fix(services): не балансировать сервис с одним IP

KPI и карточка показывают живой OK, а не гистерезис unknown; DNS по-прежнему ждёт повторные успехи.

Co-authored-by: Cursor <[email protected]>

packages/db/src/repos.ts

@@ -2341,6 +2341,113 @@ export function listIpHealthByServiceIds(
   return result;
 }
 
+function parseIpHealthState(status: string | null): IpHealthState | undefined {
+  if (
+    status === "up" ||
+    status === "down" ||
+    status === "degraded" ||
+    status === "unknown"
+  ) {
+    return status;
+  }
+  return undefined;
+}
+
+function bestAliveIpState(statuses: readonly IpHealthState[]): IpHealthState {
+  if (statuses.some((status) => status === "up")) return "up";
+  if (statuses.some((status) => status === "degraded")) return "degraded";
+  if (statuses.some((status) => status === "down")) return "down";
+  return "unknown";
+}
+
+/**
+ * Latest probe per service+IP+provider from health_probe_log, then any-up per IP.
+ * Display overlay — hysteresis in ip_health_status is unchanged (DNS).
+ */
+export function listLatestLiveHealthByServiceIds(
+  db: Db,
+  serviceIds: number[],
+): Map<number, ServiceIpHealthRow[]> {
+  const result = new Map<number, ServiceIpHealthRow[]>();
+  if (serviceIds.length === 0) return result;
+  const idList = sql.join(
+    serviceIds.map((id) => sql`${id}`),
+    sql`, `,
+  );
+  const rows = db.all<{
+    service_id: number;
+    ip: string;
+    provider: string | null;
+    status: string | null;
+    latency_ms: number | null;
+    last_error: string | null;
+    colo: string | null;
+    checked_at: string | null;
+  }>(sql`
+    SELECT sb.service_id AS service_id,
+      l.ip AS ip,
+      l.provider AS provider,
+      l.status AS status,
+      l.latency_ms AS latency_ms,
+      l.error AS last_error,
+      l.colo AS colo,
+      l.checked_at AS checked_at
+    FROM health_probe_log l
+    INNER JOIN service_bindings sb
+      ON l.scope = 'binding' AND l.ref_id = sb.id
+    INNER JOIN (
+      SELECT sb2.service_id AS service_id,
+        l2.ip AS ip,
+        l2.provider AS provider,
+        MAX(l2.id) AS max_id
+      FROM health_probe_log l2
+      INNER JOIN service_bindings sb2
+        ON l2.scope = 'binding' AND l2.ref_id = sb2.id
+      WHERE sb2.service_id IN (${idList})
+      GROUP BY sb2.service_id, l2.ip, l2.provider
+    ) latest
+      ON latest.max_id = l.id
+  `);
+
+  const byServiceIp = new Map<
+    string,
+    { serviceId: number; ip: string; probes: ServiceIpHealthRow[] }
+  >();
+  for (const row of rows) {
+    const status = parseIpHealthState(row.status);
+    if (!status) continue;
+    const key = `${row.service_id}\0${row.ip}`;
+    const probe: ServiceIpHealthRow = {
+      ip: row.ip,
+      status,
+      latency_ms: row.latency_ms,
+      last_checked_at: row.checked_at,
+      last_error: row.last_error,
+      provider: normalizeStatusProvider(row.provider),
+      colo: row.colo,
+    };
+    const bucket = byServiceIp.get(key);
+    if (bucket) bucket.probes.push(probe);
+    else {
+      byServiceIp.set(key, {
+        serviceId: row.service_id,
+        ip: row.ip,
+        probes: [probe],
+      });
+    }
+  }
+
+  for (const { serviceId, ip, probes } of byServiceIp.values()) {
+    const status = bestAliveIpState(probes.map((probe) => probe.status));
+    const preferred =
+      probes.find((probe) => probe.status === status) ?? probes[0]!;
+    const list = result.get(serviceId) ?? [];
+    list.push({ ...preferred, ip, status });
+    result.set(serviceId, list);
+  }
+  return result;
+}
+
 export function mergeHealthAggregates(
   parts: Array<HealthAggregate | undefined | null>,
 ): HealthAggregate {