File size: 14,624 Bytes
f02fdcc | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 | """
Workflow Engine — executes a WorkflowConfig graph node-by-node.
- Dependency resolution queue (Flowise AgentFlow V2 pattern)
- Saves a checkpoint BEFORE and AFTER each node (LangGraph pattern)
- Routes success/failure edges from outcome_judgment nodes
- Human approval gates pause execution and resume on external signal
- Fail-branch nodes isolate errors into error_message / error_type vars
"""
from __future__ import annotations
import asyncio
import uuid
from dataclasses import dataclass, field
from datetime import datetime, timezone
from typing import Any, Dict, List, Optional, Set
from .dsl import WorkflowConfig, NodeDef, NodeType, ErrorStrategy
from .checkpoint import CheckpointStore, ExecutionCheckpoint, checkpoint_store
from src.agents.registry import AgentRegistry
from src.agents.base import AgentContext
from src.agents.executor import AgentExecutor
from src.memory.experiential import ExperientialRepository
from src.memory.working import WorkingContextMemory
from src.memory.retriever import retrieve
from src.cognitive.loop import CognitiveLoop, CognitiveDecision
from src.cognitive.plugins.evaluation import OutcomeEvaluator
from src.config import settings
@dataclass
class WorkflowRun:
run_id: str = field(default_factory=lambda: str(uuid.uuid4()))
workflow_id: str = ""
thread_id: str = field(default_factory=lambda: str(uuid.uuid4()))
status: str = "pending" # pending | running | completed | failed | paused
flow_state: Dict[str, Any] = field(default_factory=dict)
outputs: Dict[str, Any] = field(default_factory=dict) # node_id → output
error: Optional[str] = None
paused_at: Optional[str] = None # node_id where human gate paused
created_at: datetime = field(default_factory=lambda: datetime.now(timezone.utc))
completed_at: Optional[datetime] = None
class WorkflowEngine:
def __init__(
self,
store: Optional[CheckpointStore] = None,
executor: Optional[AgentExecutor] = None,
repository: Optional[ExperientialRepository] = None,
):
self._store = store or checkpoint_store
self._executor = executor or AgentExecutor()
self._repository = repository or ExperientialRepository()
self._runs: Dict[str, WorkflowRun] = {}
self._resume_events: Dict[str, asyncio.Event] = {}
self._resume_data: Dict[str, Any] = {}
# ── Public API ────────────────────────────────────────────────────
async def run(
self,
config: WorkflowConfig,
inputs: Optional[Dict[str, Any]] = None,
) -> WorkflowRun:
run = WorkflowRun(workflow_id=config.workflow_id)
run.flow_state = {**config.flow_state, **(inputs or {})}
self._runs[run.run_id] = run
run.status = "running"
try:
await self._execute_graph(config, run)
except Exception as exc:
run.status = "failed"
run.error = str(exc)
finally:
if run.status == "running":
run.status = "completed"
run.completed_at = datetime.now(timezone.utc)
return run
def get_run(self, run_id: str) -> Optional[WorkflowRun]:
return self._runs.get(run_id)
def list_runs(self) -> List[WorkflowRun]:
return sorted(self._runs.values(), key=lambda r: r.created_at, reverse=True)
async def resume(self, run_id: str, data: Optional[Dict[str, Any]] = None) -> None:
"""Resume a paused workflow (human approval gate)."""
event = self._resume_events.get(run_id)
if event:
self._resume_data[run_id] = data or {}
event.set()
# ── Graph execution ───────────────────────────────────────────────
async def _execute_graph(self, config: WorkflowConfig, run: WorkflowRun) -> None:
node_map = {n.id: n for n in config.nodes}
edge_map = {n.id: [] for n in config.nodes}
for edge in config.edges:
edge_map[edge.source].append(edge)
# Dependency resolution — topological-like BFS
completed: Set[str] = set()
pending = [n for n in config.nodes if not n.depends_on]
step = 0
while pending:
# Find nodes whose dependencies are all met
ready = [n for n in pending if all(d in completed for d in n.depends_on)]
if not ready:
# Check if we're paused — wait for resume
if run.status == "paused":
break
break # deadlock guard
for node in ready:
pending.remove(node)
step += 1
# Checkpoint BEFORE node execution
pre_cp = ExecutionCheckpoint(
workflow_id=config.workflow_id,
run_id=run.run_id,
thread_id=run.thread_id,
node_id=node.id,
step=step,
state=dict(run.flow_state),
status="pending",
)
self._store.save(pre_cp)
# Execute node
output, error = await self._execute_node(node, run, config)
# Handle error strategies
if error:
output, error = await self._handle_error(node, run, error, config)
# Store output in run state
run.outputs[node.id] = output
run.flow_state[f"{{#{node.id}.output#}}"] = output
if error:
run.flow_state[f"{{#{node.id}.error#}}"] = error
# Commit checkpoint AFTER node execution
pre_cp.status = "committed"
pre_cp.pending = {}
completed.add(node.id)
# Resolve next nodes via edges (respecting conditions)
outcome = run.flow_state.get(f"__outcome_{node.id}")
for edge in edge_map.get(node.id, []):
target = node_map.get(edge.target)
if not target:
continue
if edge.condition in (None, "", "always"):
pending.append(target)
elif edge.condition == "success" and outcome == "success":
pending.append(target)
elif edge.condition == "failure" and outcome == "failure":
pending.append(target)
if run.status in ("paused", "failed"):
return
# ── Node dispatch ─────────────────────────────────────────────────
async def _execute_node(
