Spaces:
Running
Running
File size: 35,700 Bytes
9f6a4bd e99df76 691dc7c 2415446 8e438ab 691dc7c 261c237 2415446 691dc7c 2e7b8ca 2415446 691dc7c 2415446 691dc7c 2415446 a9e27c0 691dc7c 2415446 691dc7c 2e7b8ca 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c a9e27c0 691dc7c a1bab2d 8e438ab 2415446 2e7b8ca 2415446 261c237 691dc7c 261c237 691dc7c 261c237 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 f7872b8 2415446 691dc7c f7872b8 691dc7c f7872b8 691dc7c f7872b8 691dc7c 2415446 691dc7c 2415446 a1bab2d 2415446 691dc7c 2415446 a1bab2d 261c237 2415446 a9e27c0 691dc7c a1bab2d 261c237 691dc7c 2415446 691dc7c 2415446 2e7b8ca 691dc7c 2415446 a9e27c0 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 a1bab2d 691dc7c 2415446 8e438ab 2415446 8e438ab a1bab2d 691dc7c 2415446 a1bab2d 691dc7c 2415446 2e7b8ca 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2e7b8ca 691dc7c 6dc304d 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c 2e7b8ca 9f6a4bd 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c 6dc304d 9f6a4bd 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c 2e7b8ca 2415446 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c 2415446 a1bab2d 691dc7c a1bab2d 691dc7c 2e7b8ca 691dc7c 2e7b8ca 691dc7c a1bab2d 2e7b8ca 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c a9e27c0 691dc7c 8e438ab 2415446 691dc7c 6dc304d 2e7b8ca 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 2e7b8ca 691dc7c 2415446 691dc7c 2415446 691dc7c a9e27c0 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 691dc7c 2415446 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 | from __future__ import annotations
"""Single-owner provider generations and progressively available model metadata."""
import asyncio
from collections.abc import Awaitable, Callable, Iterable
from dataclasses import dataclass, field, replace
from functools import partial
from typing import Protocol
from loguru import logger
from free_claude_code.application.errors import ApplicationUnavailableError
from free_claude_code.application.model_metadata import (
ProviderModelInfo,
ProviderModelRefreshResult,
)
from free_claude_code.application.ports import ModelCatalogPort, ModelCatalogSnapshot
from free_claude_code.application.readiness import InitializationWait
from free_claude_code.config.model_refs import split_provider_model_ref
from free_claude_code.config.provider_catalog import PROVIDER_CATALOG
from free_claude_code.config.settings import Settings
from free_claude_code.core.async_tasks import run_sync_owned
from free_claude_code.core.json_types import JsonObject
from free_claude_code.core.token_estimation import initialize_token_estimation
from free_claude_code.core.trace import trace_event
from free_claude_code.providers.admission_policy import ProviderAdmissionLimits
from free_claude_code.providers.admission_registry import ProviderAdmissionRegistry
from free_claude_code.providers.base import BaseProvider
from free_claude_code.providers.history_negotiation import HistoryModeCache
from free_claude_code.providers.model_listing import model_infos_from_ids
from free_claude_code.providers.runtime.discovery import (
_provider_query_failure_reason,
is_optional_local_failure,
is_optional_local_provider,
model_cache_provider_ids_for_settings,
model_list_provider_ids_for_settings,
referenced_provider_ids,
)
from free_claude_code.providers.runtime.model_cache import ProviderModelCache
from free_claude_code.providers.runtime.runtime import ProviderRuntime
ProviderRuntimeFactory = Callable[
[Settings, ProviderAdmissionRegistry, HistoryModeCache], ProviderRuntime
]
ConnectedProviderIds = Callable[[], tuple[str, ...]]
CommitConfig = Callable[[], Awaitable[None]]
LOCAL_DISCOVERY_TIMEOUT_SECONDS = 5.0
class ModelCatalogPublisher(Protocol):
"""Publish a frozen model inventory without reading loop-owned mutable state."""
def publish(self, runtime: ModelCatalogPort) -> None: ...
