feat: add max_retries to OpenAI and Azure encoders

[ashraq1455]

User description

This update introduces a max_retries option for the OpenAIEncoder and AzureOpenAIEncoder classes. This option allows users to specify the maximum number of retry attempts for API calls in case of errors. By default, it is set to 3 retries. If no retries are needed, you can set max_retries to 0.

Example:

from semantic_router.encoders import OpenAIEncoder

encoder = OpenAIEncoder(name="text-embedding-3-large", max_retries=2)

---

PR Type

enhancement

---

Description

• Introduced a max_retries option for both OpenAIEncoder and AzureOpenAIEncoder classes, allowing users to specify the maximum number of retry attempts for API calls.
• Implemented exponential backoff retry logic for handling API call failures in both synchronous and asynchronous methods.
• Enhanced error logging to provide more detailed information on exceptions.
• Removed redundant error message handling to streamline code.

---

Changes walkthrough 📝

Relevant files

Enhancement

openai.py Add retry logic and error handling to OpenAIEncoder            semantic_router/encoders/openai.py Added max_retries attribute to OpenAIEncoder class with a default value of 3. Implemented retry logic using exponential backoff for API calls. Improved error logging and handling for API call failures. Removed redundant error message variable. +26/-15  zure.py Add retry logic and error handling to AzureOpenAIEncoder  semantic_router/encoders/zure.py Added max_retries attribute to AzureOpenAIEncoder class with a default value of 3. Implemented retry logic using exponential backoff for API calls. Improved error logging and handling for API call failures. Removed redundant error message variable. +25/-22

---

💡 PR-Agent usage:
Comment /help on the PR to get a list of all available PR-Agent tools and their descriptions

[github-actions]

PR Reviewer Guide 🔍

⏱️ Estimated effort to review: 3 🔵🔵🔵⚪⚪

🧪 No relevant tests

🔒 No security concerns identified

⚡ Key issues to review Exception Handling The exception handling in the retry logic raises a generic ValueError with the message "OpenAI API call failed." This could be improved by providing more specific error types or more detailed error messages to help with debugging and error resolution. Hardcoded Test Exception There is a hardcoded raise OpenAIError("Test") within the retry logic. This seems to be for testing purposes and should be removed or commented out for production code. Exception Handling Similar to openai.py, the exception handling could be more specific than just raising a ValueError with a generic message. More specific exceptions or messages could aid in better error handling and debugging. Hardcoded Test Exception The retry logic includes a raise OpenAIError("Test") which appears to be for testing. This should be removed or properly handled before moving to production.

[github-actions]

PR Code Suggestions ✨

Category	Suggestion	Score
Possible bug	Remove or replace the test exception raise inside the retry loop Replace the hardcoded exception raise inside the retry loop with a proper error handling mechanism or remove it if it was for testing purposes. semantic_router/encoders/openai.py [117] -raise OpenAIError("Test") +# Removed test exception raise Suggestion importance[1-10]: 9 Why: The hardcoded exception raise is likely for testing and should be removed or replaced with proper error handling to avoid unintended disruptions in production code.	9
Enhancement	Skip retry loop when max_retries is zero Implement a mechanism to handle the case when max_retries is zero, to skip the retry loop entirely, optimizing performance when no retries are intended. semantic_router/encoders/openai.py [115] -for j in range(self.max_retries + 1): +if self.max_retries > 0: + for j in range(self.max_retries + 1): Suggestion importance[1-10]: 8 Why: This suggestion improves performance by avoiding unnecessary iterations when no retries are intended, making the code more efficient and logical.	8
	Cap the maximum sleep time during retries to avoid excessive delays Ensure that exponential backoff sleep time increases with each retry attempt by adjusting the sleep time calculation. semantic_router/encoders/openai.py [128] -sleep(2**j) +sleep(min(2**j, 120)) # Caps the sleep time to a maximum of 120 seconds Suggestion importance[1-10]: 8 Why: Capping the sleep time during retries prevents excessive delays, which is beneficial for maintaining reasonable response times and system performance.	8
Possible issue	Validate max_retries to be a non-negative integer The max_retries attribute should be validated to ensure it is a non-negative integer. This prevents potential runtime errors or logical bugs due to negative retry attempts. semantic_router/encoders/openai.py [45] -max_retries: int = 3 +max_retries: int = max(0, int(value)) # Ensures non-negative integer Suggestion importance[1-10]: 7 Why: Ensuring max_retries is a non-negative integer is a good practice to prevent logical errors and runtime exceptions. However, the suggested code does not show how value is defined or used, which could lead to confusion.	7