fix: enforce S3 filename stem byte limit

litellm/integrations/s3.py

@@ -21,6 +21,24 @@ def _long_filename_suffix(s3_file_name: str) -> str:
     )
 
 
+def _truncate_to_utf8_bytes(value: str, max_bytes: int) -> str:
+    if max_bytes <= 0:
+        return ""
+
+    encoded_value = value.encode("utf-8")
+    if len(encoded_value) <= max_bytes:
+        return value
+
+    truncated_value = encoded_value[:max_bytes]
+    while truncated_value:
+        try:
+            return truncated_value.decode("utf-8")
+        except UnicodeDecodeError:
+            truncated_value = truncated_value[:-1]
+
+    return ""
+
+
 class S3Logger:
     # Class variables or attributes
     def __init__(
@@ -197,10 +215,14 @@ def get_s3_object_key(
     start_time: datetime,
     s3_file_name: str,
 ) -> str:
-    if len(s3_file_name) > MAX_S3_FILENAME_STEM_LENGTH:
+    if len(s3_file_name.encode("utf-8")) > MAX_S3_FILENAME_STEM_LENGTH:
         suffix = _long_filename_suffix(s3_file_name)
-        prefix_length = MAX_S3_FILENAME_STEM_LENGTH - len(suffix) - 1
-        s3_file_name = f"{s3_file_name[:prefix_length]}_{suffix}"
+        prefix_byte_length = max(
+            MAX_S3_FILENAME_STEM_LENGTH - len(suffix.encode("utf-8")) - 1,
+            0,
+        )
+        truncated_prefix = _truncate_to_utf8_bytes(s3_file_name, prefix_byte_length)
+        s3_file_name = f"{truncated_prefix}_{suffix}"
 
     s3_object_key = (
         (s3_path.rstrip("/") + "/" if s3_path else "")

tests/test_litellm/integrations/test_s3_v2.py

@@ -898,8 +898,8 @@ def test_get_s3_object_key_truncates_long_filename_components():
     )
     filename = key.split("/")[-1]
 
-    assert len(f"{original_file_name}.json") > 255
-    assert len(filename) <= 255
+    assert len(f"{original_file_name}.json".encode("utf-8")) > 255
+    assert len(filename.encode("utf-8")) <= 255
     assert filename.endswith(".json")
     assert filename.startswith(f"time-{start_time.strftime('%H-%M-%S-%f')}_resp_")
 
@@ -918,8 +918,23 @@ def test_get_s3_object_key_uses_deterministic_suffixes_for_long_filename_compone
 
     assert first_key == first_key_again
     assert first_key != second_key
-    assert len(first_key.split("/")[-1]) <= 255
-    assert len(second_key.split("/")[-1]) <= 255
+    assert len(first_key.split("/")[-1].encode("utf-8")) <= 255
+    assert len(second_key.split("/")[-1].encode("utf-8")) <= 255
+
+
+def test_get_s3_object_key_truncates_non_ascii_filename_components_by_utf8_bytes():
+    start_time = datetime(2026, 3, 26, 12, 51, 27, 995047)
+    long_response_id = "resp_" + ("漢" * 120)
+    original_file_name = f"time-{start_time.strftime('%H-%M-%S-%f')}_{long_response_id}"
+
+    key = get_s3_object_key("", "", start_time, original_file_name)
+    filename = key.split("/")[-1]
+
+    assert len(original_file_name) < 250
+    assert len(f"{original_file_name}.json".encode("utf-8")) > 255
+    assert len(filename.encode("utf-8")) <= 255
+    assert filename.endswith(".json")
+    assert filename.startswith(f"time-{start_time.strftime('%H-%M-%S-%f')}_resp_")
 
 
 def test_create_s3_batch_logging_element_limits_long_ids():
@@ -939,5 +954,5 @@ def test_create_s3_batch_logging_element_limits_long_ids():
         result = logger.create_s3_batch_logging_element(start_time, payload)
 
     assert result is not None
-    assert len(result.s3_object_key.split("/")[-1]) <= 255
+    assert len(result.s3_object_key.split("/")[-1].encode("utf-8")) <= 255
     assert result.s3_object_key.endswith(".json")