@dataclass(slots=True, eq=False)
class _ProviderGeneration:
generation_id: int
settings: Settings
runtime: ProviderRuntime
cache: ProviderModelCache
active_leases: int = 0
retired: bool = False
closed: bool = False
drained: asyncio.Event = field(default_factory=asyncio.Event)
replaced: asyncio.Event = field(default_factory=asyncio.Event)
cleanup_task: asyncio.Task[bool] | None = None
catalog_tasks: dict[str, asyncio.Task[ProviderModelRefreshResult]] = field(
default_factory=dict
)
initialized: set[str] = field(default_factory=set)
initial_ids: tuple[str, ...] = ()
catalog_started: bool = False
initial_complete: bool = False
refresh_task: asyncio.Task[ProviderModelRefreshResult] | None = None
file_state: str = "starting"
def __post_init__(self) -> None:
for definition in self.settings.custom_providers:
if definition.model_ids:
self.cache.cache_model_infos(
definition.provider_id, model_infos_from_ids(definition.model_ids)
)
self.initialized.add(definition.provider_id)
self.drained.set()
def snapshot(self) -> ModelCatalogSnapshot:
return ModelCatalogSnapshot(
self.settings, self.cache.cached_prefixed_model_infos()
)
class ProviderGenerationLease:
"""Retain one generation and freeze metadata only after its provider is ready."""
def __init__(
self, manager: ProviderRuntimeManager, generation: _ProviderGeneration
) -> None:
self._manager = manager
self._generation = generation
self._wait = InitializationWait()
self._model_infos: dict[str, dict[str, ProviderModelInfo]] = {}
self._released = False
@property
def generation_id(self) -> int:
return self._generation.generation_id
@property
def settings(self) -> Settings:
return self._generation.settings
def is_provider_cached(self, provider_id: str, api_key_number: int = 1) -> bool:
return self._generation.runtime.is_cached(provider_id, api_key_number)
async def wait_for_token_estimation(self) -> None:
self._manager._ensure_open()
await self._wait.wait(self._manager._ensure_encoder())
async def resolve_provider(
self, provider_id: str, api_key_number: int = 1
) -> BaseProvider:
self._manager._ensure_open()
generation = self._generation
if provider_id not in generation.initialized:
task = self._manager._catalog_task(generation, provider_id)
await self._wait.wait(task)
if generation.runtime.is_cached(provider_id, api_key_number):
provider = await generation.runtime.resolve_provider(
provider_id, api_key_number
)
else:
waiting = asyncio.create_task(
generation.runtime.resolve_provider(provider_id, api_key_number)
)
try:
provider = await self._wait.wait(waiting)
finally:
waiting.cancel()
await asyncio.gather(waiting, return_exceptions=True)
if provider_id not in self._model_infos:
self._model_infos[provider_id] = {
info.model_id: replace(info, model_id=f"{provider_id}/{info.model_id}")
for info in generation.cache.provider_infos(provider_id)
}
return provider
def model_info(self, provider_id: str, model_id: str) -> ProviderModelInfo | None:
return self._model_infos.get(provider_id, {}).get(model_id)
async def release(self) -> None:
if not self._released:
self._released = True
await self._manager._release(self._generation)
async def __aenter__(self) -> ProviderGenerationLease:
return self
async def __aexit__(self, *_exc: object) -> None:
await self.release()
class ProviderRuntimeManager:
"""Own provider lifetimes, shared initialization and current catalog publication."""
def __init__(
self,
settings: Settings,
*,
runtime_factory: ProviderRuntimeFactory = ProviderRuntime,
connected_provider_ids: ConnectedProviderIds = tuple,
model_catalog_publisher: ModelCatalogPublisher | None = None,
) -> None:
self._runtime_factory = runtime_factory
self._history_cache = HistoryModeCache()
self._admission_registry = ProviderAdmissionRegistry(
ProviderAdmissionLimits.from_settings(settings)
)
self._connected_provider_ids = connected_provider_ids
self._model_catalog_publisher = model_catalog_publisher
self._catalog_changed: Callable[[], None] | None = None
self._replace_lock = asyncio.Lock()
self._publication_lock = asyncio.Lock()
self._close_lock = asyncio.Lock()
self._next_generation_id = 2
self._retired: dict[int, _ProviderGeneration] = {}
self._unpublished: set[ProviderRuntime] = set()
self._publication_task: asyncio.Task[None] | None = None
self._published_version: tuple[int, int] | None = None
self._encoder_task: asyncio.Task[object] | None = None
self._catalog_revision = 0
self._closing = False
self._closed = False
self._current = _ProviderGeneration(
generation_id=1,
settings=settings,
runtime=runtime_factory(
settings, self._admission_registry, self._history_cache
),
cache=ProviderModelCache(
model_cache_provider_ids_for_settings(
settings, connected_provider_ids()
)
),
)
self._trace_published(self._current, previous=None, reason="startup")
@property
def current_generation_id(self) -> int:
return self._current.generation_id
def set_catalog_changed_callback(self, callback: Callable[[], None] | None) -> None:
"""Register a nonblocking notification for application-owned consumers."""
