Grayscale/nanobot
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

refactor(context): trim replay cap plumbing

Browse Source
This commit is contained in:
chengyongru
2026-06-29 14:23:55 +08:00
committed by Xubin Ren
parent 40282e3b74
commit 57f0c859fc
5 changed files with 29 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files
nanobot/agent/loop.py
+2 -9
View File
@@ -205,7 +205,6 @@ class AgentLoop:
timezone: str | None = None, timezone: str | None = None,
session_ttl_minutes: int = 0, session_ttl_minutes: int = 0,
consolidation_ratio: float = 0.5, consolidation_ratio: float = 0.5,
max_messages: int = 0,
hooks: list[AgentHook] | None = None, hooks: list[AgentHook] | None = None,
unified_session: bool = False, unified_session: bool = False,
disabled_skills: list[str] | None = None, disabled_skills: list[str] | None = None,
@@ -296,12 +295,7 @@ class AgentLoop:
llm_wall_timeout_for_session=lambda sk: runner_wall_llm_timeout_s(self.sessions, sk), llm_wall_timeout_for_session=lambda sk: runner_wall_llm_timeout_s(self.sessions, sk),
) )
self._unified_session = unified_session self._unified_session = unified_session
self._explicit_max_messages = max_messages > 0 self._max_messages = replay_max_messages_for_context(self.context_window_tokens)
self._max_messages = (
max_messages
if self._explicit_max_messages
else replay_max_messages_for_context(self.context_window_tokens)
)
self._running = False self._running = False
self._mcp_servers = mcp_servers or {} self._mcp_servers = mcp_servers or {}
self._mcp_stacks: dict[str, AsyncExitStack] = {} self._mcp_stacks: dict[str, AsyncExitStack] = {}
@@ -444,8 +438,7 @@ class AgentLoop:
logger.info("Runtime model switched for next turn: {} -> {}", old_model, model) logger.info("Runtime model switched for next turn: {} -> {}", old_model, model)
def _sync_replay_max_messages(self) -> None: def _sync_replay_max_messages(self) -> None:
if not self._explicit_max_messages: self._max_messages = replay_max_messages_for_context(self.context_window_tokens)
self._max_messages = replay_max_messages_for_context(self.context_window_tokens)
def _refresh_provider_snapshot(self) -> None: def _refresh_provider_snapshot(self) -> None:
if self._provider_snapshot_loader is None: if self._provider_snapshot_loader is None:
nanobot/config/loader.py
+6 -6
View File
@@ -156,12 +156,12 @@ def _migrate_config(data: dict) -> dict:
agents = data.get("agents", {}) agents = data.get("agents", {})
defaults = agents.get("defaults", {}) if isinstance(agents, dict) else {} defaults = agents.get("defaults", {}) if isinstance(agents, dict) else {}
if isinstance(defaults, dict): if isinstance(defaults, dict):
legacy_max_message_keys = [ had_legacy_max_messages = (
key for key in ("maxMessages", "max_messages") if key in defaults "maxMessages" in defaults or "max_messages" in defaults
] )
if legacy_max_message_keys: defaults.pop("maxMessages", None)
for key in legacy_max_message_keys: defaults.pop("max_messages", None)
defaults.pop(key, None) if had_legacy_max_messages:
# TODO(next version): Remove this legacy cleanup branch; the schema # TODO(next version): Remove this legacy cleanup branch; the schema
# will silently ignore this field once the warning grace period ends. # will silently ignore this field once the warning grace period ends.
logger.warning( logger.warning(
nanobot/session/manager.py
+3 -4
View File
@@ -29,7 +29,6 @@ from nanobot.utils.subagent_channel_display import scrub_subagent_announce_body
FILE_MAX_MESSAGES = 2000 FILE_MAX_MESSAGES = 2000
MIN_REPLAY_MAX_MESSAGES = 120 MIN_REPLAY_MAX_MESSAGES = 120
REPLAY_TOKENS_PER_MESSAGE = 100 REPLAY_TOKENS_PER_MESSAGE = 100
DEFAULT_REPLAY_MAX_MESSAGES = FILE_MAX_MESSAGES
_MESSAGE_TIME_PREFIX_RE = re.compile(r"^\[Message Time: [^\]]+\]\n?") _MESSAGE_TIME_PREFIX_RE = re.compile(r"^\[Message Time: [^\]]+\]\n?")
