gateway: debounce queued text follow-ups

This commit is contained in:
Paulo Nascimento
2026-05-23 21:32:19 -04:00
committed by Teknium
parent 21db250034
commit 7abd62719b
10 changed files with 607 additions and 89 deletions

View File

@@ -1,20 +1,10 @@
"""Regression test for #4469.
"""Regression tests for active-session TEXT follow-up queueing.
When the agent is actively running (session present in
``adapter._active_sessions``) and the user fires off multiple TEXT
follow-ups in rapid succession, the previous behaviour was a single-slot
replacement at ``gateway/platforms/base.py``:
self._pending_messages[session_key] = event
So three rapid messages ``A``, ``B``, ``C`` arriving while the agent was
still working on the initial turn produced a pending slot containing only
``C``; ``A`` and ``B`` were silently dropped.
The fix routes the follow-up through ``merge_pending_message_event(...,
merge_text=True)`` so TEXT events accumulate into the existing pending
event's text instead of clobbering it. Photo / media bursts continue to
merge through the same helper (they always did).
When the agent is actively running, rapid text follow-ups should survive as
one next-turn pending message instead of clobbering each other. In
``busy_text_mode=queue`` those active follow-ups first pass through a short
debounce so bursty multi-message thoughts are merged before the active drain
hands off the next turn.
"""
from __future__ import annotations
@@ -22,7 +12,7 @@ from __future__ import annotations
import asyncio
import sys
import types
from unittest.mock import AsyncMock, MagicMock
from unittest.mock import AsyncMock, MagicMock, patch
import pytest
@@ -44,16 +34,27 @@ from gateway.platforms.base import (
BasePlatformAdapter,
MessageEvent,
MessageType,
SendResult,
)
from gateway.session import SessionSource, build_session_key
def _make_event(text: str, chat_id: str = "12345") -> MessageEvent:
def _make_event(
text: str,
chat_id: str = "12345",
*,
chat_type: str = "dm",
user_id: str = "u1",
user_name: str | None = None,
thread_id: str | None = None,
) -> MessageEvent:
source = SessionSource(
platform=Platform.TELEGRAM,
chat_id=chat_id,
chat_type="dm",
user_id="u1",
chat_type=chat_type,
user_id=user_id,
user_name=user_name,
thread_id=thread_id,
)
return MessageEvent(
text=text,
@@ -63,27 +64,26 @@ def _make_event(text: str, chat_id: str = "12345") -> MessageEvent:
)
class _DummyAdapter(BasePlatformAdapter): # type: ignore[misc]
async def connect(self):
pass
async def disconnect(self):
pass
async def get_chat_info(self, chat_id):
return None
async def send(self, *args, **kwargs):
return SendResult(success=True, message_id="x")
def _make_initialized_adapter() -> BasePlatformAdapter:
return _DummyAdapter(PlatformConfig(enabled=True, token="***"), Platform.TELEGRAM)
def _make_adapter() -> BasePlatformAdapter:
"""Build a BasePlatformAdapter without running its heavy __init__.
We only need the bits ``handle_message`` touches on the active-session
path: ``_active_sessions``, ``_pending_messages``,
``_message_handler``, ``_busy_session_handler``, ``config``, ``platform``.
"""
class _DummyAdapter(BasePlatformAdapter): # type: ignore[misc]
async def connect(self):
pass
async def disconnect(self):
pass
async def get_chat_info(self, chat_id):
return None
async def send(self, *args, **kwargs):
return MagicMock(success=True, message_id="x", retryable=False)
"""Build a BasePlatformAdapter without running its heavy __init__."""
adapter = object.__new__(_DummyAdapter)
adapter.config = PlatformConfig(enabled=True, token="***")
adapter.platform = Platform.TELEGRAM
@@ -100,6 +100,10 @@ def _make_adapter() -> BasePlatformAdapter:
adapter._fatal_error_retryable = True
adapter._fatal_error_handler = None
adapter._running = True
adapter._busy_text_mode = "queue"
adapter._busy_text_debounce_seconds = 0.1
adapter._busy_text_hard_cap_seconds = 1.0
adapter._text_debounce = {}
adapter._auto_tts_default = False
adapter._auto_tts_enabled_chats = set()
adapter._auto_tts_disabled_chats = set()
@@ -107,39 +111,235 @@ def _make_adapter() -> BasePlatformAdapter:
return adapter
def _debounced_event(adapter: BasePlatformAdapter, session_key: str) -> MessageEvent:
return adapter._text_debounce[session_key].event
@pytest.mark.asyncio
async def test_rapid_text_followups_accumulate_instead_of_replacing():
"""Three rapid TEXT follow-ups during an active session must all
survive in ``adapter._pending_messages[session_key].text``."""
