Shxiao 572744975d
fix(tools): run tool assembly off-loop at async entry points (#5224)
* fix(tools): run tool assembly off-loop at async entry points

get_available_tools() may block on MCP cache initialization while it is
called on async agent-assembly paths (task_tool, durable batch execution),
stalling the calling event loop for the full discovery duration.

Dispatch the (unchanged, synchronous) assembly call to a worker thread via
asyncio.to_thread at the two async entry points so the loop keeps processing
requests, SSE frames, cancellations, and timers.

Fixes #5172

* fix(tools): offload lead-agent assembly off-loop and pin with blocking-io anchors

Review follow-up for #5224:

- run_agent now dispatches agent_factory(...) through asyncio.to_thread, so
  lead-agent assembly (including both get_available_tools call sites in
  _assemble_lead_agent) runs off the event loop — the Gateway headline
  scenario from issue #5172.
- _ensure_sync_invocable_tool takes a double-checked threading.Lock, making
  the in-place tool.func wrap on the shared tool singletons explicitly
  single-shot now that assembly can run concurrently on worker threads.
- Add backend/tests/blocking_io/test_tool_assembly_offloop.py: blocking-probe
  anchors for task_tool and SubagentBatchService._execute_item under the
  strict Blockbuster gate, plus a meta-check proving the gate trips on the
  exact syscall class (ExtensionsConfig.from_file on the loop). Verified the
  anchor goes red when the offload is flattened back to a plain call.

* fix(gateway): build checkpoint state accessor off-loop; anchor run_agent offload

Review follow-up for #5224:

- Add abuild_checkpoint_state_accessor (asyncio.to_thread around the
  unchanged sync builder) and switch every async call site to it: the
  stateless_wait route, thread_runs, both threads call sites, and the
  build_thread_checkpoint_state_accessor boundary. The agent-factory
  assembly re-enters get_available_tools() and may block on MCP cache
  initialization; repeat calls hit _state_accessor_graph_cache and only
  pay the thread hop.
- Add a third blocking-io anchor driving the real run_agent with minimal
  RunManager/bridge stubs; the factory performs a real production blocking
  read (ExtensionsConfig.from_file()) and the test asserts assembly never
  runs on the main thread. Verified the anchor goes red when the run_agent
  offload is flattened back to a plain call.
- Adapt the test_threads_router checkpoint-builder patch sites to the new
  async name.

* refactor(tools): carry assembly offloads on a dedicated bounded pool

Review follow-up for #5224:

- Add utils/assembly_io.py: a dedicated ThreadPoolExecutor (default 8
  workers, DEER_FLOW_ASSEMBLY_WORKERS-overridable, mirroring
  utils/file_io.py and tools/sync.py) with run_assembly(), which copies
  contextvars explicitly. A hung stdio MCP server parks its worker for
  the full MCP timeout; carrying assembly hops on the loop's default
  executor would let a few parked assemblies queue every other
  to_thread/run_in_executor(None, ...) caller behind them.
- Switch all four offloads (run_agent, task_tool, batch _execute_item,
  abuild_checkpoint_state_accessor) to run_assembly().
- State the cold-path behavior in the accessor docstring: the graph
  cache validates factory identity, so non-identity-stable factories may
  duplicate lead-agent assembly across concurrent readers (MCP discovery
  stays process-wide single-flight); the pool bounds the duplicates.
- Add a fourth blocking-io anchor driving build_thread_checkpoint_state_
  accessor with a per-resolution fresh factory (always a cache miss) and
  the real production blocking read; enumerate all four offloads in the
  gate's module docstring. Verified the anchor goes red when
  abuild_checkpoint_state_accessor is flattened back to a plain call.