_LOCAL_IMAGE_BREADCRUMB_RE = re.compile(r"^\[image: (?:/|~)[^\]]+\]\s*$") _LOCAL_IMAGE_BREADCRUMB_RE = re.compile(r"^\[image: (?:/|~)[^\]]+\]\s*$")
_TOOL_CALL_ECHO_RE = re.compile(r'^\s*(?:generate_image|message)\([^)]*\)\s*$') _TOOL_CALL_ECHO_RE = re.compile(r'^\s*(?:generate_image|message)\([^)]*\)\s*$')
@@ -48,7 +47,7 @@ _FORK_VOLATILE_METADATA_KEYS = {
def replay_max_messages_for_context(context_window_tokens: int | None) -> int: def replay_max_messages_for_context(context_window_tokens: int | None) -> int:
if not context_window_tokens or context_window_tokens <= 0: if not context_window_tokens or context_window_tokens <= 0:
return DEFAULT_REPLAY_MAX_MESSAGES return FILE_MAX_MESSAGES
return min( return min(
FILE_MAX_MESSAGES, FILE_MAX_MESSAGES,
max(MIN_REPLAY_MAX_MESSAGES, context_window_tokens // REPLAY_TOKENS_PER_MESSAGE), max(MIN_REPLAY_MAX_MESSAGES, context_window_tokens // REPLAY_TOKENS_PER_MESSAGE),
@@ -144,7 +143,7 @@ class Session:
def get_history( def get_history(
self, self,
max_messages: int = DEFAULT_REPLAY_MAX_MESSAGES, max_messages: int = FILE_MAX_MESSAGES,
*, *,
max_tokens: int = 0, max_tokens: int = 0,
extend_to_user: bool = False, extend_to_user: bool = False,
@@ -155,7 +154,7 @@ class Session:
token budget from the tail (``max_tokens``) when provided. token budget from the tail (``max_tokens``) when provided.
""" """
unconsolidated = self.messages[self.last_consolidated:] unconsolidated = self.messages[self.last_consolidated:]
max_messages = max_messages if max_messages > 0 else DEFAULT_REPLAY_MAX_MESSAGES max_messages = max_messages if max_messages > 0 else FILE_MAX_MESSAGES
start_idx = recent_message_start_index( start_idx = recent_message_start_index(
unconsolidated, unconsolidated,
max_messages, max_messages,
tests/agent/conftest.py
+2 -4
View File
@@ -38,7 +38,6 @@ def make_loop(
model: str = "test-model", model: str = "test-model",
context_window_tokens: int = 128_000, context_window_tokens: int = 128_000,
session_ttl_minutes: int = 0, session_ttl_minutes: int = 0,
max_messages: int = 120,
unified_session: bool = False, unified_session: bool = False,
mcp_servers: dict | None = None, mcp_servers: dict | None = None,
tools_config=None, tools_config=None,
@@ -64,7 +63,6 @@ def make_loop(
model=model, model=model,
context_window_tokens=context_window_tokens, context_window_tokens=context_window_tokens,
session_ttl_minutes=session_ttl_minutes, session_ttl_minutes=session_ttl_minutes,
max_messages=max_messages,
unified_session=unified_session, unified_session=unified_session,
) )
if mcp_servers is not None: if mcp_servers is not None:
@@ -79,8 +77,8 @@ def make_loop(
if patch_deps: if patch_deps:
with patch("nanobot.agent.loop.ContextBuilder"), \ with patch("nanobot.agent.loop.ContextBuilder"), \
patch("nanobot.agent.loop.SessionManager"), \ patch("nanobot.agent.loop.SessionManager"), \
patch("nanobot.agent.loop.SubagentManager") as MockSubMgr: patch("nanobot.agent.loop.SubagentManager") as mock_sub_mgr:
MockSubMgr.return_value.cancel_by_session = AsyncMock(return_value=0) mock_sub_mgr.return_value.cancel_by_session = AsyncMock(return_value=0)
return AgentLoop(**kwargs) return AgentLoop(**kwargs)
return AgentLoop(**kwargs) return AgentLoop(**kwargs)
tests/agent/test_max_messages_config.py
+16 -28
View File
@@ -13,17 +13,14 @@ from nanobot.bus.queue import MessageBus
from nanobot.providers.base import LLMResponse from nanobot.providers.base import LLMResponse
from nanobot.providers.factory import ProviderSnapshot from nanobot.providers.factory import ProviderSnapshot
from nanobot.session.manager import ( from nanobot.session.manager import (
DEFAULT_REPLAY_MAX_MESSAGES, FILE_MAX_MESSAGES,
Session, Session,
replay_max_messages_for_context, replay_max_messages_for_context,
) )
DEFAULT_MAX_MESSAGES = DEFAULT_REPLAY_MAX_MESSAGES
def _make_loop( def _make_loop(
tmp_path: Path, tmp_path: Path,
max_messages: int = 0,
context_window_tokens: int = 200_000, context_window_tokens: int = 200_000,
) -> AgentLoop: ) -> AgentLoop:
provider = MagicMock() provider = MagicMock()
@@ -34,7 +31,6 @@ def _make_loop(
provider=provider, provider=provider,
workspace=tmp_path, workspace=tmp_path,
model="test-model", model="test-model",
max_messages=max_messages,
context_window_tokens=context_window_tokens, context_window_tokens=context_window_tokens,
) )
@@ -67,29 +63,16 @@ class TestMaxMessagesInit:
def test_context_formula(self) -> None: def test_context_formula(self) -> None:
assert replay_max_messages_for_context(8_000) == 120 assert replay_max_messages_for_context(8_000) == 120
assert replay_max_messages_for_context(32_768) == 327 assert replay_max_messages_for_context(32_768) == 327
assert replay_max_messages_for_context(200_000) == DEFAULT_MAX_MESSAGES assert replay_max_messages_for_context(200_000) == FILE_MAX_MESSAGES
def test_default_for_200k_context_reaches_file_cap(self, tmp_path: Path) -> None: def test_default_for_200k_context_reaches_file_cap(self, tmp_path: Path) -> None:
loop = _make_loop(tmp_path) loop = _make_loop(tmp_path)
assert loop._max_messages == DEFAULT_MAX_MESSAGES assert loop._max_messages == FILE_MAX_MESSAGES
def test_default_scales_with_context_window(self, tmp_path: Path) -> None: def test_default_scales_with_context_window(self, tmp_path: Path) -> None:
loop = _make_loop(tmp_path, context_window_tokens=32_768) loop = _make_loop(tmp_path, context_window_tokens=32_768)
assert loop._max_messages == 327 assert loop._max_messages == 327
def test_positive_value_stored(self, tmp_path: Path) -> None:
loop = _make_loop(tmp_path, max_messages=25)
assert loop._max_messages == 25
def test_zero_uses_context_derived_limit(self, tmp_path: Path) -> None:
loop = _make_loop(tmp_path, max_messages=0)
assert loop._max_messages == DEFAULT_MAX_MESSAGES
def test_negative_treated_as_builtin_limit(self, tmp_path: Path) -> None:
"""Negative values should not produce negative slicing."""