"""Rapid TEXT follow-ups must all survive in the pending event."""
adapter = _make_adapter()
adapter._busy_text_mode = "" # direct-merge behavior, no debounce
first = _make_event("part one")
session_key = build_session_key(first.source)
# Mark the session as active so subsequent messages take the
# "already running" branch in handle_message.
adapter._active_sessions[session_key] = asyncio.Event()
second = _make_event("part two")
third = _make_event("part three")
await adapter.handle_message(_make_event("part two"))
await adapter.handle_message(_make_event("part three"))
await adapter.handle_message(second)
await adapter.handle_message(third)
# Both rapid follow-ups must be preserved, not just the last one.
pending = adapter._pending_messages[session_key]
assert pending.text == "part two\npart three", (
f"expected accumulated text, got {pending.text!r}"
assert pending.text == "part two\npart three"
assert not adapter._active_sessions[session_key].is_set()
@pytest.mark.asyncio
async def test_debounce_buffers_rapid_text_then_flushes_to_pending():
adapter = _make_adapter()
adapter._busy_text_debounce_seconds = 0.05
first = _make_event("part one")
session_key = build_session_key(first.source)
adapter._active_sessions[session_key] = asyncio.Event()
await adapter.handle_message(_make_event("part two"))
assert session_key in adapter._text_debounce
assert _debounced_event(adapter, session_key).text == "part two"
assert session_key not in adapter._pending_messages
await adapter.handle_message(_make_event("part three"))
assert _debounced_event(adapter, session_key).text == "part two\npart three"
await asyncio.sleep(0.15)
assert session_key not in adapter._text_debounce
assert adapter._pending_messages[session_key].text == "part two\npart three"
@pytest.mark.asyncio
async def test_debounce_resets_timer_on_new_arrival():
adapter = _make_adapter()
adapter._busy_text_debounce_seconds = 0.1
first = _make_event("one")
session_key = build_session_key(first.source)
adapter._active_sessions[session_key] = asyncio.Event()
await adapter.handle_message(first)
task1 = adapter._text_debounce[session_key].task
assert task1 is not None
assert not task1.done()
await adapter.handle_message(_make_event("two"))
task2 = adapter._text_debounce[session_key].task
assert task2 is not None
assert task2 is not task1
await asyncio.sleep(0)
assert task1.cancelled() or task1.done()
assert adapter._text_debounce[session_key].task is task2
await adapter.handle_message(_make_event("three"))
task3 = adapter._text_debounce[session_key].task
assert task3 is not None
assert task3 is not task2
await asyncio.sleep(0.2)
assert session_key not in adapter._text_debounce
assert adapter._pending_messages[session_key].text == "one\ntwo\nthree"
@pytest.mark.asyncio
async def test_active_drain_force_flushes_debounce_before_release():
adapter = _make_adapter()
adapter._busy_text_debounce_seconds = 1.0
processed: list[str] = []
async def _handler(event):
processed.append(event.text)
if event.text == "current":
await adapter.handle_message(_make_event("follow up"))
return None
adapter._message_handler = _handler
current = _make_event("current")
session_key = build_session_key(current.source)
task = asyncio.create_task(adapter._process_message_background(current, session_key))
adapter._session_tasks[session_key] = task
await asyncio.wait_for(task, timeout=1.0)
for _ in range(20):
if processed == ["current", "follow up"] and session_key not in adapter._active_sessions:
break
await asyncio.sleep(0.05)
assert processed == ["current", "follow up"]
assert session_key not in adapter._text_debounce
assert session_key not in adapter._pending_messages
assert session_key not in adapter._active_sessions
@pytest.mark.asyncio
async def test_force_flush_cancels_timer_without_duplicate_processing():
adapter = _make_adapter()
adapter._busy_text_debounce_seconds = 0.2
event = _make_event("queued once")
session_key = build_session_key(event.source)
adapter._active_sessions[session_key] = asyncio.Event()
await adapter.handle_message(event)
timer_task = adapter._text_debounce[session_key].task
flushed = await adapter._flush_text_debounce_now(session_key)
assert flushed is True
assert session_key not in adapter._text_debounce
assert adapter._pending_messages[session_key].text == "queued once"
await asyncio.sleep(0.3)
assert timer_task is not None
assert timer_task.cancelled() or timer_task.done()
assert adapter._pending_messages[session_key].text == "queued once"
@pytest.mark.asyncio
