File size: 13,114 Bytes
9abace2 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 | """
Sovereign LLM Engine — Parallel Swarm Coordinator
Part of SOVEREIGN PYTHON LLM ENGINE
Coordinates N concurrent async agent tasks using asyncio primitives.
Supports fan_out, map_reduce, and race patterns with configurable
concurrency limits, per-task timeouts, and result aggregation.
"""
from __future__ import annotations
import asyncio
import logging
import time
from dataclasses import dataclass, field
from typing import Any, Callable, Coroutine, TypeVar
logger = logging.getLogger("sovereign.swarm")
T = TypeVar("T")
R = TypeVar("R")
# Type aliases for readability
AgentCallable = Callable[..., Coroutine[Any, Any, Any]]
Mapper = Callable[[Any], Coroutine[Any, Any, Any]]
Reducer = Callable[[list[Any]], Coroutine[Any, Any, Any]]
# ==========================================
# Result Types
# ==========================================
@dataclass
class TaskOutcome:
"""Outcome of a single task within a swarm operation."""
index: int
success: bool
value: Any = None
error: str | None = None
latency_ms: float = 0.0
@dataclass
class SwarmResult:
"""
Aggregated result from a swarm operation.
Attributes:
results: Successful results in order of completion (or index for fan_out).
failures: Failed task outcomes.
latency_ms: Wall-clock time for the entire swarm operation.
winner: For race() operations — the first successful result value.
"""
results: list[Any]
failures: list[TaskOutcome]
latency_ms: float
winner: Any | None = None
@property
def success_count(self) -> int:
return len(self.results)
@property
def failure_count(self) -> int:
return len(self.failures)
@property
def total_count(self) -> int:
return self.success_count + self.failure_count
@property
def all_succeeded(self) -> bool:
return self.failure_count == 0
def __repr__(self) -> str:
return (
f"SwarmResult(success={self.success_count}, "
f"failures={self.failure_count}, "
f"latency_ms={self.latency_ms:.1f})"
)
# ==========================================
# Internal helpers
# ==========================================
async def _run_with_timeout(
coro: Coroutine[Any, Any, T],
timeout: float | None,
index: int,
semaphore: asyncio.Semaphore,
) -> TaskOutcome:
"""
Run a coroutine under a semaphore and optional timeout.
Returns a TaskOutcome regardless of success or failure.
"""
start = time.monotonic()
async with semaphore:
try:
if timeout is not None:
value = await asyncio.wait_for(coro, timeout=timeout)
else:
value = await coro
latency = (time.monotonic() - start) * 1000.0
return TaskOutcome(index=index, success=True, value=value, latency_ms=latency)
except asyncio.TimeoutError:
latency = (time.monotonic() - start) * 1000.0
return TaskOutcome(
index=index,
success=False,
error=f"Task {index} timed out after {timeout}s",
latency_ms=latency,
)
except Exception as exc:
latency = (time.monotonic() - start) * 1000.0
return TaskOutcome(
index=index,
success=False,
error=f"{type(exc).__name__}: {exc}",
latency_ms=latency,
)
# ==========================================
# Swarm
# ==========================================
class Swarm:
"""
Parallel swarm coordinator.
All operations respect a global concurrency limit (semaphore) and an
optional per-task timeout. Tasks that exceed the timeout are recorded
as failures rather than raising exceptions.
Args:
concurrency: Maximum number of tasks running simultaneously.
task_timeout: Per-task timeout in seconds (None = unlimited).
"""
def __init__(
self,
concurrency: int = 16,
task_timeout: float | None = 60.0,
) -> None:
if concurrency < 1:
raise ValueError("concurrency must be >= 1")
self._concurrency = concurrency
self._task_timeout = task_timeout
# ------------------------------------------------------------------
# fan_out
# ------------------------------------------------------------------
async def fan_out(
self,
task: Any,
agents: list[AgentCallable],
) -> SwarmResult:
"""
Dispatch the same task to every agent concurrently.
Each agent in `agents` is called as ``agent(task)``.
Results are returned in the same order as the agent list.
Failures do not block other agents.
Returns:
SwarmResult with results indexed to match agent positions.
"""
if not agents:
return SwarmResult(results=[], failures=[], latency_ms=0.0)
semaphore = asyncio.Semaphore(self._concurrency)
wall_start = time.monotonic()
coros = [
_run_with_timeout(agent(task), self._task_timeout, idx, semaphore)
for idx, agent in enumerate(agents)
]
outcomes: list[TaskOutcome] = await asyncio.gather(*coros)
wall_ms = (time.monotonic() - wall_start) * 1000.0
# Preserve original index order for results
ordered_results: list[Any] = [None] * len(agents)
failures: list[TaskOutcome] = []
for outcome in outcomes:
if outcome.success:
ordered_results[outcome.index] = outcome.value
else:
failures.append(outcome)
logger.warning(
"fan_out agent[%d] failed: %s", outcome.index, outcome.error
)
# Remove None slots for failed agents so callers get a clean list
success_results = [
ordered_results[i]
for i in range(len(agents))
if outcomes[i].success
]
return SwarmResult(
results=success_results,
failures=failures,
latency_ms=wall_ms,
)
# ------------------------------------------------------------------
# map_reduce
# ------------------------------------------------------------------
async def map_reduce(
self,
items: list[Any],
mapper: Mapper,
reducer: Reducer,
) -> SwarmResult:
"""
Map items through an async mapper concurrently, then reduce.
