diff --git a/lib/hashserv/tests.py b/lib/hashserv/tests.py
index 0fbc19c1d..41c1f249f 100644
--- a/lib/hashserv/tests.py
+++ b/lib/hashserv/tests.py
@@ -1561,6 +1561,102 @@ class TestHashEquivalenceTCPServer(HashEquivalenceTestSetup, HashEquivalenceComm
         # case it is more reliable to resolve the IP address explicitly.
         return socket.gethostbyname("localhost") + ":0"
 
+    def test_get_stream_upstream_pipelined(self):
+        # Verify that, with an upstream configured, the get-stream handler
+        # pipelines its upstream queries instead of doing one blocking
+        # round-trip per task. A latency proxy injects RTT between this server
+        # and its upstream; a serial (one round-trip per task) implementation
+        # could not beat N * RTT, so finishing far faster proves the queries
+        # are pipelined.
+        import asyncio
+
+        LATENCY = 0.01  # 10ms each direction => ~20ms round trip
+        upstream_host, upstream_port = self.server_address.rsplit(":", 1)
+        upstream_port = int(upstream_port)
+
+        ready = threading.Event()
+        proxy = {}
+
+        def run_proxy():
+            loop = asyncio.new_event_loop()
+            asyncio.set_event_loop(loop)
+
+            async def handle(creader, cwriter):
+                ureader, uwriter = await asyncio.open_connection(upstream_host, upstream_port)
+
+                async def pipe(r, w):
+                    try:
+                        while True:
+                            data = await r.read(65536)
+                            if not data:
+                                break
+                            await asyncio.sleep(LATENCY)
+                            w.write(data)
+                            await w.drain()
+                    except Exception:
+                        pass
+                    finally:
+                        try:
+                            w.close()
+                        except Exception:
+                            pass
+
+                await asyncio.gather(pipe(creader, uwriter), pipe(ureader, cwriter))
+
+            stop_event = asyncio.Event()
+            proxy["stop_event"] = stop_event
+            proxy["loop"] = loop
+
+            async def main():
+                server = await asyncio.start_server(handle, "127.0.0.1", 0)
+                proxy["port"] = server.sockets[0].getsockname()[1]
+                ready.set()
+                async with server:
+                    await stop_event.wait()
+
+            try:
+                loop.run_until_complete(main())
+            except Exception:
+                pass
+            finally:
+                loop.close()
+
+        proxy_thread = threading.Thread(target=run_proxy, daemon=True)
+        proxy_thread.start()
+        self.assertTrue(ready.wait(10), "latency proxy did not start")
+        self.addCleanup(proxy_thread.join, 10)
+        self.addCleanup(lambda: proxy["loop"].call_soon_threadsafe(proxy["stop_event"].set))
+        proxy_addr = "127.0.0.1:%d" % proxy["port"]
+
+        N = 400
+        expected = []
+        for i in range(N):
+            taskhash = hashlib.sha256(("task%d" % i).encode()).hexdigest()
+            outhash = hashlib.sha256(("out%d" % i).encode()).hexdigest()
+            unihash = hashlib.sha256(("uni%d" % i).encode()).hexdigest()
+            self.client.report_unihash(taskhash, self.METHOD, outhash, unihash)
+            expected.append((self.METHOD, taskhash, unihash))
+
+        # Downstream server with an EMPTY local DB whose upstream is the slow proxy.
+        down_server = self.start_server(upstream=proxy_addr)
+        down_client = self.start_client(down_server.address)
+
+        args = [(m, th) for (m, th, _uh) in expected]
+        start = time.time()
+        results = down_client.get_unihash_batch(args)
+        elapsed = time.time() - start
+
+        self.assertEqual(results, [uh for (_m, _th, uh) in expected])
+
+        # A serial (one round-trip per task) implementation cannot beat N*RTT.
+        # Allow a generous margin to avoid flakiness on loaded CI machines.
+        serial_lower_bound = N * 2 * LATENCY
+        self.assertLess(elapsed, serial_lower_bound / 4,
+                        "Upstream queries are not being pipelined "
+                        "(%.3fs for %d queries at %.0fms injected RTT)"
+                        % (elapsed, N, 2 * LATENCY * 1000))
+
+
 
 class TestHashEquivalenceWebsocketServer(HashEquivalenceTestSetup, HashEquivalenceCommonTests, unittest.TestCase):
     def setUp(self):