async def test_text_debounce_does_not_merge_different_senders():
adapter = _make_adapter()
adapter._busy_text_debounce_seconds = 1.0
first = _make_event(
"from alice",
chat_type="group",
user_id="alice",
user_name="Alice",
thread_id="topic-1",
)
# Interrupt event must be signalled exactly like before.
assert adapter._active_sessions[session_key].is_set()
second = _make_event(
"from bob",
chat_type="group",
user_id="bob",
user_name="Bob",
thread_id="topic-1",
)
session_key = build_session_key(first.source)
assert session_key == build_session_key(second.source)
adapter._active_sessions[session_key] = asyncio.Event()
await adapter.handle_message(first)
await adapter.handle_message(second)
assert adapter._pending_messages[session_key].text == "from alice"
assert _debounced_event(adapter, session_key).text == "from bob"
@pytest.mark.asyncio
async def test_control_and_clarify_messages_bypass_text_debounce():
adapter = _make_adapter()
started: list[str] = []
def _fake_start(event, session_key, *, interrupt_event=None):
started.append(event.text)
return True
adapter._start_session_processing = _fake_start # type: ignore[method-assign]
await adapter.handle_message(_make_event("/status"))
assert started == ["/status"]
assert adapter._text_debounce == {}
answer = _make_event("clarify answer")
session_key = build_session_key(answer.source)
adapter._active_sessions[session_key] = asyncio.Event()
adapter._message_handler = AsyncMock(return_value=None)
with patch("tools.clarify_gateway.get_pending_for_session", return_value=object()):
await adapter.handle_message(answer)
adapter._message_handler.assert_awaited_once_with(answer)
assert session_key not in adapter._text_debounce
assert session_key not in adapter._pending_messages
@pytest.mark.asyncio
async def test_debounce_skipped_when_busy_text_mode_not_queue():
adapter = _make_adapter()
adapter._busy_text_mode = ""
event = _make_event("direct merge")
session_key = build_session_key(event.source)
adapter._active_sessions[session_key] = asyncio.Event()
await adapter.handle_message(event)
assert adapter._pending_messages[session_key].text == "direct merge"
assert session_key not in adapter._text_debounce
def test_debounce_respects_env_var_override(monkeypatch):
monkeypatch.setenv("HERMES_GATEWAY_BUSY_TEXT_DEBOUNCE_SECONDS", "2.5")
adapter = _make_initialized_adapter()
assert adapter._busy_text_debounce_seconds == 2.5
@pytest.mark.asyncio
async def test_debounce_cleanup_in_cancel_background_tasks():
adapter = _make_adapter()
adapter._busy_text_debounce_seconds = 1.0
event = _make_event("cleanup test")
session_key = build_session_key(event.source)
adapter._active_sessions[session_key] = asyncio.Event()
await adapter.handle_message(event)
assert session_key in adapter._text_debounce
await adapter.cancel_background_tasks()
assert session_key not in adapter._text_debounce
@pytest.mark.asyncio
async def test_single_followup_is_stored_as_is():
"""One TEXT follow-up still lands as the event object itself
(no spurious wrapping / mutation) — guards against the merge path
breaking the simple case."""
adapter = _make_adapter()
adapter._busy_text_mode = ""
first = _make_event("only one")
session_key = build_session_key(first.source)
@@ -149,4 +349,29 @@ async def test_single_followup_is_stored_as_is():
pending = adapter._pending_messages[session_key]
assert pending is first
assert pending.text == "only one"
assert adapter._active_sessions[session_key].is_set()
assert not adapter._active_sessions[session_key].is_set()
def test_adapter_defaults_to_queue_mode(monkeypatch):
monkeypatch.delenv("HERMES_GATEWAY_BUSY_TEXT_MODE", raising=False)
adapter = _make_initialized_adapter()
assert adapter._busy_text_mode == "queue"
assert adapter._is_queue_text_debounce_candidate(_make_event("hello"))
def test_adapter_is_queue_text_debounce_candidate_by_default():
adapter = _make_adapter()
assert adapter._is_queue_text_debounce_candidate(_make_event("hello world"))
def test_command_messages_bypass_debounce_even_in_queue_mode():
adapter = _make_adapter()
assert not adapter._is_queue_text_debounce_candidate(_make_event(""))
assert not adapter._is_queue_text_debounce_candidate(_make_event("/stop"))
def test_busy_text_mode_respects_env_var_override(monkeypatch):
monkeypatch.setenv("HERMES_GATEWAY_BUSY_TEXT_MODE", "interrupt")
adapter = _make_initialized_adapter()
assert adapter._busy_text_mode == "interrupt"
assert not adapter._is_queue_text_debounce_candidate(_make_event("test"))

