test: relax the checks on range search (#36542) (#38234)

/kind improvement

pr:  #36542

tests/restful_client_v2/testcases/test_vector_operations.py

@@ -1580,10 +1580,11 @@ class TestSearchVector(TestBase):
         vector_to_search = preprocessing.normalize([np.array([random.random() for i in range(dim)])])[0].tolist()
         training_data = [item[vector_field] for item in data]
         distance_sorted = get_sorted_distance(training_data, [vector_to_search], metric_type)
-        r1, r2 = distance_sorted[0][nb//2], distance_sorted[0][nb//2+limit+int((0.2*limit))] # recall is not 100% so add 20% to make sure the range is correct
+        r1, r2 = distance_sorted[0][nb//2], distance_sorted[0][nb//2+limit+int((0.5*limit))] # recall is not 100% so add 50% to make sure the range is more than limit
         if metric_type == "L2":
             r1, r2 = r2, r1
         output_fields = get_common_fields_by_data(data, exclude_fields=[vector_field])
+        logger.info(f"r1: {r1}, r2: {r2}")
         payload = {
             "collectionName": name,
             "data": [vector_to_search],
@@ -1601,7 +1602,14 @@ class TestSearchVector(TestBase):
         assert rsp['code'] == 0
         res = rsp['data']
         logger.info(f"res: {len(res)}")
-        assert len(res) == limit
+        assert len(res) >= limit*0.8
+        # add buffer to the distance of comparison
+        if metric_type == "L2":
+            r1 = r1 + 10**-6
+            r2 = r2 - 10**-6
+        else:
+            r1 = r1 - 10**-6
+            r2 = r2 + 10**-6
         for item in res:
             distance = item.get("distance")
             if metric_type == "L2":

tests/restful_client_v2/utils/utils.py

@@ -262,6 +262,6 @@ def get_sorted_distance(train_emb, test_emb, metric_type):
         "IP": ip_distance
     }
     distance = pairwise_distances(train_emb, Y=test_emb, metric=milvus_sklearn_metric_map[metric_type], n_jobs=-1)
-    distance = np.array(distance.T, order='C', dtype=np.float16)
+    distance = np.array(distance.T, order='C', dtype=np.float32)
     distance_sorted = np.sort(distance, axis=1).tolist()
     return distance_sorted