self._catalog_changed = callback
def _ensure_open(self) -> None:
if self._closing or self._closed:
raise ApplicationUnavailableError("Provider runtime is shutting down.")
def _ensure_encoder(self) -> asyncio.Task[object]:
if self._encoder_task is None:
self._encoder_task = asyncio.create_task(
run_sync_owned(initialize_token_estimation)
)
return self._encoder_task
async def acquire(self) -> ProviderGenerationLease:
self._ensure_open()
generation = self._current
generation.active_leases += 1
generation.drained.clear()
return ProviderGenerationLease(self, generation)
def current_settings(self) -> Settings:
return self._current.settings
def _synchronize_model_cache_scope(self) -> None:
self._current.cache.set_available_providers(
model_cache_provider_ids_for_settings(
self._current.settings, self._connected_provider_ids()
)
)
def cached_model_ids(self) -> dict[str, frozenset[str]]:
self._synchronize_model_cache_scope()
return self._current.cache.cached_model_ids()
def cached_model_info(
self, provider_id: str, model_id: str
) -> ProviderModelInfo | None:
self._synchronize_model_cache_scope()
return self._current.cache.cached_model_info(provider_id, model_id)
def cached_prefixed_model_infos(self) -> tuple[ProviderModelInfo, ...]:
self._synchronize_model_cache_scope()
return self._current.cache.cached_prefixed_model_infos()
def catalog_snapshot(self) -> ModelCatalogSnapshot:
self._ensure_open()
self._synchronize_model_cache_scope()
return self._current.snapshot()
def cache_model_infos(
self, provider_id: str, model_infos: Iterable[ProviderModelInfo]
) -> None:
self._cache_result(self._current, provider_id, model_infos)
def _cache_result(
self,
generation: _ProviderGeneration,
provider_id: str,
infos: Iterable[ProviderModelInfo],
) -> None:
before = generation.cache.provider_infos(provider_id)
generation.cache.cache_model_infos(provider_id, infos)
if generation is self._current and before != generation.cache.provider_infos(
provider_id
):
self._catalog_revision += 1
self._catalog_updated()
def _catalog_task(
self,
generation: _ProviderGeneration,
provider_id: str,
*,
refresh: bool = False,
) -> asyncio.Task[ProviderModelRefreshResult]:
task = generation.catalog_tasks.get(provider_id)
if task is None or (refresh and task.done()):
task = asyncio.create_task(self._discover_provider(generation, provider_id))
generation.catalog_tasks[provider_id] = task
return task
async def _discover_provider(
self, generation: _ProviderGeneration, provider_id: str
) -> ProviderModelRefreshResult:
definition = generation.settings.custom_provider(provider_id)
if definition is not None and definition.model_ids:
generation.initialized.add(provider_id)
return ProviderModelRefreshResult(refreshed_provider_ids=(provider_id,))
try:
provider = await generation.runtime.resolve_provider(provider_id)
from free_claude_code.providers.admission import (
quiet_discovery_recovery_logs,
)
descriptor = PROVIDER_CATALOG.get(provider_id)
timeout = (
LOCAL_DISCOVERY_TIMEOUT_SECONDS
if descriptor is not None and descriptor.local
else None
)
with quiet_discovery_recovery_logs(
is_optional_local_provider(generation.settings, provider_id)
):
async with asyncio.timeout(timeout):
infos = await provider.list_model_infos()
except asyncio.CancelledError:
raise
except Exception as exc:
message = _provider_query_failure_reason(exc)
log = (
logger.debug
if is_optional_local_failure(generation.settings, provider_id, message)
else logger.warning
)
log(
"Provider model discovery skipped: provider={} reason={}",
provider_id,
message,
)
result = ProviderModelRefreshResult(
failed_provider_ids=(provider_id,),
failure_messages={provider_id: message},
)
else:
self._cache_result(generation, provider_id, infos)
logger.info(
"Provider model discovery cached: provider={} models={}",