View File

@@ -15,6 +15,7 @@ from gateway.session import SessionSource, build_session_key
class DummyTelegramAdapter(BasePlatformAdapter):
def __init__(self):
super().__init__(PlatformConfig(enabled=True, token="fake-token"), Platform.TELEGRAM)
self._busy_text_mode = ""
self.sent = []
self.typing = []
self.processing_hooks = []

View File

@@ -65,6 +65,7 @@ def _make_runner():
runner._pending_messages = {}
runner._busy_ack_ts = {}
runner._draining = False
runner._busy_text_mode = "interrupt"
runner.adapters = {}
runner.config = MagicMock()
runner.session_store = None
@@ -84,6 +85,8 @@ def _make_adapter(platform_val="telegram"):
adapter.config = MagicMock()
adapter.config.extra = {}
adapter.platform = MagicMock(value=platform_val)
adapter._text_debounce = {}
adapter._busy_text_debounce_seconds = 0.6
return adapter
@@ -186,6 +189,32 @@ class TestBusySessionAck:
assert "respond once the current task finishes" in content
assert "Interrupting" not in content
@pytest.mark.asyncio
async def test_busy_text_mode_queue_delegates_to_adapter_handle_message(self):
"""busy_text_mode=queue lets the adapter debounce text silently."""
runner, sentinel = _make_runner()
runner._busy_input_mode = "interrupt"
runner._busy_text_mode = "queue"
adapter = _make_adapter()
first = _make_event(text="part one")
second = _make_event(text="part two")
sk = build_session_key(first.source)
agent = MagicMock()
runner._running_agents[sk] = agent
runner.adapters[first.source.platform] = adapter
runner.adapters[second.source.platform] = adapter
result1 = await runner._handle_active_session_busy_message(first, sk)
result2 = await runner._handle_active_session_busy_message(second, sk)
assert result1 is False
assert result2 is False
assert sk not in adapter._pending_messages
agent.interrupt.assert_not_called()
adapter._send_with_retry.assert_not_called()
@pytest.mark.asyncio
async def test_steer_mode_calls_agent_steer_no_interrupt_no_queue(self):
"""busy_input_mode='steer' injects via agent.steer() and skips queueing."""

