diff --git a/langfuse/_task_manager/score_ingestion_consumer.py b/langfuse/_task_manager/score_ingestion_consumer.py index 1dce00f80..131fcfd3c 100644 --- a/langfuse/_task_manager/score_ingestion_consumer.py +++ b/langfuse/_task_manager/score_ingestion_consumer.py @@ -82,6 +82,19 @@ def _next(self) -> list: item_size = self._get_item_size(event) + if item_size > MAX_EVENT_SIZE_BYTES: + logger.error( + "Data error: Score event %s is %s bytes, exceeding the maximum " + "event size of %s bytes (LANGFUSE_MAX_EVENT_SIZE_BYTES). Score " + "will be dropped.", + event.get("id"), + item_size, + MAX_EVENT_SIZE_BYTES, + ) + self._ingestion_queue.task_done() + + continue + # check for serialization errors try: json.dumps(event, cls=EventSerializer) diff --git a/tests/unit/test_score_ingestion_consumer.py b/tests/unit/test_score_ingestion_consumer.py new file mode 100644 index 000000000..3981b98a8 --- /dev/null +++ b/tests/unit/test_score_ingestion_consumer.py @@ -0,0 +1,93 @@ +import logging +from queue import Queue +from unittest.mock import Mock + +from pydantic import BaseModel + +from langfuse._task_manager import score_ingestion_consumer as sic +from langfuse._task_manager.score_ingestion_consumer import ScoreIngestionConsumer + + +def _consumer(queue: Queue, client: Mock, flush_at: int = 15): + return ScoreIngestionConsumer( + ingestion_queue=queue, + identifier=0, + client=client, + public_key="pk", + flush_at=flush_at, + flush_interval=0.1, + ) + + +def _event(event_id: str, comment: str) -> dict: + return {"id": event_id, "type": "score-create", "body": {"comment": comment}} + + +def test_oversized_event_is_dropped_and_others_are_sent(monkeypatch, caplog): + monkeypatch.setattr(sic, "MAX_EVENT_SIZE_BYTES", 1_000) + queue = Queue() + client = Mock() + # flush_at=2 ends the batch once both valid events are read, independent of timing + consumer = _consumer(queue, client, flush_at=2) + + queue.put(_event("a", "small")) + queue.put(_event("big", "x" * 5_000)) + queue.put(_event("c", "small")) + + with caplog.at_level(logging.ERROR): + consumer.upload() + + batch = client.batch_post.call_args.kwargs["batch"] + assert [e["id"] for e in batch] == ["a", "c"] + assert queue.unfinished_tasks == 0 + assert len(caplog.records) == 1 + assert "Score event big is" in caplog.text + + +def test_only_oversized_event_posts_nothing(monkeypatch): + monkeypatch.setattr(sic, "MAX_EVENT_SIZE_BYTES", 1_000) + queue = Queue() + client = Mock() + consumer = _consumer(queue, client) + + queue.put(_event("big", "x" * 5_000)) + + consumer.upload() + + client.batch_post.assert_not_called() + assert queue.unfinished_tasks == 0 + + +def test_event_exactly_at_limit_is_sent(monkeypatch): + queue = Queue() + client = Mock() + consumer = _consumer(queue, client, flush_at=1) + event = _event("edge", "x" * 100) + size = consumer._get_item_size(event) + + monkeypatch.setattr(sic, "MAX_EVENT_SIZE_BYTES", size) + queue.put(event) + consumer.upload() + assert client.batch_post.call_count == 1 + + monkeypatch.setattr(sic, "MAX_EVENT_SIZE_BYTES", size - 1) + queue.put(_event("edge", "x" * 100)) + consumer.upload() + assert client.batch_post.call_count == 1 + + +def test_pydantic_body_is_measured_after_dump(monkeypatch): + class Body(BaseModel): + comment: str + + monkeypatch.setattr(sic, "MAX_EVENT_SIZE_BYTES", 1_000) + queue = Queue() + client = Mock() + consumer = _consumer(queue, client) + + queue.put({"id": "big", "type": "score-create", "body": Body(comment="x" * 5_000)}) + + consumer.upload() + + client.batch_post.assert_not_called() + assert queue.unfinished_tasks == 0