From cdb1245e7419fa8b0294da3946cbad68b3d567d5 Mon Sep 17 00:00:00 2001 From: yucheng-berri Date: Tue, 1 Sep 2026 13:30:02 -0700 Subject: [PATCH] fix(s3): bound s3 object keys and download filenames for long Responses API ids (#39164) * fix(s3): bound object keys and download filenames to s3 limits Long OpenAI-compatible Responses API ids pushed the s3 object key past s3's 1024 UTF-8 byte cap, so the PUT failed with a 400 and the log record was dropped. Keys that still fit are unchanged, byte for byte. An oversized one now keeps a readable head of the file name and appends the sha256 of the full name. A configured path/alias prefix that is long enough to overflow on its own keeps whole leading path segments, so a prefix-scoped IAM policy or lifecycle rule still matches, and ends in a short digest of the full configured value so two operators do not land in the same folder. The Content-Disposition filename carried the same unbounded id and hit s3's 2048 byte metadata-header cap, so the upload still failed with MetadataTooLarge once the key was bounded. It is bounded the same way, head plus digest, so two records downloaded from the console stay distinct files. The full response id stays in the uploaded JSON payload. * fix(s3): keep the configured prefix whole and spend the whole key budget Shorten the response id first and only trim the operator's configured prefix when the prefix itself is what does not fit, so prefix scoped IAM policies and lifecycle rules keep matching. Trim by bytes rather than whole segments so the longest possible string prefix survives, and route the audit log key through the same shared builder. * chore(s3): trim the comments and docstrings the review flagged Keep the two external facts that are not visible from the code, the 1024 byte object key cap and the 2048 byte metadata header cap, and drop the rest. --- litellm/constants.py | 6 + litellm/integrations/s3.py | 81 ++++- litellm/integrations/s3_v2.py | 20 +- tests/test_litellm/integrations/test_s3.py | 42 ++- tests/test_litellm/integrations/test_s3_v2.py | 288 ++++++++++++++++++ 5 files changed, 409 insertions(+), 28 deletions(-) diff --git a/litellm/constants.py b/litellm/constants.py index b9bea466247..9a50797f517 100644 --- a/litellm/constants.py +++ b/litellm/constants.py @@ -13,6 +13,12 @@ DEFAULT_BATCH_SIZE: Final = int(os.getenv("DEFAULT_BATCH_SIZE", 512)) DEFAULT_FLUSH_INTERVAL_SECONDS: Final = int(os.getenv("DEFAULT_FLUSH_INTERVAL_SECONDS", 5)) DEFAULT_S3_FLUSH_INTERVAL_SECONDS: Final = int(os.getenv("DEFAULT_S3_FLUSH_INTERVAL_SECONDS", 10)) DEFAULT_S3_BATCH_SIZE: Final = int(os.getenv("DEFAULT_S3_BATCH_SIZE", 512)) +# https://docs.aws.amazon.com/AmazonS3/latest/userguide/object-keys.html +MAX_S3_OBJECT_KEY_BYTES: Final = 1024 +S3_BOUNDED_OBJECT_KEY_HEAD_BYTES: Final = 64 +S3_PREFIX_DIGEST_CHARS: Final = 16 +# s3 allows 2048 bytes of combined metadata headers, which Content-Disposition counts against +MAX_S3_OBJECT_DOWNLOAD_FILENAME_BYTES: Final = 1024 DEFAULT_SQS_FLUSH_INTERVAL_SECONDS: Final = int(os.getenv("DEFAULT_SQS_FLUSH_INTERVAL_SECONDS", 10)) DEFAULT_NUM_WORKERS_LITELLM_PROXY: Final = int(os.getenv("DEFAULT_NUM_WORKERS_LITELLM_PROXY", 1)) DYNAMIC_RATE_LIMIT_ERROR_THRESHOLD_PER_MINUTE = int(os.getenv("DYNAMIC_RATE_LIMIT_ERROR_THRESHOLD_PER_MINUTE", 1)) diff --git a/litellm/integrations/s3.py b/litellm/integrations/s3.py index ddeb410c54a..8ce461eea5b 100644 --- a/litellm/integrations/s3.py +++ b/litellm/integrations/s3.py @@ -1,11 +1,18 @@ #### What this does #### # On success + failure, log events to Supabase +import hashlib from datetime import datetime from typing import Final, cast import litellm from litellm._logging import print_verbose, verbose_logger +from litellm.constants import ( + MAX_S3_OBJECT_DOWNLOAD_FILENAME_BYTES, + MAX_S3_OBJECT_KEY_BYTES, + S3_BOUNDED_OBJECT_KEY_HEAD_BYTES, + S3_PREFIX_DIGEST_CHARS, +) from litellm.types.utils import StandardLoggingPayload @@ -133,9 +140,7 @@ class S3Logger: s3_file_name, ) - s3_object_download_filename: Final = ( - "time-" + start_time.strftime("%Y-%m-%dT%H-%M-%S-%f") + "_" + payload["id"] + ".json" - ) + s3_object_download_filename: Final = get_s3_object_download_filename(start_time, payload["id"]) from litellm.litellm_core_utils.safe_json_dumps import safe_dumps @@ -198,6 +203,47 @@ def resolve_sse_params( return algorithm, valid_key_id +S3_MIN_BOUNDED_FILE_NAME_BYTES: Final = 64 + + +def _truncate_to_utf8_bytes(value: str, max_bytes: int) -> str: + """Trim `value` so its UTF-8 encoding fits `max_bytes`, never splitting a character.""" + if max_bytes <= 0: + return "" + encoded: Final = value.encode("utf-8") + if len(encoded) <= max_bytes: + return value + return encoded[:max_bytes].decode("utf-8", errors="ignore") + + +def get_s3_object_download_filename(start_time: datetime, response_id: str) -> str: + """Content-Disposition filename for the uploaded object, bounded to the metadata header cap.""" + sanitized_response_id: Final = response_id.replace("/", "_").replace('"', "_") + file_name: Final = f"time-{start_time.strftime('%Y-%m-%dT%H-%M-%S-%f')}_{response_id}" + sanitized_file_name: Final = f"time-{start_time.strftime('%Y-%m-%dT%H-%M-%S-%f')}_{sanitized_response_id}" + budget: Final = MAX_S3_OBJECT_DOWNLOAD_FILENAME_BYTES - len(b".json") + if len(sanitized_file_name.encode("utf-8")) <= budget: + return sanitized_file_name + ".json" + return _bounded_s3_file_name(file_name, sanitized_file_name, budget) + ".json" + + +def _bounded_s3_file_name(s3_file_name: str, sanitized_s3_file_name: str, max_bytes: int) -> str: + """As much of the file name as `max_bytes` allows, then the sha256 of the whole name.""" + digest: Final = hashlib.sha256(s3_file_name.encode("utf-8")).hexdigest() + head_budget: Final = min(S3_BOUNDED_OBJECT_KEY_HEAD_BYTES, max_bytes - len(digest) - 1) + head: Final = _truncate_to_utf8_bytes(sanitized_s3_file_name, head_budget) + return f"{head}_{digest}" if head else digest + + +def _bounded_s3_prefix(configured_prefix: str, max_bytes: int) -> str: + """As much of the configured prefix as fits, then a digest segment naming the full prefix.""" + digest_segment: Final = hashlib.sha256(configured_prefix.encode("utf-8")).hexdigest()[:S3_PREFIX_DIGEST_CHARS] + "/" + if max_bytes < len(digest_segment): + return "" + head: Final = _truncate_to_utf8_bytes(configured_prefix, max_bytes - len(digest_segment) - 1).rstrip("/") + return f"{head}/{digest_segment}" if head else digest_segment + + def get_s3_object_key( s3_path: str, prefix: str, @@ -205,12 +251,23 @@ def get_s3_object_key( s3_file_name: str, ) -> str: sanitized_s3_file_name: Final = s3_file_name.replace("/", "_") - s3_object_key = ( - (s3_path.rstrip("/") + "/" if s3_path else "") - + prefix - + start_time.strftime("%Y-%m-%d") - + "/" - + sanitized_s3_file_name - ) # we need the s3 key to include the time, so we log cache hits too - s3_object_key += ".json" - return s3_object_key + configured_prefix: Final = (s3_path.rstrip("/") + "/" if s3_path else "") + prefix + date_segment: Final = start_time.strftime("%Y-%m-%d") + "/" + # we need the s3 key to include the time, so we log cache hits too + s3_object_key: Final = configured_prefix + date_segment + sanitized_s3_file_name + ".json" + if len(s3_object_key.encode("utf-8")) <= MAX_S3_OBJECT_KEY_BYTES: + return s3_object_key + + # shorten the response id first and only trim the configured prefix if that is what does not + # fit, so prefix scoped IAM policies and lifecycle rules keep matching + budget: Final = MAX_S3_OBJECT_KEY_BYTES - len(date_segment.encode("utf-8")) - len(b".json") + prefix_bytes: Final = len(configured_prefix.encode("utf-8")) + if prefix_bytes + S3_MIN_BOUNDED_FILE_NAME_BYTES <= budget: + bounded_file_name: Final = _bounded_s3_file_name(s3_file_name, sanitized_s3_file_name, budget - prefix_bytes) + return configured_prefix + date_segment + bounded_file_name + ".json" + + shortest_file_name: Final = _bounded_s3_file_name( + s3_file_name, sanitized_s3_file_name, S3_MIN_BOUNDED_FILE_NAME_BYTES + ) + bounded_prefix: Final = _bounded_s3_prefix(configured_prefix, budget - len(shortest_file_name.encode("utf-8"))) + return bounded_prefix + date_segment + shortest_file_name + ".json" diff --git a/litellm/integrations/s3_v2.py b/litellm/integrations/s3_v2.py index 9f6ae72fb3a..712ce41d09e 100644 --- a/litellm/integrations/s3_v2.py +++ b/litellm/integrations/s3_v2.py @@ -16,7 +16,11 @@ from urllib.parse import quote import litellm from litellm._logging import print_verbose, verbose_logger from litellm.constants import DEFAULT_S3_BATCH_SIZE, DEFAULT_S3_FLUSH_INTERVAL_SECONDS -from litellm.integrations.s3 import get_s3_object_key, resolve_sse_params +from litellm.integrations.s3 import ( + get_s3_object_download_filename, + get_s3_object_key, + resolve_sse_params, +) from litellm.litellm_core_utils.aws_partition import get_aws_dns_suffix from litellm.litellm_core_utils.safe_json_dumps import safe_dumps from litellm.litellm_core_utils.sensitive_data_masker import SensitiveDataMasker @@ -259,11 +263,11 @@ class S3Logger(CustomBatchLogger, BaseAWSLLM): now: Final = datetime.now(timezone.utc) audit_log_id: Final = audit_log.get("id", "unknown") - s3_path = cast(str | None, self.s3_path) or "" - s3_path = s3_path.rstrip("/") + "/" if s3_path else "" - - s3_object_key: Final = ( - f"{s3_path}audit_logs/{now.strftime('%Y-%m-%d')}/{now.strftime('%H-%M-%S')}_{audit_log_id}.json" + s3_object_key: Final = get_s3_object_key( + cast(str | None, self.s3_path) or "", + "audit_logs/", + now, + f"{now.strftime('%H-%M-%S')}_{audit_log_id}", ) element: Final = s3BatchLoggingElement( @@ -463,9 +467,7 @@ class S3Logger(CustomBatchLogger, BaseAWSLLM): ) verbose_logger.debug("s3_object_key=%s", s3_object_key) - s3_object_download_filename: Final = ( - f"time-{start_time.strftime('%Y-%m-%dT%H-%M-%S-%f')}_{standard_logging_payload['id']}.json" - ) + s3_object_download_filename: Final = get_s3_object_download_filename(start_time, standard_logging_payload["id"]) return s3BatchLoggingElement( payload=dict(standard_logging_payload), diff --git a/tests/test_litellm/integrations/test_s3.py b/tests/test_litellm/integrations/test_s3.py index 7e997870852..58b15b79e76 100644 --- a/tests/test_litellm/integrations/test_s3.py +++ b/tests/test_litellm/integrations/test_s3.py @@ -2,26 +2,27 @@ from datetime import datetime from unittest.mock import MagicMock, patch import litellm +from litellm.constants import MAX_S3_OBJECT_DOWNLOAD_FILENAME_BYTES, MAX_S3_OBJECT_KEY_BYTES from litellm.integrations.s3 import S3Logger