* fix(subagents): revalidate batch item before launch; make assembly pool observable

Review follow-up for #5224:

- _execute_item() revalidates the durable state right after assembly and
  before executor.execute_async(): renew_item_lease() returns valid=False
  when cancel_batch() terminalized the item or the lease was lost while
  assembly was parked, and the launch is skipped (the canceller already
  finalized the item). Previously the launch was unconditional and the
  poll loop's cancellation checks only started after execution began.
- Regression test driving the real SQLite repository: a blocking assembly
  probe parks _execute_item, cancel_batch() lands, and the launch is
  skipped with the item staying cancelled. Verified the test goes red
  when the revalidation is removed.
- run_assembly() tracks pending assemblies and logs a throttled WARNING
  once the pending count exceeds the worker count, so assembly starvation
  (workers parked on a hung MCP server) is distinguishable from idle.
- The run_agent blocking-io anchor now binds a sentinel extension
  snapshot via ctx.extensions and asserts the factory observed it through
  get_agent_build_extensions(), pinning run_assembly()'s ContextVar
  propagation. Verified red when ctx.run is dropped.
- Document the assembly pool in backend/AGENTS.md.

* fix(utils): decrement the assembly pending count on the pool thread

The pending-assembly counter behind the starvation warning decremented
from the asyncio future's done callback, which never fires once the
submitting loop is closed while its worker is still running: the count
ratcheted up permanently and eventually fired the starvation warning
with no starvation behind it (reproduced at 97dc9bec by review).

Decrement instead from the dispatched work item: run_assembly() wraps
func so a finally drops the count under the pending lock on the pool
thread, and the done callback is gone.

Pin the counter with tests/test_assembly_io.py: a healthy call returns
the count to zero, and an abandoned loop (stopped while the worker is
parked) does not wedge it — the abandoned case goes red against the old
done-callback decrement.

* docs(utils): fix the pending-counter comment after the decrement move

The comment still described the removed done-callback decrement,
contradicting _work()'s own comment; state the actual mechanism
(increment on the loop before dispatch, decrement from the dispatched
work item's finally on a pool thread).

* test(gateway): retarget checkpoint-accessor stubs to the services seam

thread_runs and runs now call abuild_checkpoint_state_accessor, so the
upstream wait-reader, regenerate-prepare, and idempotency tests must stub
the sync builder where abuild resolves it (app.gateway.services); stubbing
the removed router re-exports fails with AttributeError at setup. The
async seam semantics are unchanged: run_assembly invokes the stubbed
sync builder off-loop and propagates its return values and exceptions.

Move the agent/tool assembly off-load note from backend/AGENTS.md to
deerflow/utils/AGENTS.md (next to assembly_io.py) so the effective
instruction chain for agents/middlewares no longer grows past the AG002
hard limit.

* fix(runtime): serialize same-key accessor assembly and release queued-cancel slots

Address the three review follow-ups on the assembly off-load:

- assembly_io: a job cancelled while still queued never runs its work
  item, so the dispatched finally never fired and _pending_assemblies
  stayed elevated until a false starvation warning. Exactly-once cleanup
  now rides the concurrent future's cancelled() state — cancel() only
  succeeds before the executor starts the item, so cancelled() is true
  precisely when the finally will never run — plus a submit-failure
  release; the one-worker queued-cancellation case is pinned red/green.
- services: overlapping cold readers sharing one cache key could both
  run full agent assembly. _state_accessor_graph now serializes per key
  through a thread-side KeyedLockTable (pool threads, no running loop)
  and re-validates factory/app-config identity under the lock, so the
  factory runs exactly once while identity changes still rebuild. Cache
  dict access is lock-guarded now that construction runs off-loop.
- guidance inventory: register deerflow/utils/AGENTS.md in
  EXPECTED_GUIDANCE_PATHS so test_repository_has_the_approved_scoped_
  guidance_shape matches the relocated assembly note (CI shard 4).

* test(keyed-lock): pin KeyedLockTable reclamation and waiter bypass directly

Thread-side counterparts of the async table's own tests: overlapping
hold() calls serialize (a late arrival joins the live entry instead of
creating a second lock that bypasses a queued waiter), the last check-in
pops the entry, and many unique keys leave the registry empty. Both
regressions verified red — popping unconditionally trips the late-arrival
test, never reclaiming trips the many-keys test.

