Skip to content

~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.

22.6 KB
Raw Blame
"""Test context overflow detection for LLM calls."""

import os
import uuid
from unittest.mock import Mock, patch

import pytest
from sqlalchemy import create_engine
from sqlalchemy.orm import sessionmaker

from local_deep_research.database.models import Base
from local_deep_research.metrics.token_counter import TokenCountingCallback


def _overflow_warning(mock_logger) -> str:
    """Return the concatenated text of the "Context overflow detected"
    warning among all ``logger.warning`` calls (``""`` if none).

    The callback also persists metrics, and that path's failure handler now
    logs via ``logger.warning`` too (#4182 sink/redaction sweep), so the
    overflow warning is no longer guaranteed to be the *last* warning call.
    Match it by its stable anchor instead of by call position.
    """
    for call in mock_logger.warning.call_args_list:
        text = " ".join(str(a) for a in call.args)
        if "Context overflow detected" in text:
            return text
    return ""


class TestContextOverflowDetection:
    """Test suite for context overflow detection."""

    @pytest.fixture
    def db_session(self):
        """Create an in-memory database session for testing."""
        engine = create_engine("sqlite:///:memory:")
        Base.metadata.create_all(engine)
        Session = sessionmaker(bind=engine)
        session = Session()
        yield session
        session.close()
        engine.dispose()

    @pytest.fixture
    def token_callback(self):
        """Create a token counting callback for testing."""
        research_id = str(uuid.uuid4())
        research_context = {
            "research_query": "Test query",
            "research_mode": "test",
            "context_limit": 2048,  # Set a specific context limit
            "username": "test_user",
            "user_password": "test_pass",
        }
        return TokenCountingCallback(research_id, research_context)

    def test_context_overflow_detection_no_overflow(self, token_callback):
        """Test that no overflow is detected for small prompts."""
        # Simulate LLM start with small prompt
        prompts = ["What is 2+2?"]
        token_callback.on_llm_start({}, prompts)

        # Create mock response with Ollama-style metadata
        mock_response = Mock()
        mock_response.llm_output = None  # Explicitly set to None
        mock_response.generations = [[Mock()]]
        mock_response.generations[0][0].message = Mock()
        mock_response.generations[0][
            0
        ].message.usage_metadata = None  # No usage_metadata
        mock_response.generations[0][0].message.response_metadata = {
            "prompt_eval_count": 5,  # Small token count
            "eval_count": 10,
            "total_duration": 1000000000,  # 1 second in nanoseconds
        }

        # Process response
        token_callback.on_llm_end(mock_response)

        # Verify no overflow detected
        assert token_callback.context_truncated is False
        assert token_callback.tokens_truncated == 0
        assert token_callback.truncation_ratio == 0.0
        assert token_callback.ollama_metrics.get("prompt_eval_count") == 5

    def test_context_overflow_detection_with_overflow(self, token_callback):
        """Test that overflow is detected when prompt approaches context limit."""
        # Simulate LLM start with large prompt
        large_text = "The quick brown fox jumps over the lazy dog. " * 500
        prompts = [large_text]
        token_callback.on_llm_start({}, prompts)

        # Create mock response indicating near-limit token usage
        mock_response = Mock()
        mock_response.llm_output = None
        mock_response.generations = [[Mock()]]
        mock_response.generations[0][0].message = Mock()
        mock_response.generations[0][0].message.usage_metadata = None
        mock_response.generations[0][0].message.response_metadata = {
            "prompt_eval_count": 1950,  # Near the 2048 limit (95%)
            "eval_count": 50,
            "total_duration": 5000000000,  # 5 seconds
            "prompt_eval_duration": 4000000000,
            "eval_duration": 1000000000,
        }

        # Process response
        token_callback.on_llm_end(mock_response)

        # Verify overflow detected
        assert token_callback.context_truncated is True
        assert token_callback.tokens_truncated > 0  # Should estimate truncation
        assert token_callback.truncation_ratio > 0
        assert token_callback.ollama_metrics["prompt_eval_count"] == 1950

    def test_ollama_raw_metrics_capture(self, token_callback):
        """Test that raw Ollama metrics are properly captured."""
        prompts = ["Test prompt"]
        token_callback.on_llm_start({}, prompts)

