[QUESTION]: When I use multiple threads to request Milvus' data, why do I have good QPS growth for 1,2,3 threads, but no QPS growth for more than 3 threads?

[HungryFour]

Is there an existing issue for this?

• [X] I have searched the existing issues

What is your question?

When I use multiple threads to request Milvus' data, why do I have good QPS growth for 1,2,3 threads, but no QPS growth for more than 3 threads? Standalone version using Docker deployment. Start with standalone_embed.sh without the user.yaml file.

    def multi_query(self, X: numpy.array, n: int, num_threads: int):

        if not self.collection:
            self.collection = Collection(self.collection_name, consistence_level="STRONG")
            self.load_collection()

        total_count = len(X)
        per_count = total_count / 10
        def single_query(idx, v, n):
            try:
                # conn = conn_pool.get_connection()
                start = time.time()
                results = self.collection.search(
                    data=[v.tolist()],
                    anns_field="vector",
                    param=self.search_params,
                    limit=n,
                    output_fields=["id", "vector"])
                total = time.time() - start
                res = [(v, r.entity.get("id"), r.entity.get("vector")) for r in results[0]]
                if (idx + 1) % per_count == 0:
                    print(
                        f"Read Processed: {round((idx + 1) / total_count * 100, 1)}%，Complete：{time.strftime('%Y-%m-%d %H:%M:%S', time.localtime())}")
                return (start, total, res)
            except Exception as e:
                print(f"Error executing query: {e}")
                return None
            # finally:
            #     if conn is not None:
            #         conn_pool.release_connection(conn)

        with ThreadPoolExecutor(max_workers=num_threads) as pool:
            results = pool.map(lambda idx_v: single_query(idx_v[0], idx_v[1], n), enumerate(X))
        self.res = [result for result in results if result is not None]

Anything else?

No response