From 9bf08d13f70a50e0153e48343c9cff5d432b1df3 Mon Sep 17 00:00:00 2001
From: alex-omophub <sdk@omophub.com>
Date: Mon, 30 Mar 2026 19:12:03 +0100
Subject: [PATCH] Add retry logic with exponential backoff and jitter for rate
 limits and server errors

- Introduced a new `_calculate_retry_delay` function to handle retry delays based on the Retry-After header and exponential backoff with jitter.
- Updated `SyncHTTPClient` and `AsyncHTTPClientImpl` to utilize the new retry delay calculation for handling rate limits (429) and server errors (502, 503, 504).
- Enhanced retry mechanism to improve resilience against temporary issues.
---
 src/omophub/_http.py | 62 ++++++++++++++++++++++++++++++++++++++------
 1 file changed, 54 insertions(+), 8 deletions(-)

diff --git a/src/omophub/_http.py b/src/omophub/_http.py
index 04cee9e..31384c2 100644
--- a/src/omophub/_http.py
+++ b/src/omophub/_http.py
@@ -2,6 +2,7 @@
 
 from __future__ import annotations
 
+import random
 import time
 from abc import ABC, abstractmethod
 from typing import TYPE_CHECKING, Any
@@ -12,6 +13,12 @@
 from ._exceptions import ConnectionError, TimeoutError
 from ._version import get_version
 
+# Retry constants (OpenAI-style exponential backoff with jitter)
+INITIAL_RETRY_DELAY = 0.5  # seconds
+MAX_RETRY_DELAY = 8.0  # seconds
+MAX_RETRY_AFTER = 60  # max seconds to respect from Retry-After header
+RETRYABLE_STATUS_CODES = (429, 502, 503, 504)
+
 if TYPE_CHECKING:
     from collections.abc import Mapping
 
@@ -24,6 +31,37 @@
     HTTP2_AVAILABLE = False
 
 
+def _calculate_retry_delay(
+    attempt: int,
+    max_retries: int,
+    response_headers: Mapping[str, str] | None = None,
+) -> float:
+    """Calculate retry delay with Retry-After support and exponential backoff + jitter.
+
+    Follows the OpenAI pattern:
+    1. If Retry-After header present and <= 60s, use it
+    2. Otherwise, exponential backoff (0.5s * 2^attempt) with 25% jitter, capped at 8s
+    """
+    # Check Retry-After header first
+    if response_headers:
+        retry_after = response_headers.get("retry-after") or response_headers.get(
+            "Retry-After"
+        )
+        if retry_after:
+            try:
+                retry_after_seconds = float(retry_after)
+                if 0 < retry_after_seconds <= MAX_RETRY_AFTER:
+                    return retry_after_seconds
+            except ValueError:
+                pass
+
+    # Exponential backoff with jitter
+    retries_done = min(max_retries - (max_retries - attempt), 1000)
+    sleep_seconds = min(INITIAL_RETRY_DELAY * (2.0**retries_done), MAX_RETRY_DELAY)
+    jitter = 1 - 0.25 * random.random()
+    return sleep_seconds * jitter
+
+
 class HTTPClient(ABC):
     """Abstract base class for HTTP clients."""
 
@@ -137,12 +175,15 @@ def request(
                     params=filtered_params if filtered_params else None,
                     json=json,
                 )
-                # Retry on server errors (502, 503, 504)
+                # Retry on rate limits (429) and server errors (502, 503, 504)
                 if (
-                    response.status_code in (502, 503, 504)
+                    response.status_code in RETRYABLE_STATUS_CODES
                     and attempt < self._max_retries
                 ):
-                    time.sleep(2**attempt * 0.5)
+                    delay = _calculate_retry_delay(
+                        attempt, self._max_retries, response.headers
+                    )
+                    time.sleep(delay)
                     continue
                 return response.content, response.status_code, response.headers
 
@@ -155,7 +196,8 @@ def request(
 
             # Exponential backoff before retry
             if attempt < self._max_retries:
-                time.sleep(2**attempt * 0.1)
+                delay = _calculate_retry_delay(attempt, self._max_retries)
+                time.sleep(delay)
 
         raise last_exception or ConnectionError("Request failed after retries")
 
@@ -229,12 +271,15 @@ async def request(
                     params=filtered_params if filtered_params else None,
                     json=json,
                 )
-                # Retry on server errors (502, 503, 504)
+                # Retry on rate limits (429) and server errors (502, 503, 504)
                 if (
-                    response.status_code in (502, 503, 504)
+                    response.status_code in RETRYABLE_STATUS_CODES
                     and attempt < self._max_retries
                 ):
-                    await asyncio.sleep(2**attempt * 0.5)
+                    delay = _calculate_retry_delay(
+                        attempt, self._max_retries, response.headers
+                    )
+                    await asyncio.sleep(delay)
                     continue
                 return response.content, response.status_code, response.headers
 
@@ -247,7 +292,8 @@ async def request(
 
             # Exponential backoff before retry
             if attempt < self._max_retries:
-                await asyncio.sleep(2**attempt * 0.1)
+                delay = _calculate_retry_delay(attempt, self._max_retries)
+                await asyncio.sleep(delay)
 
         raise last_exception or ConnectionError("Request failed after retries")