovery=%d", calls.Load(), st.discoveryWrites)
+	}
+}
+
+func TestTelemetryReadFailureAndRefreshPermission(t *testing.T) {
+	mem := store.NewMemory()
+	mem.SeedDemo()
+	tenant, _, _, _, _ := mem.DemoIDs()
+	st := &telemetryHTTPStore{Backend: mem, snapshotErr: errors.New("private database failure")}
+	server := &Server{store: st}
+	for _, handler := range []http.HandlerFunc{server.handleListPeers, server.handleListSpeakers} {
+		w := httptest.NewRecorder()
+		handler(w, telemetryRequest(http.MethodGet, "/v1/resource?live=1", tenant))
+		if w.Code != http.StatusServiceUnavailable || strings.Contains(w.Body.String(), "private") {
+			t.Fatalf("snapshot failure: %d %s", w.Code, w.Body.String())
+		}
+	}
+	r := httptest.NewRequest(http.MethodPost, "/v1/telemetry/refresh", nil)
+	w := httptest.NewRecorder()
+	server.handleTelemetryRefresh(w, r)
+	if w.Code != http.StatusUnauthorized || st.refreshCalls != 0 {
+		t.Fatal("unauthenticated refresh admitted")
+	}
+	r = r.WithContext(context.WithValue(r.Context(), authCtxKey, Auth{TenantID: tenant, Kind: AuthKindJWT}))
+	w = httptest.NewRecorder()
+	server.handleTelemetryRefresh(w, r)
+	if w.Code != http.StatusForbidden || st.refreshCalls != 0 {
+		t.Fatal("unprivileged refresh admitted")
+	}
+}