---------

Co-authored-by: Willem Jiang <willem.jiang@gmail.com>
2026-09-12 07:41:24 +08:00

216 lines
8.8 KiB
Python

import logging
import threading
from langchain.tools import BaseTool
from deerflow.config import get_app_config
from deerflow.config.app_config import AppConfig
from deerflow.mcp.tasks.runtime import is_mcp_task_runtime_available
from deerflow.reflection import resolve_variable
from deerflow.sandbox.security import is_host_bash_allowed
from deerflow.subagents.batch_runtime import is_subagent_batch_runtime_available
from deerflow.tools.builtins import (
ask_clarification_tool,
batch_status,
batch_task,
cancel_background_task,
cancel_batch,
list_background_tasks,
list_uploaded_files,
present_file_tool,
review_skill_package,
task_tool,
view_image_tool,
)
from deerflow.tools.mcp_metadata import tag_mcp_tool
from deerflow.tools.sync import make_sync_tool_wrapper
logger = logging.getLogger(__name__)
BUILTIN_TOOLS = [
present_file_tool,
ask_clarification_tool,
review_skill_package,
]
SUBAGENT_TOOLS = [
task_tool,
# task_status_tool is no longer exposed to LLM (backend handles polling internally)
]
def _is_host_bash_tool(tool: object) -> bool:
"""Return True if the tool config represents a host-bash execution surface."""
group = getattr(tool, "group", None)
use = getattr(tool, "use", None)
if group == "bash":
return True
if use == "deerflow.sandbox.tools:bash_tool":
return True
return False
_sync_invocable_tool_lock = threading.Lock()
def _ensure_sync_invocable_tool(tool: BaseTool) -> BaseTool:
"""Attach a sync wrapper to async-only tools used by sync agent callers.
The wrapped objects are process-wide singletons (BUILTIN_TOOLS /
SUBAGENT_TOOLS / MCP cache entries) and tool assembly may now run on
worker threads concurrently; double-checked locking makes the in-place
``tool.func`` wrap explicitly single-shot instead of incidental.
"""
if getattr(tool, "func", None) is not None or getattr(tool, "coroutine", None) is None:
return tool
with _sync_invocable_tool_lock:
if getattr(tool, "func", None) is None:
tool.func = make_sync_tool_wrapper(tool.coroutine, tool.name)
return tool
def get_available_tools(
groups: list[str] | None = None,
include_mcp: bool = True,
model_name: str | None = None,
subagent_enabled: bool = False,
*,
include_upload_tool: bool = True,
app_config: AppConfig | None = None,
) -> list[BaseTool]:
"""Get all available tools from config.
Note: MCP tools should be initialized at application startup using
`initialize_mcp_tools()` from deerflow.mcp module.
Args:
groups: Optional list of tool groups to filter by.
include_mcp: Whether to include tools from MCP servers (default: True).
model_name: Optional model name to determine if vision tools should be included.
subagent_enabled: Whether to include subagent tools (task, task_status).
include_upload_tool: Whether to include ``list_uploaded_files`` (default: True).
Ordinary task subagents enable it only after snapshotting the
parent's current-run upload state. Durable batch and non-standard
subagent callers without that state keep it disabled.
Returns:
List of available tools.
"""
config = app_config or get_app_config()
tool_configs = [tool for tool in config.tools if groups is None or tool.group in groups]
# Do not expose host bash by default when LocalSandboxProvider is active.
if not is_host_bash_allowed(config):
tool_configs = [tool for tool in tool_configs if not _is_host_bash_tool(tool)]
loaded_tools_raw = [(cfg, resolve_variable(cfg.use, BaseTool)) for cfg in tool_configs]
# Warn when the config ``name`` field and the tool object's ``.name``
# attribute diverge — this mismatch is the root cause of issue #1803 where
# the LLM receives one name in its tool schema but the runtime router
# recognises a different name, producing "not a valid tool" errors.
for cfg, loaded in loaded_tools_raw:
if cfg.name != loaded.name:
logger.warning(
"Tool name mismatch: config name %r does not match tool .name %r (use: %s). The tool's own .name will be used for binding.",