self, node: NodeDef, run: WorkflowRun, config: WorkflowConfig
) -> tuple[Any, Optional[str]]:
try:
if node.type == NodeType.START:
return run.flow_state, None
elif node.type == NodeType.MEMORY_RETRIEVAL:
objective = run.flow_state.get("message", run.flow_state.get("objective", ""))
working = WorkingContextMemory()
results = retrieve(objective, [], top_k=node.config.get("top_k", 5))
exp_results = self._repository.retrieve(objective, top_k=node.config.get("top_k", 5))
run.flow_state["__memory_context"] = [
{"title": r.item.title, "content": r.item.content, "score": r.relevance_score}
for r in exp_results
]
return run.flow_state["__memory_context"], None
elif node.type == NodeType.AGENT:
agent_type = node.config.get("agent_type", "assistant")
kwargs = {**node.config.get("kwargs", {})}
# Resolve DSL variable references: {{#node_id.field#}}
kwargs.update({
k: run.flow_state.get(v, v) if isinstance(v, str) and v.startswith("{{#") else v
for k, v in kwargs.items()
})
kwargs["message"] = run.flow_state.get("message", "")
try:
agent = AgentRegistry.create(agent_type)
context = AgentContext(metadata={"workflow_run_id": run.run_id})
# Chat agents need a session
from src.agents.chat import ChatSession
session = ChatSession(session_id=run.thread_id)
session.add("user", kwargs["message"])
result = await self._executor.run(agent, context, session=session, **kwargs)
output = result.output
run.flow_state[f"__outcome_{node.id}"] = (
"success" if result.status.value == "completed" else "failure"
)
return output, None
except Exception as exc:
run.flow_state[f"__outcome_{node.id}"] = "failure"
return None, str(exc)
elif node.type == NodeType.OUTCOME_JUDGMENT:
# Evaluate the most recent agent output
agent_outputs = {k: v for k, v in run.outputs.items() if v is not None}
last_output = list(agent_outputs.values())[-1] if agent_outputs else None
evaluator = OutcomeEvaluator(self._repository)
eval_result = evaluator.judge(
objective=run.flow_state.get("message", ""),
output=last_output,
error=run.flow_state.get("__last_error"),
)
run.flow_state[f"__outcome_{node.id}"] = eval_result.outcome.value
run.flow_state["__last_eval"] = {
"outcome": eval_result.outcome.value,
"score": eval_result.score,
"reflection": eval_result.reflection,
}
return run.flow_state["__last_eval"], None
elif node.type == NodeType.SAFLA_CONSOLIDATION:
branch = node.config.get("branch", "both")
last_eval = run.flow_state.get("__last_eval", {})
outcome_str = last_eval.get("outcome", "success")
from src.memory.models import MemoryOutcome
outcome = MemoryOutcome(outcome_str)
evaluator = OutcomeEvaluator(self._repository)
eval_result = evaluator.consolidate(
eval_result=type("E", (), {
"outcome": outcome,
"score": last_eval.get("score", 1.0),
"reflection": last_eval.get("reflection", ""),
"new_strategy_id": None,
"new_guardrail_ids": [],
})(),
objective=run.flow_state.get("message", ""),
output_summary=str(list(run.outputs.values())[-1] if run.outputs else ""),
)
return {
"new_strategy_id": eval_result.new_strategy_id,
"new_guardrail_ids": eval_result.new_guardrail_ids,
}, None
elif node.type == NodeType.HUMAN_APPROVAL:
run.status = "paused"
run.paused_at = node.id
event = asyncio.Event()
self._resume_events[run.run_id] = event
prompt = node.config.get("prompt", "Human review required.")
timeout = node.config.get("timeout", settings.agent_timeout)
try:
await asyncio.wait_for(event.wait(), timeout=timeout)
resume_data = self._resume_data.pop(run.run_id, {})
approved = resume_data.get("approved", True)
run.status = "running"
run.flow_state[f"__outcome_{node.id}"] = "success" if approved else "failure"
return {"approved": approved, "feedback": resume_data.get("feedback", "")}, None
except asyncio.TimeoutError:
run.flow_state[f"__outcome_{node.id}"] = "failure"
return None, f"Human approval gate timed out after {timeout}s."
elif node.type == NodeType.CONDITION:
expr = node.config.get("expression", "true")
result = bool(eval(expr, {"state": run.flow_state})) # noqa: S307 – internal DSL only
run.flow_state[f"__outcome_{node.id}"] = "success" if result else "failure"
return {"result": result}, None
elif node.type == NodeType.FAIL_BRANCH:
error_msg = run.flow_state.get("__last_error", "Unknown error")
error_type = type(error_msg).__name__ if not isinstance(error_msg, str) else "ExecutionError"
return {
"error_message": error_msg,
"error_type": error_type,
}, None
return None, f"Unknown node type: {node.type}"
except Exception as exc:
return None, str(exc)
async def _handle_error(
self, node: NodeDef, run: WorkflowRun, error: str, config: WorkflowConfig
) -> tuple[Any, Optional[str]]:
run.flow_state["__last_error"] = error
if not node.on_error:
return None, error
strategy = node.on_error.strategy
if strategy == ErrorStrategy.DEFAULT_VALUE:
return node.on_error.default_value, None
elif strategy == ErrorStrategy.ABORT:
run.status = "failed"
run.error = error
return None, error
elif strategy == ErrorStrategy.FAIL_BRANCH:
# Route to fail_branch nodes is handled by edge conditions
return None, None
elif strategy == ErrorStrategy.RETRY:
# Simple inline retry
for attempt in range(1, node.on_error.retry_limit + 1):
output, retry_err = await self._execute_node(node, run, config)
if not retry_err:
return output, None
return None, f"All {node.on_error.retry_limit} retries failed: {error}"
return None, error
# Singleton engine
workflow_engine = WorkflowEngine()
|