TEST_KMS_KEY_ARN = "arn:aws:kms:us-east-1:111122223333:key/test-key-id" -def _standard_logging_payload() -> dict: +def _standard_logging_payload(response_id: str = "chatcmpl-test-id") -> dict: return { - "id": "chatcmpl-test-id", + "id": response_id, "metadata": {"user_api_key_team_alias": None}, } -def _log_event_kwargs() -> dict: +def _log_event_kwargs(response_id: str = "chatcmpl-test-id") -> dict: return { "litellm_params": {"metadata": {}}, - "standard_logging_object": _standard_logging_payload(), + "standard_logging_object": _standard_logging_payload(response_id), } -def _run_log_event(callback_params: dict) -> MagicMock: +def _run_log_event(callback_params: dict, response_id: str = "chatcmpl-test-id") -> MagicMock: original = litellm.s3_callback_params litellm.s3_callback_params = callback_params try: @@ -30,8 +31,8 @@ def _run_log_event(callback_params: dict) -> MagicMock: mock_boto3_client.return_value = mock_s3_client logger = S3Logger() logger.log_event( - kwargs=_log_event_kwargs(), - response_obj={}, + kwargs=_log_event_kwargs(response_id), + response_obj={"id": response_id}, start_time=datetime(2026, 7, 30, 12, 0, 0), end_time=datetime(2026, 7, 30, 12, 0, 1), print_verbose=lambda *args, **kwargs: None, @@ -154,3 +155,30 @@ def test_non_string_key_id_is_dropped_and_valid_algorithm_is_kept(): put_object_kwargs = mock_s3_client.put_object.call_args.kwargs assert put_object_kwargs["ServerSideEncryption"] == "aws:kms" assert "SSEKMSKeyId" not in put_object_kwargs + + +def test_put_object_key_and_filename_are_bounded_for_an_oversized_response_id(): + """The sync logger bounds both the key and the Content-Disposition filename.""" + mock_s3_client = _run_log_event( + {"s3_bucket_name": "test-bucket", "s3_region_name": "us-west-2", "s3_path": "logs"}, + response_id="resp_" + "A" * 1100, + ) + + put_object_kwargs = mock_s3_client.put_object.call_args.kwargs + assert len(put_object_kwargs["Key"].encode("utf-8")) <= MAX_S3_OBJECT_KEY_BYTES + assert put_object_kwargs["Key"].startswith("logs/2026-07-30/time-12-00-00-000000_resp_") + filename = put_object_kwargs["ContentDisposition"].removeprefix('inline; filename="').removesuffix('"') + assert len(filename.encode("utf-8")) <= MAX_S3_OBJECT_DOWNLOAD_FILENAME_BYTES + + +def test_put_object_keeps_the_configured_path_intact_when_only_the_id_has_to_shrink(): + """A long configured s3_path survives whole when the id can be shortened instead.""" + long_path = "litellm-prod-logs/" + "t" * 921 + mock_s3_client = _run_log_event( + {"s3_bucket_name": "test-bucket", "s3_region_name": "us-west-2", "s3_path": long_path}, + response_id="resp_" + "B" * 100, + ) + + key = mock_s3_client.put_object.call_args.kwargs["Key"] + assert key.startswith(long_path + "/2026-07-30/") + assert len(key.encode("utf-8")) == MAX_S3_OBJECT_KEY_BYTES diff --git a/tests/test_litellm/integrations/test_s3_v2.py b/tests/test_litellm/integrations/test_s3_v2.py index 51671d5101e..a037284d7c1 100644 --- a/tests/test_litellm/integrations/test_s3_v2.py +++ b/tests/test_litellm/integrations/test_s3_v2.py @@ -1170,6 +1170,294 @@ def test_create_s3_batch_logging_element_flat_key_for_arn_response_id(): assert file_segment.endswith("model-invocation-job_gl18r6skk9yy.json") +# -------------------------------------------------------------- +# object keys bounded to S3's 1024 UTF-8 byte limit +# -------------------------------------------------------------- +def _oversized_response_id() -> str: + return "resp_" + "A" * 1100 + + +def