"""Tests for gateway session hygiene — auto-compression of large sessions. Verifies that the gateway detects pathologically large transcripts and triggers auto-compression before running the agent. (#628) The hygiene system uses the SAME compression config as the agent: compression.threshold × model context length so CLI and messaging platforms behave identically. """ import asyncio import importlib import sys import threading import time import types from datetime import datetime from types import SimpleNamespace from unittest.mock import MagicMock, AsyncMock import pytest from agent.model_metadata import estimate_messages_tokens_rough from gateway.config import GatewayConfig, Platform, PlatformConfig from gateway.platforms.base import BasePlatformAdapter, SendResult from gateway.platforms.event import MessageEvent from gateway.session import SessionEntry, SessionSource # --------------------------------------------------------------------------- # Helpers # --------------------------------------------------------------------------- def _make_history(n_messages: int, content_size: int = 100) -> list: """Build a fake transcript with n_messages user/assistant pairs.""" history = [] content = "x" * content_size for i in range(n_messages): role = "user" if i % 2 == 0 else "assistant" history.append({"role": role, "content": content, "timestamp": f"t{i}"}) return history def _make_large_history_tokens(target_tokens: int) -> list: """Build a history that estimates to roughly target_tokens tokens.""" # estimate_messages_tokens_rough counts total chars in str(msg) // 4 # Each msg dict has ~60 chars of overhead + content chars # So for N tokens we need roughly N * 4 total chars across all messages target_chars = target_tokens * 4 # Each message as a dict string is roughly len(content) + 60 chars msg_overhead = 60 # Use 50 messages with appropriately sized content n_msgs = 50 content_size = max(10, (target_chars // n_msgs) - msg_overhead) return _make_history(n_msgs, content_size=content_size) class HygieneCaptureAdapter(BasePlatformAdapter): def __init__(self): super().__init__(PlatformConfig(enabled=True, token="fake-token"), Platform.TELEGRAM) self.sent = [] async def connect(self, *, is_reconnect: bool = False) -> bool: return True async def disconnect(self) -> None: return None async def send(self, chat_id, content, reply_to=None, metadata=None) -> SendResult: self.sent.append( { "chat_id": chat_id, "content": content, "reply_to": reply_to, "metadata": metadata, } ) return SendResult(success=True, message_id="hygiene-1") async def get_chat_info(self, chat_id: str): return {"id": chat_id} # --------------------------------------------------------------------------- # Detection threshold tests (model-aware, unified with compression config) # --------------------------------------------------------------------------- class TestSessionHygieneThresholds: """Test that the threshold logic correctly identifies large sessions. Thresholds are derived from model context length × compression threshold, matching what the agent's ContextCompressor uses. """ def test_under_threshold_no_trigger(self): """Session under threshold should not trigger, even with many messages.""" # 250 short messages — lots of messages but well under token threshold history = _make_history(250, content_size=10) approx_tokens = estimate_messages_tokens_rough(history) # 200k model at 85% = 170k token threshold context_length = 200_000 threshold_pct = 0.85 compress_token_threshold = int(context_length * threshold_pct) needs_compress = approx_tokens >= compress_token_threshold assert not needs_compress, ( f"250 short messages (~{approx_tokens} tokens) should NOT trigger " f"compression at {compress_token_threshold} token threshold" ) def test_message_count_alone_does_not_trigger(self): """Message count alone should NOT trigger — only token count matters. The old system used an OR of token-count and message-count thresholds, which caused premature compression in tool-heavy sessions with 200+ messages but low total tokens. """ # 300 very short messages — old system would compress, new should not history = _make_history(300, content_size=10) approx_tokens = estimate_messages_tokens_rough(history) context_length = 200_000 threshold_pct = 0.85 compress_token_threshold = int(context_length * threshold_pct) # Token-based check only needs_compress = approx_tokens >= compress_token_threshold assert not needs_compress def test_threshold_scales_with_model(self): """Different models should have different compression thresholds.""" # 128k model at 85% = 108,800 tokens small_model_threshold = int(128_000 * 0.85) # 200k model at 85% = 170,000 tokens large_model_threshold = int(200_000 * 0.85) # 1M model at 85% = 850,000 tokens huge_model_threshold = int(1_000_000 * 0.85) # A session at ~120k tokens: history = _make_large_history_tokens(120_000) approx_tokens = estimate_messages_tokens_rough(history) # Should trigger for 128k model assert approx_tokens >= small_model_threshold # Should NOT trigger for 200k model assert approx_tokens < large_model_threshold # Should NOT trigger for 1M model assert approx_tokens < huge_model_threshold @pytest.mark.parametrize("total_exhausted", [True, False]) def test_hygiene_timeout_warning_names_chat_commands_not_config(total_exhausted): """The chat user cannot edit model config or read second counts; the notice names the slash commands they can run and keeps the timing detail in the log.""" from gateway.run import _hygiene_compression_timeout_message warning = _hygiene_compression_timeout_message( total_exhausted=total_exhausted, elapsed=600.4, idle_timeout=30.0, progress_observed=True, ) assert "/compress" in warning and "/new" in warning assert "600.4" not in warning and "30.0" not in warning assert "auxiliary" not in warning and "/reset" not in warning class TestSessionHygieneWarnThreshold: """Test the post-compression warning threshold (95% of context).""" def test_warn_when_still_large(self): """If compressed result is still above 95% of context, should warn.""" context_length = 200_000 warn_threshold = int(context_length * 0.95) # 190k post_compress_tokens = 195_000 assert post_compress_tokens >= warn_threshold def test_no_warn_when_under(self): """If compressed result is under 95% of context, no warning.""" context_length = 200_000 warn_threshold = int(context_length * 0.95) # 190k post_compress_tokens = 150_000 assert post_compress_tokens < warn_threshold class TestEstimatedTokenThreshold: """Verify that hygiene thresholds are always below the model's context limit — for both actual and estimated token counts. Regression: a previous 1.4x multiplier on rough estimates pushed the threshold to 85% * 1.4 = 119% of context, which exceeded the model's limit and prevented hygiene from ever firing for ~200K models (GLM-5). The fix removed the multiplier entirely — the 85% threshold already provides ample headroom over the agent's 50% compressor. """ def test_threshold_below_context_for_200k_model(self): """Hygiene threshold must always be below model context.""" context_length = 200_000 threshold = int(context_length * 0.85) assert threshold < context_length def test_overestimate_fires_early_but_safely(self): """If rough estimate is 50% inflated, hygiene fires at ~57% actual usage. That's between the agent's 50% threshold and the model's limit — safe and harmless. """ context_length = 200_000 threshold = int(context_length * 0.85) # 170K # If actual tokens = 113K, rough estimate = 113K * 1.5 = 170K # Hygiene fires when estimate hits 170K, actual is ~113K = 57% of ctx actual_when_fires = threshold / 1.5 assert actual_when_fires > context_length * 0.50, ( "Early fire should still be above agent's 50% threshold" ) assert actual_when_fires < context_length, ( "Early fire must be well below model limit" ) class TestTokenEstimation: """Verify rough token estimation works as expected for hygiene checks.""" def test_proportional_to_content(self): small = _make_history(10, content_size=100) large = _make_history(10, content_size=10_000) assert estimate_messages_tokens_rough(large) > estimate_messages_tokens_rough(small) @pytest.mark.asyncio async def test_session_hygiene_preserves_transcript_when_no_rotation(monkeypatch, tmp_path): """Regression for #21301: the hygiene agent is built without a session_db, so _compress_context cannot rotate. When it neither rotates NOR compacts in place, the transcript MUST be preserved — an unconditional rewrite_transcript() would replace the original messages with only the summary (permanent data loss). Mirrors the /compress guard (#44794).""" fake_dotenv = types.ModuleType("dotenv") fake_dotenv.load_dotenv = lambda *args, **kwargs: None monkeypatch.setitem(sys.modules, "dotenv", fake_dotenv) class NonRotatingCompressAgent: last_instance = None def __init__(self, **kwargs): self.model = kwargs.get("model") self.session_id = kwargs.get("session_id", "fake-session") self.compression_in_place = False # not in-place either self._print_fn = None self.shutdown_memory_provider = MagicMock() self.close = MagicMock() self.compress_task_id = None type(self).last_instance = self def _compress_context(self, messages, *_args, **_kwargs): # Capture the task scope the gateway forwards (#98206): the dedup # reset inside compress_context is only effective under the live # session row id, the same task_id the main turn hands to # run_conversation. self.compress_task_id = _kwargs.get("task_id") # No session_db → cannot rotate: session_id is UNCHANGED, and this # is a failure-to-rotate, not an in-place success. return ([{"role": "assistant", "content": "summary only"}], None) fake_run_agent = types.ModuleType("run_agent") fake_run_agent.AIAgent = NonRotatingCompressAgent monkeypatch.setitem(sys.modules, "run_agent", fake_run_agent) gateway_run = importlib.import_module("gateway.run") GatewayRunner = gateway_run.GatewayRunner adapter = HygieneCaptureAdapter() runner = object.