Azure OpenAI+APIM负载均衡配置后，Python代码如何适配？

已完成Azure API Management（APIM）与Azure OpenAI的负载均衡策略配置，搭建了可生成文本摘要的基础Web应用。此前代码直接调用Azure OpenAI端点并使用密钥，现需修改Python代码以适配APIM负载均衡端点。

附APIM策略代码：

<!--
    IMPORTANT:
    - Policy elements can appear only within the <inbound>, <outbound>, <backend> section elements.
    - To apply a policy to the incoming request (before it is forwarded to the backend service), place a corresponding policy element within the <inbound> section element.
    - To apply a policy to the outgoing response (before it is sent back to the caller), place a corresponding policy element within the <outbound> section element.
    - To add a policy, place the cursor at the desired insertion point and select a policy from the sidebar.
    - To remove a policy, delete the corresponding policy statement from the policy document.
    - Position the <base> element within a section element to inherit all policies from the corresponding section element in the enclosing scope.
    - Remove the <base> element to prevent inheriting policies from the corresponding section element in the enclosing scope.
    - Policies are applied in the order of their appearance, from the top down.
    - Comments within policy elements are not supported and may disappear. Place your comments between policy elements or at a higher level scope.
-->
<policies>
    <inbound>
        <base />
        <check-header name="X-Azure-FDID" failed-check-httpcode="401" failed-check-error-message="Not authorized" ignore-case="false">
            <value>aaabbbbccc</value>
        </check-header>
        <ip-filter action="allow">
            <address-range from="add1" to="add2" />
        </ip-filter>
        <set-variable name="backendUrlA" value="url1" />
        <!-- [A] Subscription 1: location East -->
        <set-variable name="backendUrlB" value="url2" />
        <!-- [B] Subscription 1: location East -->
        <set-variable name="backendA-apiKey" value="{{backendA-apiKey}}" />
        <set-variable name="backendB-apiKey" value="{{backendB-apiKey}}" />
        <!-- Load balancing logic -->
        <choose>
            <when condition="@((int)context.Request.Url.Path.IndexOf("/gpt-35-turbo") != -1)">
                <!-- Pool 1: GPT3.5 Turbo and GPT3.5 Turbo 16K -->
                <!-- Load balancing logic for gpt-35-turbo models -->
                <cache-lookup-value key="pool1Index" default-value="@((int)0)" variable-name="pool1Index" />
                <choose>
                    <when condition="@( (int)context.Variables["pool1Index"] == 0 )">
                        <set-variable name="selectedBackend" value="@((string)context.Variables["backendUrlA"])" />
                        <set-variable name="selectedBackendKey" value="@((string)context.Variables["backendA-apiKey"])" />
                    </when>
                    <otherwise>
                        <set-variable name="selectedBackend" value="@((string)context.Variables["backendUrlB"])" />
                        <set-variable name="selectedBackendKey" value="@((string)context.Variables["backendB-apiKey"])" />
                    </otherwise>
                </choose>
                <!-- Increment the pool1Index and reset to 0 when it reaches the end -->
                <set-variable name="pool1Index" value="@((int)context.Variables["pool1Index"] == 1 ? 0 : (int)context.Variables["pool1Index"] + 1)" />
                <cache-store-value key="pool1Index" value="@((int)context.Variables["pool1Index"])" duration="1440" />
            </when>
            <when condition="@((int)context.Request.Url.Path.IndexOf("/text-embedding-ada-002") != -1)">
                <!-- Pool 2: Embedding ADA-002 -->
                <!-- Load balancing logic for text-embedding-ada-002 model -->
                <cache-lookup-value key="pool2Index" default-value="@((int)0)" variable-name="pool2Index" />
                <choose>
                    <when condition="@( (int)context.Variables["pool2Index"] == 0 )">
                        <set-variable name="selectedBackend" value="@((string)context.Variables["backendUrlA"])" />
                        <set-variable name="selectedBackendKey" value="@((string)context.Variables["backendA-apiKey"])" />
                    </when>
                    <otherwise>
                        <set-variable name="selectedBackend" value="@((string)context.Variables["backendUrlB"])" />
                        <set-variable name="selectedBackendKey" value="@((string)context.Variables["backendB-apiKey"])" />
                    </otherwise>
                </choose>
                <!-- Increment the pool2Index and reset to 0 when it reaches the end -->
                <set-variable name="pool2Index" value="@((int)context.Variables["pool2Index"] == 1 ? 0 : (int)context.Variables["pool2Index"] + 1)" />
                <cache-store-value key="pool2Index" value="@((int)context.Variables["pool2Index"])" duration="1440" />
            </when>
            <otherwise>
                <!-- Direct to Pool 3: General -->
                <set-variable name="selectedBackend" value="@((string)context.Variables["backendUrlA"])" />
                <set-variable name="selectedBackendKey" value="@((string)context.Variables["backendA-apiKey"])" />
            </otherwise>
        </choose>
        <set-backend-service base-url="@((string)context.Variables["selectedBackend"])" />
        <set-header name="api-key" exists-action="override">
            <value>@((string)context.Variables["selectedBackendKey"])</value>
        </set-header>
    </inbound>
    <backend>
        <base />
    </backend>
    <outbound>
        <base />
        <!-- Add the selected backend URL to the response headers -->
        <set-header name="X-Selected-Backend" exists-action="override">
            <value>@((string)context.Variables["selectedBackend"])</value>
        </set-header>
    </outbound>
    <on-error>
        <base />
    </on-error>
</policies>