test_s3_object_key_at_the_byte_limit_is_left_alone(): + """A key that still fits is left byte-identical.""" + from litellm.constants import MAX_S3_OBJECT_KEY_BYTES + from litellm.integrations.s3 import get_s3_object_key + + start_time = datetime(2026, 8, 24, 6, 18, 41, 948021) + fixed_len = len("input/2026-08-24/.json") + file_name = "x" * (MAX_S3_OBJECT_KEY_BYTES - fixed_len) + + key = get_s3_object_key(s3_path="input", prefix="", start_time=start_time, s3_file_name=file_name) + + assert key == f"input/2026-08-24/{file_name}.json" + assert len(key.encode("utf-8")) == MAX_S3_OBJECT_KEY_BYTES + + +def test_s3_object_key_is_bounded_for_oversized_response_id(): + """An oversized Responses API id is shortened to a readable head plus a digest.""" + import hashlib + + from litellm.constants import MAX_S3_OBJECT_KEY_BYTES + from litellm.integrations.s3 import get_s3_object_key + + start_time = datetime(2026, 8, 24, 6, 18, 41, 948021) + file_name = f"time-06-18-41-948021_{_oversized_response_id()}" + + key = get_s3_object_key(s3_path="input", prefix="DefaultTeamProd/", start_time=start_time, s3_file_name=file_name) + + assert len(key.encode("utf-8")) <= MAX_S3_OBJECT_KEY_BYTES + assert key.startswith("input/DefaultTeamProd/2026-08-24/time-06-18-41-948021_resp_") + assert key.endswith(f"_{hashlib.sha256(file_name.encode('utf-8')).hexdigest()}.json") + + +@pytest.mark.parametrize( + "s3_path,prefix", + [ + ("input", ""), + ("a" * 900, ""), + ("input", "team-" + "b" * 900 + "/"), + ("c" * 600, "team-" + "d" * 600 + "/key-" + "e" * 600 + "/"), + # many short segments, so the trim lands exactly on the budget edge + ("", "ssss/" * 200), + ], +) +def test_s3_object_key_is_bounded_for_long_paths_and_aliases(s3_path: str, prefix: str): + """Long paths, team aliases and key aliases stay within the cap.""" + from litellm.constants import MAX_S3_OBJECT_KEY_BYTES + from litellm.integrations.s3 import get_s3_object_key + + key = get_s3_object_key( + s3_path=s3_path, + prefix=prefix, + start_time=datetime(2026, 8, 24, 6, 18, 41, 948021), + s3_file_name=f"time-06-18-41-948021_{_oversized_response_id()}", + ) + + assert len(key.encode("utf-8")) <= MAX_S3_OBJECT_KEY_BYTES + assert key.endswith(".json") + assert "/2026-08-24/" in key or key.startswith("2026-08-24/") + assert "/" not in key.rsplit("2026-08-24/", 1)[1] + + +def test_s3_object_key_trimmed_prefixes_stay_distinct_per_operator(): + """Prefixes that differ only past the trim point keep separate folders.""" + from litellm.constants import MAX_S3_OBJECT_KEY_BYTES + from litellm.integrations.s3 import get_s3_object_key + + start_time = datetime(2026, 8, 24, 6, 18, 41, 948021) + keys = [ + get_s3_object_key( + s3_path="input", + prefix="team-" + "b" * 1000 + suffix + "/", + start_time=start_time, + s3_file_name=f"time-06-18-41-948021_{_oversized_response_id()}", + ) + for suffix in ("-one", "-two") + ] + + assert keys[0] != keys[1] + assert all(key.startswith("input/team-" + "b" * 900) for key in keys) + assert all(len(key.encode("utf-8")) == MAX_S3_OBJECT_KEY_BYTES for key in keys) + + +def test_s3_object_key_bounded_prefix_never_splits_a_multibyte_character(): + """A multibyte prefix is trimmed on a character boundary.""" + from litellm.constants import MAX_S3_OBJECT_KEY_BYTES + from litellm.integrations.s3 import get_s3_object_key + + s3_path = "\u65e5\u672c\u8a9e" * 200 + + key = get_s3_object_key( + s3_path=s3_path, + prefix="\u30c1\u30fc\u30e0" * 200 + "/", + start_time=datetime(2026, 8, 24, 6, 18, 41, 