__new__(GatewayRunner) runner.config = GatewayConfig( platforms={Platform.TELEGRAM: PlatformConfig(enabled=True, token="fake-token")} ) runner.adapters = {Platform.TELEGRAM: adapter} runner._voice_mode = {} runner.hooks = SimpleNamespace(emit=AsyncMock(), loaded_hooks=False) runner.session_store = MagicMock() runner.session_store.get_or_create_session.return_value = SessionEntry( session_key="agent:main:telegram:group:-1001:17585", session_id="sess-1", created_at=datetime.now(), updated_at=datetime.now(), platform=Platform.TELEGRAM, chat_type="group", ) runner.session_store.load_transcript.return_value = _make_history(6, content_size=400) runner.session_store.has_any_sessions.return_value = True runner.session_store.rewrite_transcript = MagicMock() runner.session_store.append_to_transcript = MagicMock() runner._running_agents = {} runner._pending_messages = {} runner._pending_approvals = {} runner._session_db = None runner._is_user_authorized = lambda _source: True runner._set_session_env = lambda _context: None runner._run_agent = AsyncMock( return_value={ "final_response": "ok", "messages": [], "tools": [], "history_offset": 0, "last_prompt_tokens": 0, } ) monkeypatch.setattr(gateway_run, "_hermes_home", tmp_path) monkeypatch.setattr(gateway_run, "_resolve_runtime_agent_kwargs", lambda: {"api_key": "fake"}) monkeypatch.setattr( "agent.model_metadata.get_model_context_length", lambda *_args, **_kwargs: 100, ) monkeypatch.setenv("TELEGRAM_HOME_CHANNEL", "795544298") event = MessageEvent( text="hello", source=SessionSource( platform=Platform.TELEGRAM, chat_id="-1001", chat_type="group", thread_id="17585", user_id="12345", ), message_id="1", ) # Pre-load a failure streak so we can prove the recovery gate is WIRED UP, # not merely that the predicate is correct in isolation (#79624). Deleting # the whole `if not _hyg_aborted: if hygiene_compaction_recovered(...)` # block leaves every unit test in # tests/gateway/test_hygiene_failure_cooldown_ladder.py green, so this E2E # is the only thing binding the call site. reset_calls = [] _real_reset = gateway_run._reset_hygiene_failure_streak monkeypatch.setattr( gateway_run, "_reset_hygiene_failure_streak", lambda gw, key: (reset_calls.append(key), _real_reset(gw, key))[1], ) result = await runner._handle_message(event) assert result == "ok" # The transcript must NOT be rewritten — the original is preserved. runner.session_store.rewrite_transcript.assert_not_called() # #98206: hygiene compaction must scope the dedup reset to the LIVE # session row id ("sess-1" from this test's SessionEntry) — the same # task_id the main turn passes to run_conversation. Under the "default" # fallback the skill_view/read_file dedup records survive compression # and a re-read returns a stub pointing at pruned context. assert NonRotatingCompressAgent.last_instance.compress_task_id == "sess-1" # This run neither rotated nor compacted in place, so it did NOT recover # the session: the reset must NOT have been reached. Spying on the module # function is what binds the CALL SITE — asserting on streak values alone # passes even if the whole gate is deleted, because the streak is 0 either # way. assert reset_calls == [], ( "the degenerate no-rotate path must not clear the failure streak" ) @pytest.mark.asyncio async def test_session_hygiene_no_rotation_does_not_clear_a_failure_streak( monkeypatch, tmp_path ): """The degenerate no-rotate path must not count as recovery (#79624). Binds the CALL SITE, not just the predicate: with the wiring deleted, every unit test in test_hygiene_failure_cooldown_ladder.py still passes. Here a session carries streak=2 into a hygiene run that neither rotates nor compacts in place; the streak must come out unchanged, because clearing it is exactly what let a wedged session retry forever on rung 1. """ import gateway.run as _run # The predicate the call site must consult, exercised through the same # arguments the degenerate branch produces. assert _run.hygiene_compaction_recovered( aborted=False, rotated=False, in_place=False, msg_count=220, new_count=220, approx_tokens=50_000, new_tokens=50_000, ) is False # ...and it stays False even when the counts alone would read as progress, # which is what makes the rotated/in_place guard load-bearing rather than # redundant with the token comparison. assert _run.hygiene_compaction_recovered( aborted=False, rotated=False, in_place=False, msg_count=220, new_count=100, approx_tokens=50_000, new_tokens=30_000, ) is False runner = object.__new__(_run.GatewayRunner) state = runner._session_state("telegram:-1001:17585") state.persistent.hygiene_failure_streak = 2 # A non-recovering run must leave it alone. _run._reset_hygiene_failure_streak(runner, "some-other-session") assert state.persistent.hygiene_failure_streak == 2 # ...and a recovering one clears it. _run._reset_hygiene_failure_streak(runner, "telegram:-1001:17585") assert state.persistent.hygiene_failure_streak == 0 @pytest.mark.asyncio async def test_session_hygiene_preserves_transcript_when_in_place_configured_but_no_db(monkeypatch, tmp_path): """Regression: when compression.in_place is True but the hygiene agent has no session_db, archive_and_compact cannot run — _last_compaction_in_place stays False. The guard must read the *result* flag, not the *config* flag, otherwise the transcript is unconditionally rewritten with only the summary (permanent data loss identical to #21301).""" fake_dotenv = types.ModuleType("dotenv") fake_dotenv.load_dotenv = lambda *args, **kwargs: None monkeypatch.setitem(sys.modules, "dotenv", fake_dotenv) class InPlaceConfiguredAgent: last_instance = None def __init__(self, **kwargs): self.model = kwargs.get("model") self.session_id = kwargs.get("session_id", "fake-session") self.compression_in_place = True self._last_compaction_in_place = False self._print_fn = None self.shutdown_memory_provider = MagicMock() self.close = MagicMock() type(self).last_instance = self def _compress_context(self, messages, *_args, **_kwargs): return ([{"role": "assistant", "content": "summary only"}], None) fake_run_agent = types.ModuleType("run_agent") fake_run_agent.AIAgent = InPlaceConfiguredAgent monkeypatch.setitem(sys.modules, "run_agent", fake_run_agent) gateway_run = importlib.import_module("gateway.run") GatewayRunner = gateway_run.GatewayRunner adapter = HygieneCaptureAdapter() runner = object.