        # Create mock response with full Ollama metrics
        mock_response = Mock()
        mock_response.llm_output = None
        mock_response.generations = [[Mock()]]
        mock_response.generations[0][0].message = Mock()
        mock_response.generations[0][0].message.usage_metadata = None
        mock_response.generations[0][0].message.response_metadata = {
            "prompt_eval_count": 100,
            "eval_count": 200,
            "total_duration": 3000000000,
            "load_duration": 500000000,
            "prompt_eval_duration": 2000000000,
            "eval_duration": 500000000,
        }

        # Process response
        token_callback.on_llm_end(mock_response)

        # Verify all metrics captured
        assert token_callback.ollama_metrics["prompt_eval_count"] == 100
        assert token_callback.ollama_metrics["eval_count"] == 200
        assert token_callback.ollama_metrics["total_duration"] == 3000000000
        assert token_callback.ollama_metrics["load_duration"] == 500000000
        assert (
            token_callback.ollama_metrics["prompt_eval_duration"] == 2000000000
        )
        assert token_callback.ollama_metrics["eval_duration"] == 500000000

    def test_context_limit_from_research_context(self):
        """Test that context limit is properly read from research context."""
        # Create callback with context limit
        callback = TokenCountingCallback("test-id", {"context_limit": 2048})
        # Context limit is set on llm_start
        callback.on_llm_start({}, ["test"])
        assert callback.context_limit == 2048

        # Test with different limit
        callback_4k = TokenCountingCallback("test-id", {"context_limit": 4096})
        callback_4k.on_llm_start({}, ["test"])
        assert callback_4k.context_limit == 4096

        # Test with no limit
        callback_no_limit = TokenCountingCallback("test-id", {})
        callback_no_limit.on_llm_start({}, ["test"])
        assert callback_no_limit.context_limit is None

    def test_prompt_size_estimation(self, token_callback):
        """Test that prompt size is estimated correctly."""
        # Test with single prompt
        prompts = ["This is a test prompt with approximately 10 words."]
        token_callback.on_llm_start({}, prompts)

        # Rough estimate: ~4 chars per token
        expected_tokens = len(prompts[0]) // 4
        assert (
            abs(token_callback.original_prompt_estimate - expected_tokens) < 5
        )

        # Test with multiple prompts
        token_callback.original_prompt_estimate = 0
        prompts = ["First prompt.", "Second prompt.", "Third prompt."]
        total_chars = sum(len(p) for p in prompts)
        token_callback.on_llm_start({}, prompts)

        expected_tokens = total_chars // 4
        assert (
            abs(token_callback.original_prompt_estimate - expected_tokens) < 5
        )

    @patch("local_deep_research.metrics.token_counter.logger")
    def test_overflow_warning_logged(self, mock_logger, token_callback):
        """Test that overflow detection logs a warning."""
        # Create large prompt
        large_text = "word " * 10000
        prompts = [large_text]
        token_callback.on_llm_start({}, prompts)

        # Mock response at context limit
        mock_response = Mock()
        mock_response.llm_output = None
        mock_response.generations = [[Mock()]]
        mock_response.generations[0][0].message = Mock()
        mock_response.generations[0][0].message.usage_metadata = None
        mock_response.generations[0][0].message.response_metadata = {
            "prompt_eval_count": 2000,  # At 95% of 2048 limit
            "eval_count": 10,
        }

        token_callback.on_llm_end(mock_response)

        # Verify warning was logged with structured fields
        mock_logger.warning.assert_called()
        # The new logger.warning is called with multiple string args that get
        # concatenated by f-strings; inspect each positional arg for the fields.
        warning_call = _overflow_warning(mock_logger)
        assert "Context overflow detected" in warning_call
        assert "[provider-confirmed]" in warning_call
        assert "prompt_tokens=2000" in warning_call
        assert "context_limit=2048" in warning_call

