Environment:Tensorflow Serving Python Client Environment

Knowledge Sources	TensorFlow Serving Python API Setup
Domains	Infrastructure, ML_Serving
Last Updated	2026-02-13 17:00 GMT

Overview

Python 3.9+ client environment with TensorFlow, gRPC >= 1.24.3, and Protobuf >= 4.21.6 for interacting with TensorFlow Serving APIs.

Description

This environment provides the Python runtime and package dependencies required to use the TensorFlow Serving Python client API (`tensorflow-serving-api` pip package). It includes gRPC for making prediction requests, Protobuf for message serialization, and TensorFlow for tensor manipulation and model export. Example clients also require NumPy and Pillow (for image-based models like ResNet).

Usage

Use this environment when writing Python client code to interact with TensorFlow Serving, including sending prediction requests via gRPC or REST, exporting models with `SavedModelBuilder`, and generating model warmup data. This is the prerequisite for all Python-based example scripts in the repository.

System Requirements

Category	Requirement	Notes
Python	>= 3.9	Tested with 3.9, 3.10+
Network	Access to TF Serving endpoint	gRPC (port 8500) or REST (port 8501)

Dependencies

Python Packages

`tensorflow` >= 1.2.0, < 3
`grpcio` >= 1.24.3, < 2.0
`protobuf` >= 4.21.6
`tensorflow-serving-api` (this package bundles the above)
`numpy` (required by example clients)
`Pillow` (required by ResNet example client for image processing)
`requests` (required by REST API example client)

Credentials

No credentials required for local serving. For remote endpoints:

TLS/SSL: Optional `ssl_config_file` for secure gRPC channels
ALTS: Optional `--use_alts_credentials` for Google ALTS authentication

Quick Install

# Install the TensorFlow Serving Python API
pip install tensorflow-serving-api

# For example clients (additional dependencies)
pip install numpy Pillow requests

Code Evidence

Package dependencies from `setup.py:62-67`:

REQUIRED_PACKAGES = [
    # Match versions to what TF needs here:
    #   https://github.com/tensorflow/tensorflow/blob/master/tensorflow/tools/pip_package/setup.py.tpl
    'grpcio >= 1.24.3, < 2.0',
    'protobuf>=4.21.6',
] + _TF_REQ

Python version constraint from `setup.py:83`:

python_requires='>=3.9',

TensorFlow version range from `setup.py:36`:

_TF_VERSION = '>=1.2.0,<3'

gRPC client imports from `mnist_client.py:28-39`:

import grpc
import numpy
import tensorflow as tf
from tensorflow_serving.apis import predict_pb2
from tensorflow_serving.apis import prediction_service_pb2_grpc

ResNet client imports from `resnet_client_grpc.py:21-30`:

import grpc
import numpy as np
from PIL import Image
import requests
import tensorflow as tf
from tensorflow_serving.apis import predict_pb2
from tensorflow_serving.apis import prediction_service_pb2_grpc

Common Errors

Error Message	Cause	Solution
`ModuleNotFoundError: No module named 'tensorflow_serving'`	tensorflow-serving-api not installed	`pip install tensorflow-serving-api`
`grpc._channel._InactiveRpcError`	Server not reachable on specified port	Verify server is running and port matches (default gRPC: 8500)
`TypeError: Descriptors cannot not be created directly`	Protobuf version mismatch	Upgrade: `pip install protobuf>=4.21.6`

Compatibility Notes

GPU builds prior to TF 2.1: Must use `tensorflow-gpu` package instead of `tensorflow`. From TF 2.1+, the unified `tensorflow` package includes GPU support.
TF v1 vs v2 API: Example scripts use `tf.compat.v1` APIs. Some examples disable eager execution with `tf.compat.v1.disable_default_session()`.

Related Pages

Page Connections

Double-click a node to navigate. Hold to expand connections.

Principle

Implementation

Heuristic

Environment