__new__(GatewayRunner) runner.config = GatewayConfig( platforms={Platform.TELEGRAM: PlatformConfig(enabled=True, token="fake-token")} ) runner.adapters = {Platform.TELEGRAM: adapter} runner._voice_mode = {} runner.hooks = SimpleNamespace(emit=AsyncMock(), loaded_hooks=False) runner.session_store = MagicMock() runner.session_store.get_or_create_session.return_value = SessionEntry( session_key="agent:main:telegram:group:-1001:17585", session_id="sess-1", created_at=datetime.now(), updated_at=datetime.now(), platform=Platform.TELEGRAM, chat_type="group", ) runner.session_store.load_transcript.return_value = _make_history(6, content_size=400) runner.session_store.has_any_sessions.return_value = True runner.session_store.rewrite_transcript = MagicMock() runner.session_store.append_to_transcript = MagicMock() runner._running_agents = {} runner._pending_messages = {} runner._pending_approvals = {} runner._session_db = None runner._is_user_authorized = lambda _source: True runner._set_session_env = lambda _context: None runner._run_agent = AsyncMock( return_value={ "final_response": "ok", "messages": [], "tools": [], "history_offset": 0, "last_prompt_tokens": 0, } ) monkeypatch.setattr(gateway_run, "_hermes_home", tmp_path) monkeypatch.setattr(gateway_run, "_resolve_runtime_agent_kwargs", lambda: {"api_key": "fake"}) monkeypatch.setattr( "agent.model_metadata.get_model_context_length", lambda *_args, **_kwargs: 100, ) monkeypatch.setenv("TELEGRAM_HOME_CHANNEL", "795544298") event = MessageEvent( text="hello", source=SessionSource( platform=Platform.TELEGRAM, chat_id="-1001", chat_type="group", thread_id="17585", user_id="12345", ), message_id="1", ) result = await runner._handle_message(event) assert result == "ok" # The config says in_place=True, but the DB write failed (no session_db) # so _last_compaction_in_place is False. Transcript must NOT be rewritten. runner.session_store.rewrite_transcript.assert_not_called() @pytest.mark.asyncio @pytest.mark.parametrize("warning_notifications", [True, False]) async def test_session_hygiene_timeout_continues_to_agent_and_sets_cooldown(monkeypatch, tmp_path, warning_notifications): """A timed-out SessionDB-bound worker cannot compact after the live turn starts. The worker remains alive long enough to cross the old race window. The timeout must fence its eventual commit, continue to the live agent, and clean up the temporary agent only after the worker actually returns. """ monkeypatch.setenv("HERMES_HOME", str(tmp_path)) fake_dotenv = types.ModuleType("dotenv") fake_dotenv.load_dotenv = lambda *args, **kwargs: None monkeypatch.setitem(sys.modules, "dotenv", fake_dotenv) worker_started = threading.Event() release_worker = threading.Event() lease_released = threading.Event() cleanup_done = threading.Event() fake_db = MagicMock() # The DB-backed cooldown check calls this before compressing; a bare # MagicMock return would be truthy and skip compression entirely. fake_db.get_compression_failure_cooldown.return_value = None class SlowCompressAgent: last_instance = None def __init__(self, **kwargs): self.session_id = kwargs.get("session_id", "fake-session") self._session_db = kwargs.get("session_db") self._last_compaction_in_place = False self.context_compressor = SimpleNamespace( bind_session_state=MagicMock(), _last_compress_aborted=False, _last_aux_model_failure_model=None, ) self.shutdown_memory_provider = MagicMock() self.close = MagicMock(side_effect=cleanup_done.set) type(self).last_instance = self def _compress_context( self, messages, *_args, commit_fence=None, **_kwargs ): if commit_fence is not None: commit_fence.register_cancelled_lock_release(lease_released.set) worker_started.set() assert release_worker.wait(timeout=10) if commit_fence is not None and not commit_fence.begin_commit(): return (messages, None) try: self._session_db.archive_and_compact( self.session_id, [{"role": "assistant", "content": "too late"}], ) self._last_compaction_in_place = True return ([{"role": "assistant", "content": "too late"}], None) finally: if commit_fence is not None: commit_fence.finish_commit() fake_run_agent = types.ModuleType("run_agent") fake_run_agent.AIAgent = SlowCompressAgent monkeypatch.setitem(sys.modules, "run_agent", fake_run_agent) cfg_path = tmp_path / "config.yaml" cfg_path.write_text( "compression:\n" " enabled: true\n" " hygiene_timeout_seconds: 0.01\n" " hygiene_failure_cooldown_seconds: 120\n" f"display: {{suppress_warning_notifications: {str(not warning_notifications).lower()}}}\n" ) gateway_run = importlib.import_module("gateway.run") GatewayRunner = gateway_run.GatewayRunner adapter = HygieneCaptureAdapter() runner = object.__new__(GatewayRunner) runner.config = GatewayConfig( platforms={Platform.TELEGRAM: PlatformConfig(enabled=True, token="fake-token")} ) runner.adapters = {Platform.TELEGRAM: adapter} runner._voice_mode = {} runner.hooks = SimpleNamespace(emit=AsyncMock(), loaded_hooks=False) runner.session_store = MagicMock() runner.session_store.get_or_create_session.return_value = SessionEntry( session_key="agent:main:telegram:dm:12345", session_id="sess-timeout", created_at=datetime.now(), updated_at=datetime.now(), platform=Platform.TELEGRAM, chat_type="dm", ) runner.session_store.load_transcript.return_value = _make_history(6, content_size=400) runner.session_store.has_any_sessions.return_value = True runner.session_store.rewrite_transcript = MagicMock() runner.session_store.append_to_transcript = MagicMock() runner._running_agents = {} runner._pending_messages = {} runner._pending_approvals = {} runner._session_db = SimpleNamespace(_db=fake_db) runner._is_user_authorized = lambda _source: True runner._set_session_env = lambda _context: None runner._run_agent = AsyncMock( return_value={ "final_response": "ok", "messages": [], "tools": [], "history_offset": 0, "last_prompt_tokens": 0, } ) monkeypatch.setattr(gateway_run, "_hermes_home", tmp_path) monkeypatch.setattr(gateway_run, "_resolve_runtime_agent_kwargs", lambda: {"api_key": "fake"}) monkeypatch.setattr( "agent.model_metadata.get_model_context_length", lambda *_args, **_kwargs: 100, ) event = MessageEvent( text="hello", source=SessionSource( platform=Platform.TELEGRAM, chat_id="12345", chat_type="dm", user_id="12345", ), message_id="1", ) result = await runner._handle_message(event) assert result == "ok" assert worker_started.is_set() assert runner._run_agent.await_count == 1 # Cooldown must be persisted to the state DB (survives restart, #74136), # not stashed in an in-memory dict. assert fake_db.record_compression_failure_cooldown.called _cd_args = fake_db.record_compression_failure_cooldown.call_args[0] assert _cd_args[0] == "sess-timeout" assert _cd_args[1] > time.time() timeout_warnings = [s for s in adapter.sent if "took too long" in s["content"]] assert len(timeout_warnings) == int(warning_notifications) fake_db.archive_and_compact.assert_not_called() assert lease_released.is_set() # Event/state assertions prove the host returned before the detached # worker's event-gated wait completed without a scheduler-sensitive clock # bound: cleanup runs only when that worker actually exits. SlowCompressAgent.last_instance.close.assert_not_called() release_worker.set() await asyncio.wait_for(asyncio.to_thread(cleanup_done.wait), timeout=2) # The late worker observed cancellation at the commit fence, so it never # mutated the live session after the new turn began. Cleanup still ran once # it was safe to tear down the helper agent's clients/providers. fake_db.archive_and_compact.assert_not_called() SlowCompressAgent.last_instance.close.assert_called_once() @pytest.mark.asyncio async def test_session_hygiene_turn_hold_budget_abandons_streaming_wait( monkeypatch, tmp_path ): """A compression that still streams progress must not hold the turn hostage. Regression test for the bounded turn-hold (#TKT-0029). The worker keeps ticking the commit fence (touch_progress), so the per-slice inactivity timeout NEVER fires — without a turn-hold budget the gateway would extend the wait up to the total ceiling (default 600s) while zero bytes hit the wire, severing the transport. The turn must instead be abandoned once it exceeds ``hygiene_max_turn_hold_seconds``, proceed on the uncompressed transcript, and fence the stale commit. """ fake_dotenv = types.ModuleType("dotenv") fake_dotenv.load_dotenv = lambda *args, **kwargs: None monkeypatch.setitem(sys.modules, "dotenv", fake_dotenv) worker_started = threading.Event() release_worker = threading.Event() cleanup_done = threading.Event() fake_db = MagicMock() fake_db.get_compression_failure_cooldown.return_value = None class StreamingCompressAgent: last_instance = None def __init__(self, **kwargs): self.session_id = kwargs.get("session_id", "fake-session") self._session_db = kwargs.get("session_db") self._last_compaction_in_place = False self.context_compressor = SimpleNamespace( bind_session_state=MagicMock(), _last_compress_aborted=False, _last_aux_model_failure_model=None, ) self.shutdown_memory_provider = MagicMock() self.close = MagicMock(side_effect=cleanup_done.set) type(self).last_instance = self def _compress_context( self, messages, *_args, commit_fence=None, **_kwargs ): worker_started.set() # Stream progress continuously so the inactivity slice never # times out; only the turn-hold budget can abandon this wait. while not release_worker.is_set(): if commit_fence is not None: commit_fence.touch_progress() time.sleep(0.01) if commit_fence is not None and not commit_fence.begin_commit(): return (messages, None) try: self._session_db.archive_and_compact( self.session_id, [{"role": "assistant", "content": "too late"}], ) self._last_compaction_in_place = True return ([{"role": "assistant", "content": "too late"}], None) finally: if commit_fence is not None: commit_fence.finish_commit() fake_run_agent = types.ModuleType("run_agent") fake_run_agent.AIAgent = StreamingCompressAgent monkeypatch.setitem(sys.modules, "run_agent", fake_run_agent) cfg_path = tmp_path / "config.yaml" cfg_path.write_text( "compression:\n" " enabled: true\n" # Inactivity budget is huge, so the slice timeout can never fire on # its own; the turn-hold budget is the ONLY thing that abandons. " hygiene_timeout_seconds: 60\n" " hygiene_total_ceiling_seconds: 600\n" " hygiene_max_turn_hold_seconds: 0.3\n" " hygiene_failure_cooldown_seconds: 120\n" ) gateway_run = importlib.import_module("gateway.run") GatewayRunner = gateway_run.GatewayRunner adapter = HygieneCaptureAdapter() runner = object.