    @patch("local_deep_research.metrics.token_counter.logger")
    def test_estimated_overflow_for_non_ollama_provider(
        self, mock_logger, token_callback
    ):
        """Detection fires from prompt estimate when provider doesn't echo prompt_eval_count."""
        # Use preset_model/preset_provider so on_llm_start initializes by_model.
        token_callback.preset_model = "gpt-4"
        token_callback.preset_provider = "openai"

        # Large prompt: ~10000 tokens (well over 2048 context_limit)
        large_text = "word " * 10000
        token_callback.on_llm_start({}, [large_text])

        # OpenAI-style response: usage_metadata, no prompt_eval_count
        mock_response = Mock()
        mock_response.llm_output = None
        mock_response.generations = [[Mock()]]
        mock_response.generations[0][0].message = Mock()
        mock_response.generations[0][0].message.usage_metadata = {
            "input_tokens": 100,
            "output_tokens": 50,
            "total_tokens": 150,
        }
        mock_response.generations[0][0].message.response_metadata = {}

        token_callback.on_llm_end(mock_response)

        # Estimation path should mark truncation and log
        assert token_callback.context_truncated is True
        assert token_callback.tokens_truncated > 0
        mock_logger.warning.assert_called()
        warning_call = _overflow_warning(mock_logger)
        assert "[estimated]" in warning_call
        assert "provider=openai" in warning_call
        assert "context_limit=2048" in warning_call

    @patch("local_deep_research.metrics.token_counter.logger")
    def test_provider_confirmed_total_context_overflow(
        self, mock_logger, token_callback
    ):
        """[total-context] fires when input+output exceeds limit but input alone doesn't.

        Input must stay strictly under the input-only threshold (80% of
        context_limit, set by PR #3792 / #3840) so the input-only branch
        does NOT fire and we exercise the total-context elif path.
        """
        large_text = "word " * 500
        token_callback.on_llm_start({}, [large_text])

        # Input is below 80% of 2048 (1638); but input + output >= 95% of 2048 (1945).
        mock_response = Mock()
        mock_response.llm_output = None
        mock_response.generations = [[Mock()]]
        mock_response.generations[0][0].message = Mock()
        mock_response.generations[0][0].message.usage_metadata = None
        mock_response.generations[0][0].message.response_metadata = {
            "prompt_eval_count": 1500,  # 73% of 2048 — below 80% input-only threshold
            "eval_count": 500,  # 1500+500=2000 >= 95% of 2048 (1945)
            "total_duration": 3000000000,
        }

        token_callback.on_llm_end(mock_response)

        assert token_callback.context_truncated is True
        mock_logger.warning.assert_called()
        warning_call = _overflow_warning(mock_logger)
        assert "[total-context]" in warning_call
        assert "prompt_tokens=1500" in warning_call
        assert "completion_tokens=500" in warning_call
        assert "total_tokens=2000" in warning_call
        assert "context_limit=2048" in warning_call

    @patch("local_deep_research.metrics.token_counter.logger")
    def test_estimated_total_context_overflow_for_non_ollama(
        self, mock_logger, token_callback
    ):
        """[total-context] fires when hosted provider input+output exceeds limit.

        Input alone stays below 80% of 2048 (input-only threshold, set by
        #3792/#3840) so the elif total-context branch is exercised.
        """
        token_callback.preset_model = "gpt-4"
        token_callback.preset_provider = "openai"

        # Prompt estimate below context_limit but not by much
        medium_text = "word " * 1500  # ~1875 estimated tokens
        token_callback.on_llm_start({}, [medium_text])

        # Response reports actual tokens: input below 80% but input+output >= 95%.
        mock_response = Mock()
        mock_response.llm_output = None
        mock_response.generations = [[Mock()]]
        mock_response.generations[0][0].message = Mock()
        mock_response.generations[0][0].message.usage_metadata = {
            "input_tokens": 1500,  # 73% of 2048 — below 80% input-only threshold
            "output_tokens": 600,  # 1500+600=2100 > 2048 (so tokens_truncated > 0)
            "total_tokens": 2100,
        }
        mock_response.generations[0][0].message.response_metadata = {}

        token_callback.on_llm_end(mock_response)

        assert token_callback.context_truncated is True
        assert token_callback.tokens_truncated > 0
        mock_logger.warning.assert_called()
        warning_call = _overflow_warning(mock_logger)
        assert "[total-context]" in warning_call
        assert "prompt_tokens=1500" in warning_call
        assert "completion_tokens=600" in warning_call
        assert "total_tokens=2100" in warning_call
        assert "context_limit=2048" in warning_call

    @patch("local_deep_research.metrics.token_counter.logger")
    def test_estimated_total_context_overflow_via_llm_output(
        self, mock_logger, token_callback
    ):
        """[estimated-total-context] fires when token_usage comes from llm_output.

