Fix pr #6095: Chore: clean up LLM (prompt caching, supports fn callin…

…g), leftover renames

openhands/llm/exceptions.py

@@ -0,0 +1,14 @@
+from litellm.exceptions import APIError
+
+
+class CloudflareBlockError(APIError):
+    """Exception raised when Cloudflare blocks the request."""
+
+    def __init__(self, *args, **kwargs):
+        super().__init__(*args, **kwargs)
+
+    def __instancecheck__(self, instance):
+        return (
+            super().__instancecheck__(instance)
+            and 'Attention Required! | Cloudflare' in str(instance)
+        )

openhands/llm/llm.py

@@ -30,6 +30,7 @@
 from openhands.core.logger import openhands_logger as logger
 from openhands.core.message import Message
 from openhands.llm.debug_mixin import DebugMixin
+from openhands.llm.exceptions import CloudflareBlockError
 from openhands.llm.fn_call_converter import (
     STOP_WORDS,
     convert_fncall_messages_to_non_fncall_messages,
@@ -147,6 +148,7 @@ def __init__(
         @self.retry_decorator(
             num_retries=self.config.num_retries,
             retry_exceptions=LLM_RETRY_EXCEPTIONS,
+            exclude_exceptions=(CloudflareBlockError,),
             retry_min_wait=self.config.retry_min_wait,
             retry_max_wait=self.config.retry_max_wait,
             retry_multiplier=self.config.retry_multiplier,

openhands/llm/retry_mixin.py

@@ -1,4 +1,3 @@
-from litellm.exceptions import APIError
 from tenacity import (
     retry,
     retry_if_exception,
@@ -19,23 +18,24 @@ def retry_decorator(self, **kwargs):
 
         Args:
             **kwargs: Keyword arguments to override default retry behavior.
-                      Keys: num_retries, retry_exceptions, retry_min_wait, retry_max_wait, retry_multiplier
+                      Keys: num_retries, retry_exceptions, exclude_exceptions, retry_min_wait, retry_max_wait, retry_multiplier
 
         Returns:
             A retry decorator with the parameters customizable in configuration.
         """
         num_retries = kwargs.get('num_retries')
         retry_exceptions: tuple = kwargs.get('retry_exceptions', ())
+        exclude_exceptions: tuple = kwargs.get('exclude_exceptions', ())
         retry_min_wait = kwargs.get('retry_min_wait')
         retry_max_wait = kwargs.get('retry_max_wait')
         retry_multiplier = kwargs.get('retry_multiplier')
 
         def _filter_exceptions(e):
-            # For Cloudflare blocks, don't retry - just return False
-            if isinstance(e, APIError) and 'Attention Required! | Cloudflare' in str(e):
+            # First check if the exception is in the exclude list
+            if isinstance(e, exclude_exceptions):
                 return False
 
-            # Otherwise, return True if we want to retry, which means e is in retry_exceptions
+            # Then check if it's in the retry list
             return isinstance(e, retry_exceptions)
 
         return retry(

pyproject.toml

@@ -101,6 +101,7 @@ reportlab = "*"
 [tool.coverage.run]
 concurrency = ["gevent"]
 
+
 [tool.poetry.group.runtime.dependencies]
 jupyterlab = "*"
 notebook = "*"
@@ -129,6 +130,7 @@ ignore = ["D1"]
 [tool.ruff.lint.pydocstyle]
 convention = "google"
 
+
 [tool.poetry.group.evaluation.dependencies]
 streamlit = "*"
 whatthepatch = "*"