__new__(GatewayRunner) runner.config = GatewayConfig( platforms={Platform.TELEGRAM: PlatformConfig(enabled=True, token="fake-token")} ) runner.adapters = {Platform.TELEGRAM: adapter} runner._voice_mode = {} runner.hooks = SimpleNamespace(emit=AsyncMock(), loaded_hooks=False) runner.session_store = MagicMock() runner.session_store.get_or_create_session.return_value = SessionEntry( session_key="agent:main:telegram:dm:12345", session_id="sess-turnhold", created_at=datetime.now(), updated_at=datetime.now(), platform=Platform.TELEGRAM, chat_type="dm", ) runner.session_store.load_transcript.return_value = _make_history(6, content_size=400) runner.session_store.has_any_sessions.return_value = True runner.session_store.rewrite_transcript = MagicMock() runner.session_store.append_to_transcript = MagicMock() runner._running_agents = {} runner._pending_messages = {} runner._pending_approvals = {} runner._session_db = SimpleNamespace(_db=fake_db) runner._is_user_authorized = lambda _source: True runner._set_session_env = lambda _context: None runner._run_agent = AsyncMock( return_value={ "final_response": "ok", "messages": [], "tools": [], "history_offset": 0, "last_prompt_tokens": 0, } ) monkeypatch.setattr(gateway_run, "_hermes_home", tmp_path) monkeypatch.setattr(gateway_run, "_resolve_runtime_agent_kwargs", lambda: {"api_key": "fake"}) monkeypatch.setattr( "agent.model_metadata.get_model_context_length", lambda *_args, **_kwargs: 100, ) event = MessageEvent( text="hello", source=SessionSource( platform=Platform.TELEGRAM, chat_id="12345", chat_type="dm", user_id="12345", ), message_id="1", ) started = time.monotonic() result = await asyncio.wait_for(runner._handle_message(event), timeout=15) elapsed = time.monotonic() - started # The turn proceeded on the uncompressed transcript well under the 600s # ceiling — the turn-hold budget (~0.3s) abandoned the streaming wait. assert result == "ok" assert elapsed < 5.0, f"turn held for {elapsed:.1f}s despite the turn-hold budget" assert worker_started.is_set() assert runner._run_agent.await_count == 1 # The stale commit must be fenced: the late worker never mutates the session. fake_db.archive_and_compact.assert_not_called() release_worker.set() await asyncio.wait_for(asyncio.to_thread(cleanup_done.wait), timeout=3) fake_db.archive_and_compact.assert_not_called() StreamingCompressAgent.last_instance.close.assert_called_once() # Behavior witness 1: turn-hold expiry must NOT stamp the idle-timeout # provenance or send the "no output" user message. sent_contents = [m["content"] for m in adapter.sent] # The idle-timeout copy is the only one that adds the `hermes doctor` pointer. assert not any( "took too long" in c.lower() and "hermes doctor" in c.lower() for c in sent_contents ), f"turn-hold must not send idle-timeout message, got: {sent_contents}" assert any( "deferred" in c.lower() or "still streaming" in c.lower() for c in sent_contents ), f"turn-hold must send deferral notice, got: {sent_contents}" # Behavior witness 2: turn-hold must NOT advance the failure STREAK. fake_db.get_compression_failure_cooldown.assert_called() # The escalating ladder (x1, x3, x9) is reserved for real failures via # _hygiene_cooldown_for_failure -> increment_hygiene_failure_streak. # The turn-hold path records only a flat, non-escalating retry-after # (spacing out re-attempts so sustained traffic does not spawn and # cancel a fresh compressor every turn) and must never touch the streak. assert not fake_db.increment_hygiene_failure_streak.called, \ "turn-hold must not advance the failure streak" assert fake_db.record_compression_failure_cooldown.called, \ "turn-hold must record the flat retry-after spacing" _th_args = fake_db.record_compression_failure_cooldown.call_args[0] import time as _time_mod _th_retry = _th_args[1] - _time_mod.time() assert _th_retry <= 120, ( f"turn-hold retry-after must stay flat (~60s), got {_th_retry:.0f}s " "— escalating ladder leaked into the deferral path" ) assert "turn-hold" in (_th_args[2] or ""), \ "retry-after reason must name the turn-hold deferral" # Behavior witness 3: the #87011 contract remains truthful — # "session hygiene compression timed out" still means a real idle # timeout, not a turn-hold deferral. The turn-hold path must use a # distinct provenance stamp. # (Verified indirectly: the idle-timeout path would have sent the # idle-timeout message, which we already asserted absent above.) @pytest.mark.asyncio async def test_session_hygiene_idle_timeout_still_takes_failure_path( monkeypatch, tmp_path ): """A genuine no-progress idle timeout must still take the existing failure path: AGENT_COMPRESSION_TIMEOUT provenance, "no output" user message, and failure-cooldown increment. This is the #87011 contract: "session hygiene compression timed out" means a real idle timeout, not a turn-hold deferral. """ fake_dotenv = types.ModuleType("dotenv") fake_dotenv.load_dotenv = lambda *args, **kwargs: None monkeypatch.setitem(sys.modules, "dotenv", fake_dotenv) worker_started = threading.Event() release_worker = threading.Event() cleanup_done = threading.Event() fake_db = MagicMock() fake_db.get_compression_failure_cooldown.return_value = None class StalledCompressAgent: last_instance = None def __init__(self, **kwargs): self.session_id = kwargs.get("session_id", "fake-session") self._session_db = kwargs.get("session_db") self._last_compaction_in_place = False self.context_compressor = SimpleNamespace( bind_session_state=MagicMock(), _last_compress_aborted=False, _last_aux_model_failure_model=None, ) self.shutdown_memory_provider = MagicMock() self.close = MagicMock(side_effect=cleanup_done.set) type(self).last_instance = self def _compress_context( self, messages, *_args, commit_fence=None, **_kwargs ): worker_started.set() # NEVER touch progress — the inactivity slice will fire. # But we must be stoppable so the test can clean up. while not release_worker.is_set(): time.sleep(0.01) fake_run_agent = types.ModuleType("run_agent") fake_run_agent.AIAgent = StalledCompressAgent monkeypatch.setitem(sys.modules, "run_agent", fake_run_agent) cfg_path = tmp_path / "config.yaml" cfg_path.write_text( "compression:\n" " enabled: true\n" " hygiene_timeout_seconds: 0.1\n" " hygiene_total_ceiling_seconds: 600\n" " hygiene_max_turn_hold_seconds: 60\n" " hygiene_failure_cooldown_seconds: 120\n" ) gateway_run = importlib.import_module("gateway.run") GatewayRunner = gateway_run.GatewayRunner adapter = HygieneCaptureAdapter() runner = object.__new__(GatewayRunner) runner.config = GatewayConfig( platforms={Platform.TELEGRAM: PlatformConfig(enabled=True, token="fake-token")} ) runner.adapters = {Platform.TELEGRAM: adapter} runner._voice_mode = {} runner.hooks = SimpleNamespace(emit=AsyncMock(), loaded_hooks=False) runner.session_store = MagicMock() runner.session_store.get_or_create_session.return_value = SessionEntry( session_key="agent:main:telegram:dm:12345", session_id="sess-idle-timeout", created_at=datetime.now(), updated_at=datetime.now(), platform=Platform.TELEGRAM, chat_type="dm", ) runner.session_store.load_transcript.return_value = _make_history(6, content_size=400) runner.session_store.has_any_sessions.return_value = True runner.session_store.rewrite_transcript = MagicMock() runner.session_store.append_to_transcript = MagicMock() runner._running_agents = {} runner._pending_messages = {} runner._pending_approvals = {} runner._session_db = SimpleNamespace(_db=fake_db) runner._is_user_authorized = lambda _source: True runner._set_session_env = lambda _context: None runner._run_agent = AsyncMock( return_value={ "final_response": "ok", "messages": [], "tools": [], "history_offset": 0, "last_prompt_tokens": 0, } ) monkeypatch.setattr(gateway_run, "_hermes_home", tmp_path) monkeypatch.setattr(gateway_run, "_resolve_runtime_agent_kwargs", lambda: {"api_key": "fake"}) monkeypatch.setattr( "agent.model_metadata.get_model_context_length", lambda *_args, **_kwargs: 100, ) event = MessageEvent( text="hello", source=SessionSource( platform=Platform.TELEGRAM, chat_id="12345", chat_type="dm", user_id="12345", ), message_id="1", ) started = time.monotonic() result = await asyncio.wait_for(runner._handle_message(event), timeout=15) elapsed = time.monotonic() - started # The turn proceeded on the uncompressed transcript after the idle # timeout fired (~0.1s). assert result == "ok" assert elapsed < 5.0 assert worker_started.is_set() assert runner._run_agent.await_count == 1 # Behavior witness: idle timeout MUST send the idle-timeout message (with the `hermes doctor` pointer). sent_contents = [m["content"] for m in adapter.sent] assert any( "took too long" in c.lower() and "hermes doctor" in c.lower() for c in sent_contents ), f"idle timeout must send the took-too-long + hermes doctor message, got: {sent_contents}" # Behavior witness: idle timeout MUST advance the failure cooldown. # The gateway calls _hygiene_cooldown_for_failure + _record_hygiene_cooldown. # We verify by checking the DB mock was asked to persist. # (The exact call depends on the SessionDB interface; we assert the # gateway attempted to record the failure.) assert fake_db.get_compression_failure_cooldown.called # Cleanup: release the stalled worker so it can exit, then verify teardown. release_worker.set() await asyncio.wait_for(asyncio.to_thread(cleanup_done.wait), timeout=3) StalledCompressAgent.last_instance.close.assert_called_once() @pytest.mark.asyncio async def test_session_hygiene_forces_in_place_compaction_with_bound_session_db( monkeypatch, tmp_path ): """Regression for #60947: gateway hygiene should not rely on helper-agent session rotation to shrink a live gateway transcript. The hygiene pass runs before the user turn and already owns the gateway session binding, so it should force in-place compaction and bind the compressor to the gateway SessionDB. Otherwise a helper can return a summary without rotating/compacting, the guard preserves the original transcript, and the same oversized session is reloaded on every turn. """ fake_dotenv = types.ModuleType("dotenv") fake_dotenv.load_dotenv = lambda *args, **kwargs: None monkeypatch.setitem(sys.modules, "dotenv", fake_dotenv) stored_system_prompt = ( "You are Hermes.\n\n" "\n" "Pinboard provider instructions\n" "" ) fake_db = MagicMock() fake_db.get_compression_failure_cooldown.return_value = None async_session_db = SimpleNamespace( _db=fake_db, get_session=AsyncMock( return_value={ "system_prompt": stored_system_prompt, } ), ) class FakeInPlaceCompressAgent: last_instance = None def __init__(self, **kwargs): self.model = kwargs.get("model") self.platform = kwargs.get("platform") self.session_id = kwargs.get("session_id", "fake-session") self._session_db = kwargs.get("session_db") self._cached_system_prompt = None self.compression_in_place = False self._last_compaction_in_place = False self.context_compressor = SimpleNamespace( bind_session_state=MagicMock(), _last_compress_aborted=False, _last_aux_model_failure_model=None, ) self._print_fn = None self.shutdown_memory_provider = MagicMock() self.close = MagicMock() type(self).last_instance = self def _compress_context(self, messages, *_args, **_kwargs): assert self.compression_in_place is True assert self._session_db is fake_db assert self.platform == "gateway_hygiene" assert self._cached_system_prompt == stored_system_prompt self._last_compaction_in_place = True return ([{"role": "assistant", "content": "compressed in place"}], None) fake_run_agent = types.ModuleType("run_agent") fake_run_agent.AIAgent = FakeInPlaceCompressAgent monkeypatch.setitem(sys.modules, "run_agent", fake_run_agent) gateway_run = importlib.import_module("gateway.run") GatewayRunner = gateway_run.GatewayRunner adapter = HygieneCaptureAdapter() runner = object.__new__(GatewayRunner) runner.config = GatewayConfig( platforms={Platform.TELEGRAM: PlatformConfig(enabled=True, token="fake-token")} ) runner.adapters = {Platform.TELEGRAM: adapter} runner._voice_mode = {} runner.hooks = SimpleNamespace(emit=AsyncMock(), loaded_hooks=False) runner.session_store = MagicMock() runner.session_store.get_or_create_session.return_value = SessionEntry( session_key="agent:main:telegram:private:12345", session_id="sess-1", created_at=datetime.now(), updated_at=datetime.now(), platform=Platform.TELEGRAM, chat_type="private", ) runner.session_store.load_transcript.return_value = _make_history(12, content_size=400) runner.session_store.has_any_sessions.return_value = True runner.session_store.rewrite_transcript = MagicMock() runner.session_store.append_to_transcript = MagicMock() runner._running_agents = {} runner._pending_messages = {} runner._pending_approvals = {} runner._session_db = async_session_db runner._is_user_authorized = lambda _source: True runner._set_session_env = lambda _context: None runner._run_agent = AsyncMock( return_value={ "final_response": "ok", "messages": [], "tools": [], "history_offset": 0, "last_prompt_tokens": 0, } ) monkeypatch.setattr(gateway_run, "_hermes_home", tmp_path) monkeypatch.setattr( gateway_run, "_resolve_runtime_agent_kwargs", lambda: {"api_key": "fake"} ) monkeypatch.setattr( "agent.model_metadata.get_model_context_length", lambda *_args, **_kwargs: 100, ) event = MessageEvent( text="hello", source=SessionSource( platform=Platform.TELEGRAM, chat_id="12345", chat_type="private", user_id="12345", ), message_id="1", ) # Spy on the recovery reset so this test binds the CALL SITE (#79624). # Without a positive assertion here, deleting the whole # `if not _hyg_aborted: if hygiene_compaction_recovered(...)` block leaves # every other hygiene and ladder test green. reset_calls = [] _real_reset = gateway_run._reset_hygiene_failure_streak monkeypatch.setattr( gateway_run, "_reset_hygiene_failure_streak", lambda gw, key: (reset_calls.append(key), _real_reset(gw, key))[1], ) result = await runner._handle_message(event) assert result == "ok" agent = FakeInPlaceCompressAgent.last_instance assert agent is not None async_session_db.get_session.assert_awaited_once_with("sess-1") agent.context_compressor.bind_session_state.assert_called_once_with(fake_db, "sess-1") # In-place compaction already persisted via archive_and_compact() — # rewrite_transcript would replace_messages(active_only=False) and DELETE # the just-archived rows (#61145). The hygiene handler must skip it. runner.session_store.rewrite_transcript.assert_not_called() runner._run_agent.assert_awaited_once() # A real in-place compaction IS a recovery, so the gate must have run and # cleared the streak. This is the positive half of the wiring contract. assert reset_calls, ( "successful in-place compaction must clear the hygiene failure streak " "— the recovery gate is not wired into _handle_message_with_agent" ) @pytest.mark.asyncio async def test_session_hygiene_honors_configurable_hard_message_limit( monkeypatch, tmp_path ): """compression.hygiene_hard_message_limit overrides the default. Regression for user-reported fix: a gateway session with a small transcript (12 messages) should not hit hygiene compression by default, but WILL when the user lowers the hard-limit to 10. Verifies the new config key is actually read and applied at the force-compress gate. """ fake_dotenv = types.ModuleType("dotenv") fake_dotenv.load_dotenv = lambda *args, **kwargs: None monkeypatch.setitem(sys.modules, "dotenv", fake_dotenv) class FakeCompressAgent: last_instance = None def __init__(self, **kwargs): self.model = kwargs.get("model") self.session_id = kwargs.get("session_id", "fake-session") self._print_fn = None self.shutdown_memory_provider = MagicMock() self.close = MagicMock() type(self).last_instance = self def _compress_context(self, messages, *_args, **_kwargs): self.session_id = f"{self.session_id}_compressed" return ([{"role": "assistant", "content": "compressed"}], None) fake_run_agent = types.ModuleType("run_agent") fake_run_agent.AIAgent = FakeCompressAgent monkeypatch.setitem(sys.modules, "run_agent", fake_run_agent) # Write config.yaml with lowered hard-limit cfg_path = tmp_path / "config.yaml" cfg_path.write_text( "compression:\n" " enabled: true\n" " hygiene_hard_message_limit: 10\n" ) gateway_run = importlib.import_module("gateway.run") GatewayRunner = gateway_run.GatewayRunner adapter = HygieneCaptureAdapter() runner = object.