View File

@@ -47,6 +47,7 @@ def _make_adapter():
"""Create a minimal adapter for testing the active-session guard."""
config = PlatformConfig(enabled=True, token="test-token")
adapter = _StubAdapter(config, Platform.TELEGRAM)
adapter._busy_text_mode = ""
adapter.sent_responses = []
async def _mock_handler(event):

View File

@@ -45,6 +45,7 @@ def _run_gateway_import(hermes_home: Path, initial_env: dict[str, str]) -> dict[
"HERMES_AGENT_TIMEOUT",
"HERMES_AGENT_TIMEOUT_WARNING",
"HERMES_GATEWAY_BUSY_INPUT_MODE",
"HERMES_GATEWAY_BUSY_TEXT_MODE",
"HERMES_TIMEZONE",
):
v = os.environ.get(k)
@@ -143,6 +144,15 @@ def test_config_display_busy_input_mode_wins_over_stale_env(hermes_home: Path) -
assert env.get("HERMES_GATEWAY_BUSY_INPUT_MODE") == "interrupt"
def test_config_display_busy_text_mode_wins_over_stale_env(hermes_home: Path) -> None:
_write_config(hermes_home, display_cfg={"busy_text_mode": "queue"})
_write_env(hermes_home, {"HERMES_GATEWAY_BUSY_TEXT_MODE": "interrupt"})
env = _run_gateway_import(hermes_home, initial_env={})
assert env.get("HERMES_GATEWAY_BUSY_TEXT_MODE") == "queue"
def test_config_timezone_wins_over_stale_env(hermes_home: Path) -> None:
_write_config(hermes_home, timezone="America/Los_Angeles")
_write_env(hermes_home, {"HERMES_TIMEZONE": "UTC"})

View File

@@ -103,6 +103,7 @@ class TestInterruptKeyConsistency:
async def test_handle_message_stores_under_session_key(self):
"""handle_message stores pending messages under session_key, not chat_id."""
adapter = StubAdapter()
adapter._busy_text_mode = ""
adapter.set_message_handler(lambda event: asyncio.sleep(0, result=None))
source = _source("-1001234", "group")
@@ -120,8 +121,8 @@ class TestInterruptKeyConsistency:
# NOT stored under chat_id
assert source.chat_id not in adapter._pending_messages
# Interrupt event was set
assert adapter._active_sessions[session_key].is_set()
# Text follow-ups queue silently and do not interrupt the active turn.
assert adapter._active_sessions[session_key].is_set() is False
@pytest.mark.asyncio
async def test_photo_followup_is_queued_without_interrupt(self):

View File

@@ -116,6 +116,24 @@ def test_load_busy_input_mode_prefers_env_then_config_then_default(tmp_path, mon
assert gateway_run.GatewayRunner._load_busy_input_mode() == "interrupt"
def test_load_busy_text_mode_defaults_to_queue_and_allows_interrupt(tmp_path, monkeypatch):
monkeypatch.setattr(gateway_run, "_hermes_home", tmp_path)
monkeypatch.delenv("HERMES_GATEWAY_BUSY_TEXT_MODE", raising=False)
assert gateway_run.GatewayRunner._load_busy_text_mode() == "queue"
(tmp_path / "config.yaml").write_text(
"display:\n busy_text_mode: interrupt\n", encoding="utf-8"
)
assert gateway_run.GatewayRunner._load_busy_text_mode() == "interrupt"
monkeypatch.setenv("HERMES_GATEWAY_BUSY_TEXT_MODE", "queue")
assert gateway_run.GatewayRunner._load_busy_text_mode() == "queue"
monkeypatch.setenv("HERMES_GATEWAY_BUSY_TEXT_MODE", "bogus")
assert gateway_run.GatewayRunner._load_busy_text_mode() == "queue"
def test_load_restart_drain_timeout_prefers_env_then_config_then_default(
tmp_path, monkeypatch, caplog
):

View File

@@ -53,6 +53,7 @@ class _StubAdapter(BasePlatformAdapter):
def _make_adapter():
config = PlatformConfig(enabled=True, token="test-token")
adapter = _StubAdapter(config, Platform.TELEGRAM)
adapter._busy_text_mode = ""
adapter.sent_responses = []
async def _mock_send_retry(chat_id, content, **kwargs):
@@ -396,4 +397,3 @@ class TestOldTaskCannotClobberNewerGuard:
# default path) still work.
adapter._release_session_guard(sk)
assert sk not in adapter._active_sessions