948021), + s3_file_name=f"time-06-18-41-948021_{_oversized_response_id()}", + ) + + assert len(key.encode("utf-8")) <= MAX_S3_OBJECT_KEY_BYTES + assert key.startswith(s3_path[:100]) + assert "\ufffd" not in key + + +def test_s3_object_key_stays_unique_for_ids_sharing_a_head(): + """Ids sharing a visible head still get distinct keys.""" + from litellm.integrations.s3 import get_s3_object_key + + start_time = datetime(2026, 8, 24, 6, 18, 41, 948021) + keys = { + get_s3_object_key( + s3_path="input", + prefix="", + start_time=start_time, + s3_file_name=f"time-06-18-41-948021_{_oversized_response_id()}{suffix}", + ) + for suffix in ("first", "second", "third") + } + + assert len(keys) == 3 + + +def test_s3_object_key_bounding_matches_the_documented_layout(): + """The bounded key is `//_.json`.""" + import hashlib + + from litellm.integrations.s3 import get_s3_object_key + + file_name = f"time-06-18-41-948021_{_oversized_response_id()}" + + key = get_s3_object_key( + s3_path="input", + prefix="team/", + start_time=datetime(2026, 8, 24, 6, 18, 41, 948021), + s3_file_name=file_name, + ) + + digest = hashlib.sha256(file_name.encode("utf-8")).hexdigest() + assert key == f"input/team/2026-08-24/{file_name[:64]}_{digest}.json" + + +def test_s3_object_key_keeps_the_configured_prefix_when_only_the_id_overflows(): + """A 940 byte configured prefix survives whole when only the id overflows.""" + from litellm.constants import MAX_S3_OBJECT_KEY_BYTES + from litellm.integrations.s3 import get_s3_object_key + + prefix = "team-" + "b" * 934 + "/" + + key = get_s3_object_key( + s3_path="", + prefix=prefix, + start_time=datetime(2026, 8, 24, 6, 18, 41, 948021), + s3_file_name=f"time-06-18-41-948021_{_oversized_response_id()}", + ) + + assert key.startswith(prefix + "2026-08-24/") + assert len(key.encode("utf-8")) == MAX_S3_OBJECT_KEY_BYTES + + +def test_s3_object_key_spends_the_whole_budget_when_the_prefix_must_be_trimmed(): + """A trimmed prefix keeps every byte the budget allows, not whole segments.""" + from litellm.constants import MAX_S3_OBJECT_KEY_BYTES + from litellm.integrations.s3 import get_s3_object_key + + s3_path = "p" * 400 + "/" + "q" * 600 + + key = get_s3_object_key( + s3_path=s3_path, + prefix="", + start_time=datetime(2026, 8, 24, 6, 18, 41, 948021), + s3_file_name="time-06-18-41-948021_abc", + ) + + assert len(key.encode("utf-8")) == MAX_S3_OBJECT_KEY_BYTES + assert key.startswith("p" * 400 + "/" + "q" * 500) + + +def test_s3_object_key_keeps_a_single_segment_path_as_far_as_it_fits(): + """A path with no separator is kept as far as it fits, never dropped to the bucket root.""" + from litellm.constants import MAX_S3_OBJECT_KEY_BYTES + from litellm.integrations.s3 import get_s3_object_key + + key = get_s3_object_key( + s3_path="a" * 1050, + prefix="", + start_time=datetime(2026, 8, 24, 6, 18, 41, 948021), + s3_file_name="time-06-18-41-948021_chatcmpl-xyz", + ) + + assert len(key.encode("utf-8")) == MAX_S3_OBJECT_KEY_BYTES + assert key.startswith("a" * 900) + + +def test_create_s3_batch_logging_element_bounds_key_and_keeps_full_response_id(): + """The batch element bounds the key and keeps the full response id in the payload.""" + from litellm.constants import MAX_S3_OBJECT_KEY_BYTES + + logger = S3Logger(s3_use_team_prefix=True, s3_use_key_prefix=True) + response_id = _oversized_response_id() + payload = StandardLoggingPayload( + id=response_id, + metadata={"user_api_key_team_alias": "DefaultTeamProd", "user_api_key_alias": "prod-key"}, + messages=[], + ) + + result = logger.create_s3_batch_logging_element(datetime(2026, 