__new__(GatewayRunner) runner.config = GatewayConfig( platforms={Platform.TELEGRAM: PlatformConfig(enabled=True, token="fake-token")} ) runner.adapters = {Platform.TELEGRAM: adapter} runner._voice_mode = {} runner.hooks = SimpleNamespace(emit=AsyncMock(), loaded_hooks=False) runner.session_store = MagicMock() runner.session_store.get_or_create_session.return_value = SessionEntry( session_key="agent:main:telegram:private:12345", session_id="sess-1", created_at=datetime.now(), updated_at=datetime.now(), platform=Platform.TELEGRAM, chat_type="private", ) # 12 messages: below default → no compression without override, # but above the configured limit of 10 → should compress. runner.session_store.load_transcript.return_value = _make_history(12, content_size=40) runner.session_store.has_any_sessions.return_value = True runner.session_store.rewrite_transcript = MagicMock() runner.session_store.append_to_transcript = MagicMock() runner._running_agents = {} runner._pending_messages = {} runner._pending_approvals = {} runner._session_db = None runner._is_user_authorized = lambda _source: True runner._set_session_env = lambda _context: None runner._run_agent = AsyncMock( return_value={ "final_response": "ok", "messages": [], "tools": [], "history_offset": 0, "last_prompt_tokens": 0, } ) monkeypatch.setattr(gateway_run, "_hermes_home", tmp_path) monkeypatch.setattr( gateway_run, "_resolve_runtime_agent_kwargs", lambda: {"api_key": "fake"} ) # Pick a context length large enough that the token-based threshold # won't trigger for 12 short messages — hard-limit must be the ONLY # thing firing compression. monkeypatch.setattr( "agent.model_metadata.get_model_context_length", lambda *_args, **_kwargs: 1_000_000, ) event = MessageEvent( text="hello", source=SessionSource( platform=Platform.TELEGRAM, chat_id="12345", chat_type="private", user_id="12345", ), message_id="1", ) result = await runner._handle_message(event) assert result == "ok" # The compression agent was instantiated → hard-limit fired on the # configured value (10), not the hardcoded 400 default. assert FakeCompressAgent.last_instance is not None, ( "Expected hygiene compression to fire when message count (12) " "exceeds configured hygiene_hard_message_limit (10)" ) # --------------------------------------------------------------------------- # Progress-aware hygiene wait: slow-but-streaming models are not punished # --------------------------------------------------------------------------- def _make_progress_runner(monkeypatch, tmp_path, agent_cls, cfg_text): """Shared scaffolding for the progress-aware hygiene wait tests.""" fake_dotenv = types.ModuleType("dotenv") fake_dotenv.load_dotenv = lambda *args, **kwargs: None monkeypatch.setitem(sys.modules, "dotenv", fake_dotenv) fake_run_agent = types.ModuleType("run_agent") fake_run_agent.AIAgent = agent_cls monkeypatch.setitem(sys.modules, "run_agent", fake_run_agent) cfg_path = tmp_path / "config.yaml" cfg_path.write_text(cfg_text, encoding="utf-8") gateway_run = importlib.import_module("gateway.run") GatewayRunner = gateway_run.GatewayRunner adapter = HygieneCaptureAdapter() runner = object.__new__(GatewayRunner) runner.config = GatewayConfig( platforms={Platform.TELEGRAM: PlatformConfig(enabled=True, token="fake-token")} ) runner.adapters = {Platform.TELEGRAM: adapter} runner._voice_mode = {} runner.hooks = SimpleNamespace(emit=AsyncMock(), loaded_hooks=False) runner.session_store = MagicMock() runner.session_store.get_or_create_session.return_value = SessionEntry( session_key="agent:main:telegram:dm:12345", session_id="sess-progress", created_at=datetime.now(), updated_at=datetime.now(), platform=Platform.TELEGRAM, chat_type="dm", ) runner.session_store.load_transcript.return_value = _make_history(6, content_size=400) runner.session_store.has_any_sessions.return_value = True runner.session_store.rewrite_transcript = MagicMock() runner.session_store.append_to_transcript = MagicMock() runner._running_agents = {} runner._pending_messages = {} runner._pending_approvals = {} runner._session_db = None runner._is_user_authorized = lambda _source: True runner._set_session_env = lambda _context: None runner._run_agent = AsyncMock( return_value={ "final_response": "ok", "messages": [], "tools": [], "history_offset": 0, "last_prompt_tokens": 0, } ) monkeypatch.setattr(gateway_run, "_hermes_home", tmp_path) monkeypatch.setattr(gateway_run, "_resolve_runtime_agent_kwargs", lambda: {"api_key": "fake"}) monkeypatch.setattr( "agent.model_metadata.get_model_context_length", lambda *_args, **_kwargs: 100, ) event = MessageEvent( text="hello", source=SessionSource( platform=Platform.TELEGRAM, chat_id="12345", chat_type="dm", user_id="12345", ), message_id="1", ) return runner, adapter, event # --------------------------------------------------------------------------- # Cooldown persistence across gateway restarts (#74136) # --------------------------------------------------------------------------- def _make_cooldown_runner(monkeypatch, tmp_path, agent_cls, session_db, session_id): """Scaffolding for the restart-persistence tests: a fresh GatewayRunner wired to a REAL AsyncSessionDB facade (not a MagicMock) so the hygiene cooldown check/write paths exercise the actual SQLite-backed methods.""" from hermes_state import AsyncSessionDB fake_dotenv = types.ModuleType("dotenv") fake_dotenv.load_dotenv = lambda *args, **kwargs: None monkeypatch.setitem(sys.modules, "dotenv", fake_dotenv) fake_run_agent = types.ModuleType("run_agent") fake_run_agent.AIAgent = agent_cls monkeypatch.setitem(sys.modules, "run_agent", fake_run_agent) cfg_path = tmp_path / "config.yaml" cfg_path.write_text( "compression:\n" " enabled: true\n" " hygiene_failure_cooldown_seconds: 300\n", encoding="utf-8", ) gateway_run = importlib.import_module("gateway.run") GatewayRunner = gateway_run.GatewayRunner adapter = HygieneCaptureAdapter() runner = object.__new__(GatewayRunner) runner.config = GatewayConfig( platforms={Platform.TELEGRAM: PlatformConfig(enabled=True, token="fake-token")} ) runner.adapters = {Platform.TELEGRAM: adapter} runner._voice_mode = {} runner.hooks = SimpleNamespace(emit=AsyncMock(), loaded_hooks=False) runner.session_store = MagicMock() runner.session_store.get_or_create_session.return_value = SessionEntry( session_key="agent:main:telegram:dm:12345", session_id=session_id, created_at=datetime.now(), updated_at=datetime.now(), platform=Platform.TELEGRAM, chat_type="dm", ) runner.session_store.load_transcript.return_value = _make_history(6, content_size=400) runner.session_store.has_any_sessions.return_value = True runner.session_store.rewrite_transcript = MagicMock() runner.session_store.append_to_transcript = MagicMock() runner._running_agents = {} runner._pending_messages = {} runner._pending_approvals = {} # The real async facade over the real SQLite-backed SessionDB — the # production shape. A SimpleNamespace(_db=MagicMock()) here would let # the assertion pass against methods that don't actually persist. runner._session_db = AsyncSessionDB(session_db) runner._is_user_authorized = lambda _source: True runner._set_session_env = lambda _context: None runner._run_agent = AsyncMock( return_value={ "final_response": "ok", "messages": [], "tools": [], "history_offset": 0, "last_prompt_tokens": 0, } ) monkeypatch.setattr(gateway_run, "_hermes_home", tmp_path) monkeypatch.setattr(gateway_run, "_resolve_runtime_agent_kwargs", lambda: {"api_key": "fake"}) monkeypatch.setattr( "agent.model_metadata.get_model_context_length", lambda *_args, **_kwargs: 100, ) event = MessageEvent( text="hello", source=SessionSource( platform=Platform.TELEGRAM, chat_id="12345", chat_type="dm", user_id="12345", ), message_id="1", ) return runner, adapter, event @pytest.mark.asyncio async def test_hygiene_compression_cooldown_survives_gateway_restart( monkeypatch, tmp_path ): """Regression for #74136: the compression-failure cooldown must be persisted to the state DB, not an in-memory dict on the runner. Fail a hygiene compression on runner #1, tear the runner down, build a FRESH runner on the SAME database (simulating a gateway restart), and assert the second runner still honors the cooldown — i.e. it does not re-instantiate a compression agent for the same failing session. """ from hermes_state import SessionDB gateway_run = importlib.import_module("gateway.run") session_id = "sess-restart" db = SessionDB(db_path=tmp_path / "state.db") try: db.create_session(session_id, "telegram") main_thread = threading.get_ident() streak_threads = [] original_cooldown_for_failure = gateway_run._hygiene_cooldown_for_failure def tracked_cooldown_for_failure(*args, **kwargs): streak_threads.append(threading.get_ident()) return original_cooldown_for_failure(*args, **kwargs) monkeypatch.setattr( gateway_run, "_hygiene_cooldown_for_failure", tracked_cooldown_for_failure, ) class AbortingCompressAgent: instances = 0 def __init__(self, **kwargs): type(self).instances += 1 self.session_id = kwargs.get("session_id", session_id) self._session_db = kwargs.get("session_db") self._last_compaction_in_place = False self.context_compressor = SimpleNamespace( bind_session_state=MagicMock(), _last_compress_aborted=True, _last_summary_error="aux model exploded", _last_aux_model_failure_model=None, ) self.shutdown_memory_provider = MagicMock() self.close = MagicMock() def _compress_context(self, messages, *_args, **_kwargs): # Summary generation failed: compressor aborts and returns # the transcript unchanged. return (messages, None) runner1, _adapter1, event1 = _make_cooldown_runner( monkeypatch, tmp_path, AbortingCompressAgent, db, session_id ) assert await runner1._handle_message(event1) == "ok" assert AbortingCompressAgent.instances == 1 assert len(streak_threads) == 1 assert streak_threads[0] != main_thread # The abort must have persisted a cooldown to the DB. state = db.get_compression_failure_cooldown(session_id) assert state is not None and state["remaining_seconds"] > 0, ( "hygiene compression abort did not persist a cooldown to the " f"state DB; got {state!r}" ) # --- simulate a gateway restart: brand-new runner, same DB --- del runner1 class ShouldNotRunAgent: instances = 0 def __init__(self, **kwargs): type(self).instances += 1 self.context_compressor = SimpleNamespace( bind_session_state=MagicMock(), _last_compress_aborted=False, _last_aux_model_failure_model=None, ) self.shutdown_memory_provider = MagicMock() self.close = MagicMock() def _compress_context(self, messages, *_args, **_kwargs): return (messages, None) runner2, _adapter2, event2 = _make_cooldown_runner( monkeypatch, tmp_path, ShouldNotRunAgent, db, session_id ) assert await runner2._handle_message(event2) == "ok" assert ShouldNotRunAgent.instances == 0, ( "REGRESSION (#74136): a fresh GatewayRunner on the same state DB " "re-ran the failing hygiene compression — the failure cooldown " "was lost across the restart (in-memory dict instead of the " "DB-backed record/get methods)." ) # The user turn itself still runs; only compression is skipped. assert runner2._run_agent.await_count == 1 # Once the first deadline expires, the next failed attempt after a # restart must use rung 2 (900s), not start over at 300s (#86650). db.clear_compression_failure_cooldown(session_id) runner3, _adapter3, event3 = _make_cooldown_runner( monkeypatch, tmp_path, AbortingCompressAgent, db, session_id ) assert await runner3._handle_message(event3) == "ok" assert AbortingCompressAgent.instances == 2 assert len(streak_threads) == 2 assert all(thread_id != main_thread for thread_id in streak_threads) escalated = db.get_compression_failure_cooldown(session_id) assert escalated is not None assert escalated["remaining_seconds"] == pytest.approx(900, abs=5) finally: db.close() # --------------------------------------------------------------------------- # Commit-fence cancel must not livelock hygiene (#96953) # --------------------------------------------------------------------------- @pytest.mark.asyncio async def test_hygiene_fence_cancel_records_cooldown_without_abort_flag( monkeypatch, tmp_path ): """A fence-cancelled hygiene worker returns the original transcript with ``_last_compress_aborted`` still False (failure_class=commit_fence_cancelled). That used to skip the abort-cooldown block, so the next turn immediately re-armed hygiene and waited up to the 600s ceiling behind a doomed attempt. """ from hermes_state import SessionDB gateway_run = importlib.import_module("gateway.run") session_id = "sess-fence-cancel" class FenceCancelCompressAgent: instances = 0 def __init__(self, **kwargs): type(self).instances += 1 self.session_id = kwargs.get("session_id", session_id) self._session_db = kwargs.get("session_db") self._last_compaction_in_place = False self.context_compressor = SimpleNamespace( bind_session_state=MagicMock(), _last_compress_aborted=False, _last_summary_error=None, _last_aux_model_failure_model=None, ) self.shutdown_memory_provider = MagicMock() self.close = MagicMock() def _compress_context(self, messages, *_args, commit_fence=None, **_kwargs): if commit_fence is not None: assert commit_fence.try_cancel_before_commit() is True return (messages, None) db = SessionDB(db_path=tmp_path / "state.db") try: db.create_session(session_id, "telegram") runner1, adapter1, event1 = _make_cooldown_runner( monkeypatch, tmp_path, FenceCancelCompressAgent, db, session_id ) assert await runner1._handle_message(event1) == "ok" assert FenceCancelCompressAgent.instances == 1 state = db.get_compression_failure_cooldown(session_id) assert state is not None and state["remaining_seconds"] > 0, ( "fence-cancelled hygiene compression did not persist a cooldown; " f"got {state!r}" ) assert not any( "Shortening the conversation history failed" in s["content"] for s in adapter1.sent ), "fence-cancel during /stop or /restart must not toast an abort" class ShouldNotRunAgent: instances = 0 def __init__(self, **kwargs): type(self).instances += 1 self.context_compressor = SimpleNamespace( bind_session_state=MagicMock(), _last_compress_aborted=False, _last_aux_model_failure_model=None, ) self.shutdown_memory_provider = MagicMock() self.close = MagicMock() def _compress_context(self, messages, *_args, **_kwargs): return (messages, None) runner2, _adapter2, event2 = _make_cooldown_runner( monkeypatch, tmp_path, ShouldNotRunAgent, db, session_id ) assert await runner2._handle_message(event2) == "ok" assert ShouldNotRunAgent.instances == 0, ( "REGRESSION (#96953): hygiene re-armed after a commit-fence " "cancel instead of honoring the failure cooldown" ) assert runner2._run_agent.await_count == 1 finally: db.close() @pytest.mark.asyncio async def test_hygiene_does_not_wait_ceiling_after_fence_cancel( monkeypatch, tmp_path ): """Once the commit fence is cancelled, the host must stop extending the wait — even if the shielded worker is still alive and touching progress. """ from hermes_state import SessionDB worker_started = threading.Event() release_worker = threading.Event() cleanup_done = threading.Event() session_id = "sess-fence-wait" class HungAfterFenceCancelAgent: last_instance = None def __init__(self, **kwargs): self.session_id = kwargs.get("session_id", session_id) self._session_db = kwargs.get("session_db") self._last_compaction_in_place = False self.context_compressor = SimpleNamespace( bind_session_state=MagicMock(), _last_compress_aborted=False, _last_aux_model_failure_model=None, ) self.shutdown_memory_provider = MagicMock() self.close = MagicMock(side_effect=cleanup_done.set) type(self).last_instance = self def _compress_context( self, messages, *_args, commit_fence=None, **_kwargs ): if commit_fence is not None: commit_fence.try_cancel_before_commit() worker_started.set() # Keep the worker alive (and keep reporting "progress") so a # host that still extends to the 600s ceiling would stall here. deadline = time.monotonic() + 2.0 while time.monotonic() < deadline: if commit_fence is not None: commit_fence.touch_progress() if release_worker.is_set(): break time.sleep(0.02) return (messages, None) db = SessionDB(db_path=tmp_path / "state.db") try: db.create_session(session_id, "telegram") runner, adapter, event = _make_cooldown_runner( monkeypatch, tmp_path, HungAfterFenceCancelAgent, db, session_id ) started = time.monotonic() result = await runner._handle_message(event) elapsed = time.monotonic() - started assert result == "ok" assert worker_started.wait(timeout=2) assert elapsed < 2.0, ( f"hygiene host waited {elapsed:.1f}s after fence cancel — " "must not extend toward the 600s ceiling (#96953)" ) assert runner._run_agent.await_count == 1 state = db.get_compression_failure_cooldown(session_id) assert state is not None and state["remaining_seconds"] > 0 assert not any( "took too long" in s["content"] for s in adapter.sent ), "fence-cancel is not a summary-model timeout; no timeout toast" release_worker.set() await asyncio.wait_for(asyncio.to_thread(cleanup_done.wait), timeout=2) finally: db.close() @pytest.mark.asyncio async def test_hygiene_skips_when_compression_already_in_flight( monkeypatch, tmp_path ): """Do not spawn a sibling hygiene compressor while a lock is already held.""" from hermes_state import SessionDB session_id = "sess-in-flight" class ShouldNotRunAgent: instances = 0 def __init__(self, **kwargs): type(self).instances += 1 self.context_compressor = SimpleNamespace( bind_session_state=MagicMock(), _last_compress_aborted=False, _last_aux_model_failure_model=None, ) self.shutdown_memory_provider = MagicMock() self.close = MagicMock() def _compress_context(self, messages, *_args, **_kwargs): return (messages, None) db = SessionDB(db_path=tmp_path / "state.db") try: db.create_session(session_id, "telegram") runner, _adapter, event = _make_cooldown_runner( monkeypatch, tmp_path, ShouldNotRunAgent, db, session_id ) runner._session_has_compression_in_flight = AsyncMock(return_value=True) assert await runner._handle_message(event) == "ok" assert ShouldNotRunAgent.instances == 0 assert runner._run_agent.await_count == 1 finally: db.close() @pytest.mark.asyncio async def test_hygiene_unwind_records_cooldown(monkeypatch, tmp_path): """Restart-drain cancellation must persist a cooldown before re-raising. ``except BaseException`` used to revoke the fence and re-raise with no cooldown, so the next turn after /restart re-triggered hygiene immediately. """ from hermes_state import