cfg.name,
loaded.name,
cfg.use,
)
loaded_tools = [_ensure_sync_invocable_tool(t) for _, t in loaded_tools_raw]
# Conditionally add tools based on config
builtin_tools = BUILTIN_TOOLS.copy()
if is_mcp_task_runtime_available():
builtin_tools.extend((list_background_tasks, cancel_background_task))
if include_upload_tool:
builtin_tools.append(list_uploaded_files)
skill_evolution_config = getattr(config, "skill_evolution", None)
if getattr(skill_evolution_config, "enabled", False):
from deerflow.tools.skill_manage_tool import skill_manage_tool
builtin_tools.append(skill_manage_tool)
# Add subagent tools only if enabled via runtime parameter
if subagent_enabled:
builtin_tools.extend(SUBAGENT_TOOLS)
if is_subagent_batch_runtime_available():
builtin_tools.extend((batch_task, batch_status, cancel_batch))
logger.info("Including native subagent tools")
# If no model_name specified, use the first model (default)
if model_name is None and config.models:
model_name = config.models[0].name
# Add view_image_tool only if the model supports vision
model_config = config.get_model_config(model_name) if model_name else None
if model_config is not None and model_config.supports_vision:
builtin_tools.append(view_image_tool)
logger.info(f"Including view_image_tool for model '{model_name}' (supports_vision=True)")
# Get cached MCP tools if enabled
# NOTE: We use ExtensionsConfig.from_file() instead of config.extensions
# to always read the latest configuration from disk. This ensures that changes
# made through the Gateway API (which runs in a separate process) are immediately
# reflected when loading MCP tools.
mcp_tools = []
if include_mcp:
try:
from deerflow.config.extensions_config import ExtensionsConfig
from deerflow.mcp.cache import get_cached_mcp_tools
extensions_config = ExtensionsConfig.from_file()
if extensions_config.get_enabled_mcp_servers():
mcp_tools = get_cached_mcp_tools()
if mcp_tools:
logger.info(f"Using {len(mcp_tools)} cached MCP tool(s)")
# Tag MCP-sourced tools so deferred-tool assembly at each
# agent construction site can identify them. Lead agents
# assemble their full configured MCP catalog and apply active
# skill policy at runtime; subagents may pass an already
# policy-filtered list because their skills load at startup.
for t in mcp_tools:
tag_mcp_tool(t)
except ImportError:
logger.warning("MCP module not available. Install 'langchain-mcp-adapters' package to enable MCP tools.")
except Exception as e:
logger.error(f"Failed to get cached MCP tools: {e}")
# Add invoke_acp_agent tool if any ACP agents are configured
acp_tools: list[BaseTool] = []
try:
from deerflow.tools.builtins.invoke_acp_agent_tool import build_invoke_acp_agent_tool
if app_config is None:
from deerflow.config.acp_config import get_acp_agents
acp_agents = get_acp_agents()
else:
acp_agents = getattr(config, "acp_agents", {}) or {}
if acp_agents:
acp_tools.append(build_invoke_acp_agent_tool(acp_agents))
logger.info(f"Including invoke_acp_agent tool ({len(acp_agents)} agent(s): {list(acp_agents.keys())})")
except Exception as e:
logger.warning(f"Failed to load ACP tool: {e}")
logger.info(f"Total tools loaded: {len(loaded_tools)}, built-in tools: {len(builtin_tools)}, MCP tools: {len(mcp_tools)}, ACP tools: {len(acp_tools)}")
# Deduplicate by tool name — config-loaded tools take priority, followed by
# built-ins, MCP tools, and ACP tools. Duplicate names cause the LLM to
# receive ambiguous or concatenated function schemas (issue #1803).
all_tools = [_ensure_sync_invocable_tool(t) for t in loaded_tools + builtin_tools + mcp_tools + acp_tools]
seen_names: set[str] = set()
unique_tools: list[BaseTool] = []
for t in all_tools:
if t.name not in seen_names:
unique_tools.append(t)
seen_names.add(t.name)
else:
logger.warning(
"Duplicate tool name %r detected and skipped — check your config.yaml and MCP server registrations (issue #1803).",
t.name,
)
return unique_tools