核心修改点 #

根据你的APIM策略，代码需要做以下关键调整：

• 替换请求端点为APIM的服务端点（而非直接调用Azure OpenAI端点）
• 添加APIM强制验证的X-Azure-FDID请求头（值为策略中指定的aaabbbbccc）
• 移除原有的Azure OpenAI密钥配置（APIM会自动注入后端实例的密钥）
• 保持原有请求体结构不变（APIM会转发请求到对应后端）

代码示例1：使用OpenAI Python SDK #

from openai import AzureOpenAI

# 初始化客户端时指定APIM端点，无需设置api_key
client = AzureOpenAI(
    azure_endpoint="https://你的APIM实例名称.azure-api.net/openai",  # 替换为你的APIM端点
    api_version="2024-02-15-preview",  # 保持与原代码一致的API版本
    # 注意：不需要设置api_key，APIM会自动处理后端密钥
)

# 构造请求时添加X-Azure-FDID头
response = client.chat.completions.create(
    model="gpt-35-turbo",
    messages=[
        {"role": "system", "content": "请生成文本摘要"},
        {"role": "user", "content": "你的输入文本内容"}
    ],
    extra_headers={
        "X-Azure-FDID": "aaabbbbccc"  # 必须与APIM策略中的值一致
    }
)

# 可选：查看响应头中的X-Selected-Backend，验证负载均衡是否生效
print("当前选中的后端实例:", response.headers.get("X-Selected-Backend"))
print("摘要结果:", response.choices[0].message.content)

代码示例2：使用requests库直接调用 #

import requests

APIM_ENDPOINT = "https://你的APIM实例名称.azure-api.net/openai/deployments/gpt-35-turbo/chat/completions"  # 替换为你的APIM完整端点
API_VERSION = "2024-02-15-preview"

headers = {
    "Content-Type": "application/json",
    "X-Azure-FDID": "aaabbbbccc",  # 必须添加的验证头
    # 注意：不需要添加api-key头，APIM会自动注入
}

payload = {
    "messages": [
        {"role": "system", "content": "请生成文本摘要"},
        {"role": "user", "content": "你的输入文本内容"}
    ],
    "temperature": 0.7
}

response = requests.post(
    f"{APIM_ENDPOINT}?api-version={API_VERSION}",
    headers=headers,
    json=payload
)

# 检查响应状态
if response.status_code == 200:
    result = response.json()
    print("当前选中的后端实例:", response.headers.get("X-Selected-Backend"))
    print("摘要结果:", result["choices"][0]["message"]["content"])
else:
    print(f"请求失败: {response.status_code} - {response.text}")

关键说明 #

1. IP白名单验证：确保你的Web应用服务器IP在APIM策略的<ip-filter>允许范围内，否则会被拒绝访问。
2. 负载均衡验证：通过响应头X-Selected-Backend可以查看当前请求被路由到哪个Azure OpenAI实例，确认负载均衡逻辑是否正常工作。
3. 模型路径匹配：APIM策略通过请求路径中的模型名称（如/gpt-35-turbo）分配后端池，确保请求URL中的模型名称与策略中的匹配。

内容的提问来源于stack exchange，提问作者FarzadAvari