loop = _make_loop(tmp_path, max_messages=-5)
assert loop._max_messages == DEFAULT_MAX_MESSAGES
def test_provider_refresh_resyncs_context_derived_limit(self, tmp_path: Path) -> None: def test_provider_refresh_resyncs_context_derived_limit(self, tmp_path: Path) -> None:
old_provider = MagicMock() old_provider = MagicMock()
old_provider.get_default_model.return_value = "old-model" old_provider.get_default_model.return_value = "old-model"
@@ -112,7 +95,7 @@ class TestMaxMessagesInit:
assert loop._max_messages == 327 assert loop._max_messages == 327
loop._refresh_provider_snapshot() loop._refresh_provider_snapshot()
assert loop._max_messages == DEFAULT_MAX_MESSAGES assert loop._max_messages == FILE_MAX_MESSAGES
class TestGetHistoryWithMaxMessages: class TestGetHistoryWithMaxMessages:
@@ -121,7 +104,7 @@ class TestGetHistoryWithMaxMessages:
def test_default_uses_builtin_limit(self) -> None: def test_default_uses_builtin_limit(self) -> None:
session = _populated_session(80) session = _populated_session(80)
history = session.get_history() history = session.get_history()
assert len(history) <= DEFAULT_MAX_MESSAGES assert len(history) <= FILE_MAX_MESSAGES
def test_explicit_max_messages_limits_output(self) -> None: def test_explicit_max_messages_limits_output(self) -> None:
session = _populated_session(40) # 80 messages total session = _populated_session(40) # 80 messages total
@@ -137,7 +120,7 @@ class TestGetHistoryWithMaxMessages:
def test_max_messages_zero_uses_builtin_limit(self) -> None: def test_max_messages_zero_uses_builtin_limit(self) -> None:
session = _populated_session(80) # 160 messages total session = _populated_session(80) # 160 messages total
history = session.get_history(max_messages=0) history = session.get_history(max_messages=0)
assert len(history) <= DEFAULT_MAX_MESSAGES assert len(history) <= FILE_MAX_MESSAGES
def test_small_session_unaffected(self) -> None: def test_small_session_unaffected(self) -> None:
"""When session has fewer messages than max_messages, all are returned.""" """When session has fewer messages than max_messages, all are returned."""
@@ -152,7 +135,8 @@ class TestMaxMessagesIntegration:
@pytest.mark.asyncio @pytest.mark.asyncio
async def test_process_message_passes_limit_to_history_call(self, tmp_path: Path) -> None: async def test_process_message_passes_limit_to_history_call(self, tmp_path: Path) -> None:
"""The real message path should pass max_messages into session history replay.""" """The real message path should pass max_messages into session history replay."""
loop = _make_loop(tmp_path, max_messages=25) loop = _make_loop(tmp_path)
loop._max_messages = 25
loop.provider.chat_with_retry = AsyncMock( loop.provider.chat_with_retry = AsyncMock(
return_value=LLMResponse(content="ok", tool_calls=[], usage={}) return_value=LLMResponse(content="ok", tool_calls=[], usage={})
) )
@@ -171,8 +155,11 @@ class TestMaxMessagesIntegration:
assert mock_hist.call_args.kwargs["extend_to_user"] is False assert mock_hist.call_args.kwargs["extend_to_user"] is False
@pytest.mark.asyncio @pytest.mark.asyncio
async def test_zero_limit_passes_builtin_limit_to_history_call(self, tmp_path: Path) -> None: async def test_default_limit_passes_context_derived_limit_to_history_call(
loop = _make_loop(tmp_path, max_messages=0) self,
tmp_path: Path,
) -> None:
loop = _make_loop(tmp_path)
loop.provider.chat_with_retry = AsyncMock( loop.provider.chat_with_retry = AsyncMock(
return_value=LLMResponse(content="ok", tool_calls=[], usage={}) return_value=LLMResponse(content="ok", tool_calls=[], usage={})
) )
@@ -186,7 +173,7 @@ class TestMaxMessagesIntegration:
) )
assert result is not None assert result is not None
assert mock_hist.call_args.kwargs["max_messages"] == DEFAULT_MAX_MESSAGES assert mock_hist.call_args.kwargs["max_messages"] == FILE_MAX_MESSAGES
assert mock_hist.call_args.kwargs["extend_to_user"] is False assert mock_hist.call_args.kwargs["extend_to_user"] is False
@pytest.mark.asyncio @pytest.mark.asyncio
@@ -195,7 +182,8 @@ class TestMaxMessagesIntegration:
tmp_path: Path, tmp_path: Path,
) -> None: ) -> None:
"""A live user turn should not extend history to an older long tool turn.""" """A live user turn should not extend history to an older long tool turn."""
loop = _make_loop(tmp_path, max_messages=6) loop = _make_loop(tmp_path)
loop._max_messages = 6
loop.provider.chat_with_retry = AsyncMock( loop.provider.chat_with_retry = AsyncMock(
return_value=LLMResponse(content="ok", tool_calls=[], usage={}) return_value=LLMResponse(content="ok", tool_calls=[], usage={})
) )