        Distinct from the [total-context] path: this exercises the post-loop
        estimation block (token_counter.py ~line 440) where token_usage is
        sourced from response.llm_output rather than from the per-generation
        usage_metadata that _check_context_overflow consumes.

        Conditions for this path:
          - token_usage populated from llm_output (not from generations)
          - generations have NO usage_metadata AND NO response_metadata
            (so _check_context_overflow is never called)
          - original_prompt_estimate <= context_limit (so the [estimated]
            input-only branch above doesn't fire)
          - prompt_tokens + completion_tokens >= 95% of context_limit
        """
        token_callback.preset_model = "gpt-4"
        token_callback.preset_provider = "openai"

        # Estimate well below context_limit — input-only [estimated] stays quiet.
        small_text = "word " * 100  # ~125 estimated tokens
        token_callback.on_llm_start({}, [small_text])

        # token_usage flows in via llm_output; generations carry no metadata
        # so _check_context_overflow is never called.
        mock_response = Mock()
        mock_response.llm_output = {
            "token_usage": {
                "prompt_tokens": 1500,  # 73% of 2048 — below 80%
                "completion_tokens": 600,  # 1500+600=2100, past context_limit
                "total_tokens": 2100,
            }
        }
        # Empty generations list → loop never enters _check_context_overflow.
        mock_response.generations = []

        token_callback.on_llm_end(mock_response)

        assert token_callback.context_truncated is True
        assert token_callback.tokens_truncated > 0
        mock_logger.warning.assert_called()
        warning_call = _overflow_warning(mock_logger)
        assert "[estimated-total-context]" in warning_call
        assert "prompt_tokens=1500" in warning_call
        assert "completion_tokens=600" in warning_call
        assert "total_tokens=2100" in warning_call
        assert "context_limit=2048" in warning_call

    @patch("local_deep_research.metrics.token_counter.logger")
    def test_estimated_path_fires_on_subsequent_calls_after_first_truncation(
        self, mock_logger, token_callback
    ):
        """Regression: TokenCountingCallback is reused across LLM calls in a
        research session (see config/llm_config.py wrap_llm). on_llm_start
        must reset context_truncated/tokens_truncated/truncation_ratio,
        otherwise the post-loop estimation block's `if not context_truncated`
        guard silently disables [estimated] / [estimated-total-context] for
        every call after the first one that truncates.
        """
        token_callback.preset_model = "gpt-4"
        token_callback.preset_provider = "openai"

        def llm_output_overflow_response():
            mock_response = Mock()
            mock_response.llm_output = {
                "token_usage": {
                    "prompt_tokens": 1500,
                    "completion_tokens": 600,
                    "total_tokens": 2100,
                }
            }
            mock_response.generations = []
            return mock_response

        # Call 1: triggers [estimated-total-context] → context_truncated=True.
        token_callback.on_llm_start({}, ["word " * 100])
        token_callback.on_llm_end(llm_output_overflow_response())
        first_call_count = mock_logger.warning.call_count
        assert first_call_count >= 1
        assert token_callback.context_truncated is True

        # Call 2: same overflow shape — must ALSO log, not be silenced by
        # leftover state from Call 1.
        token_callback.on_llm_start({}, ["word " * 100])
        # Reset asserted by the second on_llm_start clearing prior state:
        assert token_callback.context_truncated is False
        token_callback.on_llm_end(llm_output_overflow_response())
        assert mock_logger.warning.call_count > first_call_count
        assert token_callback.context_truncated is True