8, 24, 6, 18, 41, 948021), payload) + + assert result is not None + assert len(result.s3_object_key.encode("utf-8")) <= MAX_S3_OBJECT_KEY_BYTES + assert result.s3_object_key.startswith("DefaultTeamProd/prod-key/2026-08-24/") + assert result.payload["id"] == response_id + + +def test_s3_object_download_filename_is_bounded_for_oversized_response_id(): + """The Content-Disposition filename is bounded too, or the PUT fails with MetadataTooLarge.""" + from litellm.constants import MAX_S3_OBJECT_DOWNLOAD_FILENAME_BYTES + from litellm.integrations.s3 import get_s3_object_download_filename + + file_name = get_s3_object_download_filename(datetime(2026, 8, 24, 6, 18, 41, 948021), _oversized_response_id()) + + assert len(file_name.encode("utf-8")) <= MAX_S3_OBJECT_DOWNLOAD_FILENAME_BYTES + assert file_name.startswith("time-2026-08-24T06-18-41-948021_resp_") + assert file_name.endswith(".json") + + +def test_s3_object_download_filenames_stay_distinct_when_shortened(): + """Shortened filenames stay distinct.""" + from litellm.integrations.s3 import get_s3_object_download_filename + + start_time = datetime(2026, 8, 24, 6, 18, 41, 948021) + file_names = { + get_s3_object_download_filename(start_time, _oversized_response_id() + suffix) + for suffix in ("first", "second", "third") + } + + assert len(file_names) == 3 + + +def test_s3_object_download_filename_short_id_is_unchanged(): + """An ordinary response id keeps the filename it had before.""" + from litellm.integrations.s3 import get_s3_object_download_filename + + file_name = get_s3_object_download_filename(datetime(2026, 8, 24, 6, 18, 41, 948021), "resp_abc123") + + assert file_name == "time-2026-08-24T06-18-41-948021_resp_abc123.json" + + +def test_create_s3_batch_logging_element_bounds_the_download_filename(): + """The batch element carries a bounded Content-Disposition filename.""" + from litellm.constants import MAX_S3_OBJECT_DOWNLOAD_FILENAME_BYTES + + logger = S3Logger() + payload = StandardLoggingPayload(id=_oversized_response_id(), metadata={}, messages=[]) + + result = logger.create_s3_batch_logging_element(datetime(2026, 8, 24, 6, 18, 41, 948021), payload) + + assert result is not None + assert len(result.s3_object_download_filename.encode("utf-8")) <= MAX_S3_OBJECT_DOWNLOAD_FILENAME_BYTES + + +@pytest.mark.asyncio +async def test_audit_log_object_key_is_bounded_for_a_long_configured_path(): + """Audit log keys are bounded by the same builder.""" + from litellm.constants import MAX_S3_OBJECT_KEY_BYTES + + logger = S3Logger() + logger.s3_path = "audit-archive/" + "z" * 1100 + + await logger.async_log_audit_log_event({"id": "1a4f7bd0-6f1e-4d0a-9b3c-9f2e1d5a7c88"}) + + assert len(logger.log_queue) == 1 + assert len(logger.log_queue[0].s3_object_key.encode("utf-8")) <= MAX_S3_OBJECT_KEY_BYTES + assert logger.log_queue[0].s3_object_key.startswith("audit-archive/" + "z" * 900) + + +def test_s3_object_download_filename_drops_characters_that_break_the_header(): + """A quote or separator in the response id cannot escape the quoted header value.""" + from litellm.integrations.s3 import get_s3_object_download_filename + + file_name = get_s3_object_download_filename(datetime(2026, 8, 24, 6, 18, 41, 948021), 'resp_a"b/c') + + assert file_name == "time-2026-08-24T06-18-41-948021_resp_a_b_c.json" + + # -------------------------------------------------------------- # params_source / s3_callback_params_override (audit-log decoupling) # --------------------------------------------------------------