File size: 10,730 Bytes
e6c104f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 | """Translate LangGraph `astream_events` into the SSE event contract.
The first version of this adapter *predicted* the next node after each update.
After `verify` that guess is often wrong — the router and the adapter can
disagree about whether to replan, run another tool, or synthesise. Each wrong
guess emitted a `node_start synthesizer` the UI never closed, so the trace
showed "Synthesizer attempt 8 / running" while the graph was still on tools.
`astream_events` reports the node LangGraph actually entered, so a start is
only emitted for a real visit.
"""
from __future__ import annotations
import asyncio
import time
import uuid
from typing import Any, AsyncIterator, Dict, Optional
from .agent.graph import get_graph, run_config
from .config import get_settings
from .events import (
Audit,
Clarification,
Error,
Heartbeat,
NodeEnd,
NodeName,
NodeStart,
Plan,
PlanStep,
RunEnd,
RunStart,
Token,
ToolEnd,
ToolStart,
sse,
)
GRAPH_TO_UI = {
"gatekeeper": NodeName.gatekeeper,
"planner": NodeName.planner,
"execute_tool": NodeName.tool_executor,
"verify": NodeName.auditor,
"synthesize": NodeName.synthesizer,
}
PING_S = 5.0
def _preview(output: Any, width: int = 240) -> str:
text = output if isinstance(output, str) else str(output)
return " ".join(text.split())[:width]
def _node_name(event: Dict[str, Any]) -> Optional[str]:
"""Return a supervisor node name, or None for nested LLM/tool events."""
name = event.get("name")
node = (event.get("metadata") or {}).get("langgraph_node")
if name in GRAPH_TO_UI and node == name:
return name
return None
async def live_stream(query: str, conversation_id: Optional[str] = None) -> AsyncIterator[str]:
"""Yield SSE frames for one live supervisor run."""
settings = get_settings()
seq = 0
started = time.perf_counter()
last_mark = started
deadline = started + settings.agent_timeout_s
node_timings_ms: Dict[str, int] = {}
attempts: Dict[str, int] = {}
llm_calls = 0
output_tokens = 0
streamed_tokens = False
replan_count = 0
run_id = f"run_{uuid.uuid4().hex[:12]}"
conversation_id = conversation_id or f"conv_{uuid.uuid4().hex[:12]}"
def emit(event) -> str:
nonlocal seq
seq += 1
event.seq = seq
return sse(event)
yield emit(RunStart(seq=0, run_id=run_id, conversation_id=conversation_id, query=query))
iterator = get_graph().astream_events(
{"original_request": query, "replan_count": 0},
run_config(),
version="v2",
).__aiter__()