    @patch("local_deep_research.metrics.token_counter.logger")
    def test_estimated_overflow_skipped_when_context_limit_none(
        self, mock_logger, token_callback
    ):
        """Estimation path should not fire when context_limit is not set."""
        token_callback.preset_model = "gpt-4"
        token_callback.preset_provider = "openai"
        # Remove context_limit
        token_callback.research_context = {}

        large_text = "word " * 10000
        token_callback.on_llm_start({}, [large_text])

        mock_response = Mock()
        mock_response.llm_output = None
        mock_response.generations = [[Mock()]]
        mock_response.generations[0][0].message = Mock()
        mock_response.generations[0][0].message.usage_metadata = {
            "input_tokens": 100,
            "output_tokens": 50,
            "total_tokens": 150,
        }
        mock_response.generations[0][0].message.response_metadata = {}

        token_callback.on_llm_end(mock_response)

        assert token_callback.context_truncated is False

    @patch("local_deep_research.metrics.token_counter.logger")
    def test_estimated_overflow_zero_prompt_estimate(
        self, mock_logger, token_callback
    ):
        """Estimation path should not fire when prompt estimate is 0."""
        token_callback.preset_model = "gpt-4"
        token_callback.preset_provider = "openai"
        token_callback.original_prompt_estimate = 0

        mock_response = Mock()
        mock_response.llm_output = None
        mock_response.generations = [[Mock()]]
        mock_response.generations[0][0].message = Mock()
        mock_response.generations[0][0].message.usage_metadata = {
            "input_tokens": 100,
            "output_tokens": 50,
            "total_tokens": 150,
        }
        mock_response.generations[0][0].message.response_metadata = {}

        token_callback.on_llm_end(mock_response)

        assert token_callback.context_truncated is False
        assert token_callback.tokens_truncated == 0

    @patch("local_deep_research.metrics.token_counter.logger")
    def test_estimated_overflow_without_on_llm_start(
        self, mock_logger, token_callback
    ):
        """Estimation path should not crash if on_llm_start was never called."""
        # Don't call on_llm_start — fields stay at defaults
        mock_response = Mock()
        mock_response.llm_output = None
        mock_response.generations = [[Mock()]]
        mock_response.generations[0][0].message = Mock()
        mock_response.generations[0][0].message.usage_metadata = {
            "input_tokens": 100,
            "output_tokens": 50,
            "total_tokens": 150,
        }
        mock_response.generations[0][0].message.response_metadata = {}

        # Should not raise and should not mark truncation
        token_callback.on_llm_end(mock_response)
        assert token_callback.context_truncated is False


@pytest.mark.skipif(
    os.environ.get("SKIP_OLLAMA_TESTS", "true").lower() == "true",
    reason="Ollama integration tests skipped",
)
class TestContextOverflowIntegration:
    """Integration tests with actual Ollama (when available)."""

    @pytest.mark.slow
    def test_ollama_context_overflow_real(self):
        """Test with real Ollama instance if available."""
        from langchain_ollama import ChatOllama
        from local_deep_research.llm.providers.implementations.ollama import (
            OllamaProvider,
        )

        if not OllamaProvider.is_available():
            pytest.skip("Ollama not available")

        # Create LLM with small context window
        llm = ChatOllama(
            model="llama3.2:latest",
            num_ctx=512,  # Very small context for testing
            temperature=0.1,
        )

        # Create callback
        research_id = str(uuid.uuid4())
        callback = TokenCountingCallback(research_id, {"context_limit": 512})

        # Create prompt that will likely overflow
        large_prompt = "Please analyze this text: " + ("word " * 200)

        # Run with callback
        try:
            _ = llm.invoke(large_prompt, config={"callbacks": [callback]})

            # Check if overflow was detected
            if callback.ollama_metrics.get("prompt_eval_count"):
                prompt_tokens = callback.ollama_metrics["prompt_eval_count"]
                if prompt_tokens >= 512 * 0.80:
                    assert callback.context_truncated is True
                    print(f"✅ Overflow detected: {prompt_tokens}/512 tokens")
                else:
                    print(f"ℹ️ No overflow: {prompt_tokens}/512 tokens")
        except AssertionError:
            raise
        except Exception as e:
            pytest.skip(f"Ollama test failed: {e}")