Each item in `items` is passed to ``mapper(item)`` concurrently.
Successful mapped values are collected and passed to ``reducer(values)``.
The reducer result is the single entry in SwarmResult.results.
Failures in individual map tasks are recorded; the reducer only
receives successfully mapped values. If all maps fail, the reducer
is not called and SwarmResult.results is empty.
Returns:
SwarmResult where results[0] is the reducer output (if any map succeeded).
"""
if not items:
reduced = await reducer([])
return SwarmResult(results=[reduced], failures=[], latency_ms=0.0)
semaphore = asyncio.Semaphore(self._concurrency)
wall_start = time.monotonic()
map_coros = [
_run_with_timeout(mapper(item), self._task_timeout, idx, semaphore)
for idx, item in enumerate(items)
]
map_outcomes: list[TaskOutcome] = await asyncio.gather(*map_coros)
mapped_values: list[Any] = []
failures: list[TaskOutcome] = []
for outcome in map_outcomes:
if outcome.success:
mapped_values.append(outcome.value)
else:
failures.append(outcome)
logger.warning(
"map_reduce mapper[%d] failed: %s", outcome.index, outcome.error
)
results: list[Any] = []
if mapped_values:
try:
reduced_value = await reducer(mapped_values)
results = [reduced_value]
except Exception as exc:
logger.exception("map_reduce reducer failed: %s", exc)
failures.append(
TaskOutcome(
index=-1,
success=False,
error=f"Reducer {type(exc).__name__}: {exc}",
)
)
wall_ms = (time.monotonic() - wall_start) * 1000.0
return SwarmResult(results=results, failures=failures, latency_ms=wall_ms)
# ------------------------------------------------------------------
# race
# ------------------------------------------------------------------
async def race(
self,
tasks: list[Coroutine[Any, Any, Any]],
) -> SwarmResult:
"""
Run tasks concurrently and return the first successful result.
All remaining tasks are cancelled once a winner is found.
If all tasks fail, SwarmResult.winner is None and all failures
are recorded.
Returns:
SwarmResult with winner set to the first successful value.
results contains only the winning value (or empty if all failed).
"""
if not tasks:
return SwarmResult(results=[], failures=[], latency_ms=0.0, winner=None)
semaphore = asyncio.Semaphore(self._concurrency)
wall_start = time.monotonic()
# Wrap each coroutine in a task
loop = asyncio.get_running_loop()
pending: set[asyncio.Task[TaskOutcome]] = set()
for idx, coro in enumerate(tasks):
t = loop.create_task(
_run_with_timeout(coro, self._task_timeout, idx, semaphore)
)
pending.add(t)
winner: Any | None = None
winner_found = False
failures: list[TaskOutcome] = []
while pending:
done, pending = await asyncio.wait(
pending, return_when=asyncio.FIRST_COMPLETED
)
for finished in done:
outcome: TaskOutcome = finished.result()
if outcome.success and not winner_found:
winner = outcome.value
winner_found = True
logger.debug(
"race winner: task[%d] in %.1fms",
outcome.index,
outcome.latency_ms,
)
# Cancel remaining tasks
for remaining in pending:
remaining.cancel()
# Drain cancelled tasks
if pending:
await asyncio.gather(*pending, return_exceptions=True)
pending = set()
break
elif not outcome.success:
failures.append(outcome)
wall_ms = (time.monotonic() - wall_start) * 1000.0
if winner_found:
return SwarmResult(
results=[winner],
failures=failures,
latency_ms=wall_ms,
winner=winner,
)
else:
return SwarmResult(
results=[],
failures=failures,
latency_ms=wall_ms,
winner=None,
)
# ------------------------------------------------------------------
# Batch helper
# ------------------------------------------------------------------
async def batch(
self,
coros: list[Coroutine[Any, Any, Any]],
) -> SwarmResult:
"""
Run an arbitrary list of coroutines concurrently with the swarm
concurrency and timeout settings. Results are in completion order.
This is a lower-level primitive used internally; fan_out and
map_reduce are preferred for agent patterns.
"""
if not coros:
return SwarmResult(results=[], failures=[], latency_ms=0.0)
semaphore = asyncio.Semaphore(self._concurrency)
wall_start = time.monotonic()
wrapped = [
_run_with_timeout(coro, self._task_timeout, idx, semaphore)
for idx, coro in enumerate(coros)
]
outcomes: list[TaskOutcome] = await asyncio.gather(*wrapped)
results = [o.value for o in outcomes if o.success]
failures = [o for o in outcomes if not o.success]
wall_ms = (time.monotonic() - wall_start) * 1000.0
return SwarmResult(results=results, failures=failures, latency_ms=wall_ms)
|