# wait_for() cancels its awaitable on timeout. Cancelling anext() closes
# the async generator, so a RunPod call that takes >PING_S would abort the
# whole graph. Shield the next event and reuse the same task across pings.
pending: Optional[asyncio.Task] = None
try:
while True:
remaining = deadline - time.perf_counter()
if remaining <= 0:
if pending is not None:
pending.cancel()
yield emit(
Error(
seq=0,
code="timeout",
message=f"Agent exceeded its {settings.agent_timeout_s:.0f}s ceiling.",
retryable=True,
)
)
return
if pending is None:
pending = asyncio.ensure_future(anext(iterator))
try:
event = await asyncio.wait_for(
asyncio.shield(pending), timeout=min(PING_S, remaining)
)
pending = None
except asyncio.TimeoutError:
if time.perf_counter() >= deadline:
pending.cancel()
yield emit(
Error(
seq=0,
code="timeout",
message=f"Agent exceeded its {settings.agent_timeout_s:.0f}s ceiling.",
retryable=True,
)
)
return
yield emit(Heartbeat(seq=0))
continue
except StopAsyncIteration:
pending = None
break
kind = event.get("event")
graph_node = _node_name(event)
if (
kind == "on_chat_model_stream"
and (event.get("metadata") or {}).get("langgraph_node") == "synthesize"
):
chunk = (event.get("data") or {}).get("chunk")
text = getattr(chunk, "content", None) or ""
if isinstance(text, list):
text = "".join(
part.get("text", "") if isinstance(part, dict) else str(part)
for part in text
)
if text:
streamed_tokens = True
output_tokens += 1
yield emit(Token(seq=0, text=text))
continue
if graph_node is None:
continue
ui_node = GRAPH_TO_UI[graph_node]
if kind == "on_chain_start":
attempts[graph_node] = attempts.get(graph_node, 0) + 1
last_mark = time.perf_counter()
yield emit(
NodeStart(seq=0, node=ui_node, attempt=attempts[graph_node])
)
continue
if kind != "on_chain_end":
continue
now = time.perf_counter()
duration_ms = int((now - last_mark) * 1000)
node_timings_ms[ui_node.value] = (
node_timings_ms.get(ui_node.value, 0) + duration_ms
)
payload = (event.get("data") or {}).get("output") or {}
if not isinstance(payload, dict):
payload = {}
yield emit(
NodeEnd(
seq=0,
node=ui_node,
attempt=attempts.get(graph_node, 1),
duration_ms=duration_ms,
)
)
if graph_node == "gatekeeper":
llm_calls += 1
question = payload.get("clarification_question")
if question:
yield emit(Clarification(seq=0, question=question))
scoped = payload.get("final_response")
if scoped:
streamed_tokens = True
output_tokens = len(str(scoped).split())
yield emit(Token(seq=0, text=str(scoped)))
elif graph_node == "planner":
llm_calls += 1
replan_count = int(payload.get("replan_count") or replan_count)
steps = [
PlanStep(
tool_name=s["tool_name"],
tool_input=s.get("tool_input") or "",
)
for s in payload.get("plan") or []
if isinstance(s, dict) and s.get("tool_name") != "FINISH"
]
yield emit(
Plan(seq=0, attempt=attempts.get(graph_node, 1), steps=steps)
)
elif graph_node == "execute_tool":
steps = payload.get("intermediate_steps") or []
if steps:
last = steps[-1]
output = last.get("tool_output")
failed = isinstance(output, str) and (
output.startswith("Tool raised")
or output.startswith("Unknown tool")
)
yield emit(
ToolStart(
seq=0,
tool_name=last.get("tool_name") or "unknown",
tool_input=last.get("tool_input") or "",
)
)
yield emit(
ToolEnd(
seq=0,
tool_name=last.get("tool_name") or "unknown",
duration_ms=duration_ms,
ok=not failed,
preview=_preview(output),
result_count=len(output) if isinstance(output, list) else None,
error=(str(output)[:200] if failed else None),
)
)
elif graph_node == "verify":
llm_calls += 1
history = payload.get("verification_history") or []
if history:
audit = history[-1]
score = int(audit.get("confidence_score") or 1)
action = (
"replan" if score < settings.audit_pass_score else "accept"
)
yield emit(
Audit(
seq=0,
attempt=attempts.get(graph_node, 1),
confidence_score=score,
is_relevant=bool(audit.get("is_relevant")),
reasoning=str(audit.get("reasoning") or ""),
action=action,
)
)
elif graph_node == "synthesize":
llm_calls += 1
if not streamed_tokens:
answer = payload.get("final_response") or ""
output_tokens = len(str(answer).split())
if answer:
yield emit(Token(seq=0, text=str(answer)))
except asyncio.CancelledError:
raise
except Exception as exc: # noqa: BLE001 - the client needs the error on the stream
yield emit(
Error(seq=0, code=type(exc).__name__, message=str(exc)[:400], retryable=True)
)
return
yield emit(
RunEnd(
seq=0,
run_id=run_id,
duration_ms=int((time.perf_counter() - started) * 1000),
node_timings_ms=node_timings_ms,
llm_calls=llm_calls,
output_tokens=output_tokens,
replans=replan_count,
)
)
|