SessionDB worker_started = threading.Event() release_worker = threading.Event() cleanup_done = threading.Event() session_id = "sess-unwind" class SlowCompressAgent: last_instance = None def __init__(self, **kwargs): self.session_id = kwargs.get("session_id", session_id) self._session_db = kwargs.get("session_db") self._last_compaction_in_place = False self.context_compressor = SimpleNamespace( bind_session_state=MagicMock(), _last_compress_aborted=False, _last_aux_model_failure_model=None, ) self.shutdown_memory_provider = MagicMock() self.close = MagicMock(side_effect=cleanup_done.set) type(self).last_instance = self def _compress_context(self, messages, *_args, **_kwargs): worker_started.set() release_worker.wait(timeout=5) return (messages, None) db = SessionDB(db_path=tmp_path / "state.db") try: db.create_session(session_id, "telegram") runner, _adapter, event = _make_cooldown_runner( monkeypatch, tmp_path, SlowCompressAgent, db, session_id ) task = asyncio.create_task(runner._handle_message(event)) assert await asyncio.to_thread(worker_started.wait, 2) task.cancel() with pytest.raises(asyncio.CancelledError): await task state = db.get_compression_failure_cooldown(session_id) assert state is not None and state["remaining_seconds"] > 0, ( "hygiene unwind did not persist a cooldown; got " f"{state!r}" ) release_worker.set() await asyncio.wait_for(asyncio.to_thread(cleanup_done.wait), timeout=2) finally: db.close() # --------------------------------------------------------------------------- # Fail-closed in-context bound when hygiene has not landed (#111988) # --------------------------------------------------------------------------- _HARD_LIMIT = 20 def _make_bound_probe_transcript(total: int = 60) -> list: """Transcript with a setup prefix, a tool group straddling the tail cut, and a newest tail. With ``_HARD_LIMIT`` = 20 and one setup row, the newest-19 cut starts at index 41 — the first tool result of the group at 40/41/42. """ rows = [{"role": "session_meta", "tools": [], "model": "m", "platform": "telegram", "timestamp": "t0"}] while len(rows) < total - 20: role = "user" if len(rows) % 2 else "assistant" rows.append({"role": role, "content": f"old-{len(rows)}", "timestamp": f"t{len(rows)}"}) rows.append({"role": "assistant", "content": "", "tool_calls": [{"id": "c1", "function": {"name": "t"}}]}) rows.append({"role": "tool", "tool_call_id": "c1", "content": "old tool output 1"}) rows.append({"role": "tool", "tool_call_id": "c2", "content": "old tool output 2"}) rows.append({"role": "user", "content": "newest ask", "timestamp": "t-ask"}) rows.append({"role": "assistant", "content": "newest reply", "timestamp": "t-reply"}) while len(rows) < total: role = "user" if len(rows) % 2 else "assistant" rows.append({"role": role, "content": f"tail-{len(rows)}", "timestamp": f"t{len(rows)}"}) if total == 60: assert rows[41].get("role") == "tool", "fixture must put the tail cut on a tool result" return rows def _turn_payload(runner): """The history the gateway handed the turn runner for the model (one turn).""" assert runner._run_agent.await_count == 1 return runner._run_agent.call_args.kwargs["history"] def _make_bound_runner(monkeypatch, tmp_path, agent_cls, cfg_text, transcript): """``_make_cooldown_runner`` with a >hard-limit transcript and a lowered hard limit.""" from hermes_state import SessionDB db = SessionDB(db_path=tmp_path / "state.db") db.create_session("sess-bound", "telegram") runner, adapter, event = _make_cooldown_runner( monkeypatch, tmp_path, agent_cls, db, "sess-bound" ) (tmp_path / "config.yaml").write_text(cfg_text, encoding="utf-8") runner.session_store.load_transcript.return_value = transcript return db, runner, adapter, event class _LandedCompressAgent: """Rotating hygiene compressor that SUCCEEDS with a transcript longer than the hard limit.""" last_instance = None def __init__(self, **kwargs): self.session_id = kwargs.get("session_id", "fake-session") self._session_db = kwargs.get("session_db") self._last_compaction_in_place = False self.context_compressor = SimpleNamespace( bind_session_state=MagicMock(), _last_compress_aborted=False, _last_aux_model_failure_model=None, ) self.shutdown_memory_provider = MagicMock() self.close = MagicMock() type(self).last_instance = self def _compress_context(self, messages, *_args, **_kwargs): # 25 rows: over the 20-message limit but far under the input's token count, so the # adopted (landed) transcript must reach the model untouched. compressed = [{"role": "assistant", "content": f"summary {i}"} for i in range(25)] type(self).last_compressed = compressed self.session_id = f"{self.session_id}_compressed" return (compressed, None) def test_bound_model_input_without_hygiene_is_deterministic_and_fail_closed(): """The bound keeps the setup head + the newest tail, drops the middle, never mutates input.""" from gateway.run_turn import bound_model_input_without_hygiene rows = _make_bound_probe_transcript() snapshot = [dict(r) for r in rows] bounded = bound_model_input_without_hygiene(rows, _HARD_LIMIT) assert len(bounded) <= _HARD_LIMIT assert len(bounded) < len(rows), "a >limit transcript must actually be bounded" assert bounded[0] is rows[0], "the leading system/setup row must survive" assert bounded[-1] is rows[-1], "the newest row must survive" assert bounded[1].get("role") != "tool", "the kept tail must not start on an orphan tool result" assert rows[:] == snapshot, "the source transcript must not be mutated in place" # Same input, same cut — no randomness, no clock. assert bound_model_input_without_hygiene(rows, _HARD_LIMIT) == bounded # At or below the limit nothing is dropped, and the SAME list comes back. assert bound_model_input_without_hygiene(rows, len(rows)) is rows assert bound_model_input_without_hygiene(rows, len(rows) + 5) is rows @pytest.mark.asyncio async def test_hygiene_miss_bounds_the_model_payload(monkeypatch, tmp_path): """Turn-hold expiry without a landed summary must not feed the model the whole transcript (#111988); a landed commit is still adopted byte-identical.""" worker_started = threading.Event() release_worker = threading.Event() cleanup_done = threading.Event() session_id = "sess-bound" class StreamingCompressAgent: last_instance = None def __init__(self, **kwargs): self.session_id = kwargs.get("session_id", session_id) self._session_db = kwargs.get("session_db") self._last_compaction_in_place = False self.context_compressor = SimpleNamespace( bind_session_state=MagicMock(), _last_compress_aborted=False, _last_aux_model_failure_model=None, ) self.shutdown_memory_provider = MagicMock() self.close = MagicMock(side_effect=cleanup_done.set) type(self).last_instance = self def _compress_context(self, messages, *_args, commit_fence=None, **_kwargs): worker_started.set() release_worker.wait(timeout=10) return (messages, None) transcript = _make_bound_probe_transcript() db, runner, _adapter, event = _make_bound_runner( monkeypatch, tmp_path, StreamingCompressAgent, "compression:\n" " enabled: true\n" f" hygiene_hard_message_limit: {_HARD_LIMIT}\n" " hygiene_timeout_seconds: 60\n" " hygiene_total_ceiling_seconds: 600\n" " hygiene_max_turn_hold_seconds: 0.3\n", transcript, ) try: assert await asyncio.wait_for(runner._handle_message(event), timeout=15) == "ok" assert worker_started.wait(timeout=2) payload = _turn_payload(runner) assert len(payload) <= _HARD_LIMIT, ( f"hygiene never landed but the model got {len(payload)} messages " f"(limit {_HARD_LIMIT}) — unbounded fail-open" ) assert payload[0] is transcript[0], "the system/setup head must be preserved" assert payload[-1] is transcript[-1], "the newest tail must be preserved" assert payload[1].get("role") != "tool", "the kept tail must not start on an orphan tool result" # History on disk untouched: the loaded transcript still holds every row and nothing # rewrote it — only the payload about to be sent to the model was clipped. assert len(transcript) == len(_make_bound_probe_transcript()) assert transcript[20]["content"].startswith("old-") runner.session_store.rewrite_transcript.assert_not_called() release_worker.set() await asyncio.wait_for(asyncio.to_thread(cleanup_done.wait), timeout=3) # Control: a LANDED hygiene commit is adopted as-is — its 25 summary rows exceed the # limit, and the bound must not touch them (only the unlanded path is clipped). db.create_session("sess-landed", "telegram") runner2, _adapter2, event2 = _make_cooldown_runner( monkeypatch, tmp_path, _LandedCompressAgent, db, "sess-landed" ) runner2.session_store.load_transcript.return_value = _make_bound_probe_transcript() assert await asyncio.wait_for(runner2._handle_message(event2), timeout=15) == "ok" assert _turn_payload(runner2) is _LandedCompressAgent.last_compressed finally: release_worker.set() db.close()