_fire_reasoning_delta called stream_reasoning_deltas_enabled() on every reasoning token; each call took _CONFIG_LOCK and paid load_config()'s full deepcopy on a cache hit, so every concurrently streaming thread in a WebUI/gateway process serialized behind one lock on the token path. Cache the opt-in on the agent for the life of one stream (reset in _reset_stream_delivery_tracking, so the next request re-reads config) and use load_config_readonly() for the lookup itself. (cherry picked from commit 20fab007e5c3e399e3923bb6d66649918292825c)
178 lines
6.6 KiB
Python
178 lines
6.6 KiB
Python
"""Asynchronous per-consumer plugin observers for streaming LLM output.
|
|
|
|
Each registered hook callback gets its own bounded queue + daemon worker thread
|
|
so plugin code never runs inline on the token path. Queues drop the oldest
|
|
pending event when full; dispatchers for callbacks that are no longer
|
|
registered are stopped lazily on the next lookup.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
import contextvars
|
|
import logging
|
|
import queue
|
|
import threading
|
|
from dataclasses import dataclass
|
|
from typing import Any, Callable
|
|
|
|
from hermes_cli.middleware import OBSERVER_SCHEMA_VERSION
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
_QUEUE_SIZE = 1024
|
|
_STOP = object()
|
|
|
|
|
|
@dataclass
|
|
class _ConsumerDispatcher:
|
|
hook_name: str
|
|
callback: Callable[..., Any]
|
|
events: "queue.Queue[tuple[contextvars.Context, dict[str, Any]] | object]"
|
|
thread: threading.Thread | None = None
|
|
|
|
|
|
_dispatcher_lock = threading.Lock()
|
|
_dispatchers: dict[tuple[str, int], _ConsumerDispatcher] = {}
|
|
|
|
|
|
def _callback_name(callback: Callable[..., Any]) -> str:
|
|
return getattr(callback, "__name__", repr(callback))
|
|
|
|
|
|
def _put_drop_oldest(events: "queue.Queue[Any]", item: Any) -> bool:
|
|
"""put_nowait; on a full queue evict the oldest pending event and retry once."""
|
|
try:
|
|
events.put_nowait(item)
|
|
return True
|
|
except queue.Full:
|
|
try:
|
|
events.get_nowait()
|
|
events.task_done()
|
|
except queue.Empty:
|
|
pass
|
|
try:
|
|
events.put_nowait(item)
|
|
return True
|
|
except queue.Full:
|
|
return False
|
|
|
|
|
|
def _worker(dispatcher: _ConsumerDispatcher) -> None:
|
|
while True:
|
|
item = dispatcher.events.get()
|
|
try:
|
|
if item is _STOP:
|
|
return
|
|
context, payload = item
|
|
payload = dict(payload)
|
|
payload.setdefault("telemetry_schema_version", OBSERVER_SCHEMA_VERSION)
|
|
try:
|
|
# The worker outlives every turn and serves every profile; run the callback in the
|
|
# enqueuing turn's contextvars so it sees that turn's profile scope (home override,
|
|
# secrets), not an unbound context that fail-closed plugin bindings refuse (#118538).
|
|
# ``copy()``: one snapshot fans out to N consumer threads and a Context can only be
|
|
# entered by one thread at a time.
|
|
context.copy().run(dispatcher.callback, **payload)
|
|
except Exception as exc:
|
|
# Fires once per streaming delta: a mis-declared callback fails identically every
|
|
# time, so it goes through the manager's warn-once reporter (#111922).
|
|
from hermes_cli.plugins import get_plugin_manager
|
|
|
|
get_plugin_manager()._report_hook_failure(dispatcher.hook_name, dispatcher.callback, payload, exc)
|
|
finally:
|
|
dispatcher.events.task_done()
|
|
|
|
|
|
def _registered_callbacks(hook_name: str) -> tuple[Callable[..., Any], ...]:
|
|
try:
|
|
from hermes_cli import plugins
|
|
return plugins.iter_hook_callbacks(hook_name)
|
|
except Exception:
|
|
logger.debug("plugin stream hook callback lookup failed: %s", hook_name, exc_info=True)
|
|
return ()
|
|
|
|
|
|
def _stop_dispatcher(dispatcher: _ConsumerDispatcher, timeout: float = 1.0) -> None:
|
|
_put_drop_oldest(dispatcher.events, _STOP)
|
|
if dispatcher.thread is not None:
|
|
dispatcher.thread.join(timeout=timeout)
|
|
|
|
|
|
def _start_dispatcher(hook_name: str, callback: Callable[..., Any]) -> _ConsumerDispatcher:
|
|
dispatcher = _ConsumerDispatcher(hook_name=hook_name, callback=callback, events=queue.Queue(maxsize=_QUEUE_SIZE))
|
|
dispatcher.thread = threading.Thread(
|
|
target=_worker, args=(dispatcher,), daemon=True, name=f"plugin-stream-hook:{hook_name}"
|
|
)
|
|
dispatcher.thread.start()
|
|
return dispatcher
|
|
|
|
|
|
def _dispatchers_for(hook_name: str) -> list[_ConsumerDispatcher]:
|
|
"""Live dispatcher per registered callback (restarting dead workers); stale
|
|
ones for unregistered callbacks are stopped outside the lock."""
|
|
callbacks = _registered_callbacks(hook_name)
|
|
if not callbacks:
|
|
return []
|
|
|
|
callback_ids = {id(callback) for callback in callbacks}
|
|
ready: list[_ConsumerDispatcher] = []
|
|
with _dispatcher_lock:
|
|
stale = [_dispatchers.pop(key) for key in list(_dispatchers) if key[0] == hook_name and key[1] not in callback_ids]
|
|
for callback in callbacks:
|
|
key = (hook_name, id(callback))
|
|
dispatcher = _dispatchers.get(key)
|
|
if dispatcher is None or dispatcher.thread is None or not dispatcher.thread.is_alive():
|
|
dispatcher = _dispatchers[key] = _start_dispatcher(hook_name, callback)
|
|
ready.append(dispatcher)
|
|
|
|
for dispatcher in stale:
|
|
_stop_dispatcher(dispatcher, timeout=0.2)
|
|
return ready
|
|
|
|
|
|
def enqueue_plugin_stream_hook(hook_name: str, **payload: Any) -> bool:
|
|
"""Queue an observer hook for each consumer without running plugin code inline."""
|
|
queued = False
|
|
item = (contextvars.copy_context(), dict(payload))
|
|
for dispatcher in _dispatchers_for(hook_name):
|
|
if _put_drop_oldest(dispatcher.events, item):
|
|
queued = True
|
|
else:
|
|
logger.debug(
|
|
"plugin stream hook queue full after drop-oldest: %s callback=%s",
|
|
hook_name, _callback_name(dispatcher.callback),
|
|
)
|
|
return queued
|
|
|
|
|
|
def has_stream_observer_hooks() -> bool:
|
|
return any(_registered_callbacks(name) for name in ("on_stream_start", "on_stream_delta", "on_stream_end"))
|
|
|
|
|
|
def has_reasoning_stream_observer_hooks() -> bool:
|
|
return stream_reasoning_deltas_enabled() and bool(_registered_callbacks("on_stream_delta"))
|
|
|
|
|
|
def stream_reasoning_deltas_enabled() -> bool:
|
|
"""Return True only when the user opted plugins into reasoning deltas.
|
|
|
|
Read-only scalar lookup: skips ``load_config()``'s deepcopy. Callers on the token path
|
|
should still cache the result per stream (``_fire_reasoning_delta`` does)."""
|
|
try:
|
|
from hermes_cli import config as config_mod
|
|
config = config_mod.load_config_readonly()
|
|
return bool(config_mod.cfg_get(config, "plugins", "stream_reasoning_deltas", default=False))
|
|
except Exception:
|
|
logger.debug("failed to read plugins.stream_reasoning_deltas", exc_info=True)
|
|
return False
|
|
|
|
|
|
def shutdown_plugin_stream_hook_dispatcher(timeout: float = 1.0) -> None:
|
|
"""Stop background stream hook dispatchers; used by tests and clean shutdown paths."""
|
|
global _dispatchers
|
|
with _dispatcher_lock:
|
|
dispatchers = list(_dispatchers.values())
|
|
_dispatchers = {}
|
|
for dispatcher in dispatchers:
|
|
_stop_dispatcher(dispatcher, timeout=timeout)
|