provider_id,
len(infos),
)
result = ProviderModelRefreshResult(refreshed_provider_ids=(provider_id,))
generation.initialized.add(provider_id)
return result
def _start_pass(
self, generation: _ProviderGeneration, *, refresh: bool = False
) -> asyncio.Task[ProviderModelRefreshResult]:
if not generation.catalog_started:
generation.catalog_started = True
generation.initial_ids = tuple(
dict.fromkeys(
(
*referenced_provider_ids(generation.settings),
*model_list_provider_ids_for_settings(
generation.settings, self._connected_provider_ids()
),
)
)
)
if generation is self._current:
self._catalog_updated()
elif (
refresh
and generation is self._current
and self._published_version
!= (generation.generation_id, self._catalog_revision)
):
self._request_publication()
if generation.refresh_task is not None and not generation.refresh_task.done():
return generation.refresh_task
for provider_id in generation.initial_ids:
self._catalog_task(generation, provider_id, refresh=refresh)
generation.refresh_task = asyncio.create_task(self._finish_pass(generation))
return generation.refresh_task
async def _finish_pass(
self, generation: _ProviderGeneration
) -> ProviderModelRefreshResult:
while True:
tasks = tuple(
self._catalog_task(generation, provider_id)
for provider_id in generation.initial_ids
)
results = await asyncio.gather(*tasks, return_exceptions=True)
if tasks != tuple(
generation.catalog_tasks[provider_id]
for provider_id in generation.initial_ids
):
continue
for result in results:
if isinstance(result, asyncio.CancelledError):
raise result
generation.initial_complete = True
if generation.initial_complete and tasks == tuple(
generation.catalog_tasks[provider_id]
for provider_id in generation.initial_ids
):
break
failures = {
provider: message
for result in results
if isinstance(result, ProviderModelRefreshResult)
for provider, message in result.failure_messages.items()
}
failed_ids = tuple(
provider
for result in results
if isinstance(result, ProviderModelRefreshResult)
for provider in result.failed_provider_ids
if not is_optional_local_failure(
generation.settings, provider, failures.get(provider)
)
)
return ProviderModelRefreshResult(
refreshed_provider_ids=tuple(
provider
for result in results
if isinstance(result, ProviderModelRefreshResult)
for provider in result.refreshed_provider_ids
),
failed_provider_ids=failed_ids,
failure_messages={provider: failures[provider] for provider in failed_ids},
)
def start_model_list_refresh(self) -> None:
self._ensure_open()
self._ensure_encoder()
self._start_pass(self._current)
async def refresh_model_list_cache(self) -> ProviderModelRefreshResult:
self._ensure_open()
async with self._replace_lock:
task = self._start_pass(self._current, refresh=True)
return await asyncio.shield(task)
async def refresh_provider(self, provider_id: str) -> ProviderModelRefreshResult:
self._ensure_open()
generation = self._current
generation.active_leases += 1
generation.drained.clear()
try:
task = self._catalog_task(generation, provider_id, refresh=True)
return await asyncio.shield(task)
finally:
await self._release(generation)
async def wait_for_model_catalog(self, model: str) -> ModelCatalogSnapshot:
"""Wait only for a selection's metadata, following settings replacement."""
wait = InitializationWait()
while True:
snapshot = self.catalog_snapshot()
try:
provider_id, _ = split_provider_model_ref(model)
except ValueError:
return snapshot
generation = self._current
if provider_id not in (
*referenced_provider_ids(generation.settings),
*model_list_provider_ids_for_settings(
generation.settings, self._connected_provider_ids()
),
):
return snapshot
if provider_id not in generation.initialized:
try:
await self._wait_for_catalog_work(
generation, self._catalog_task(generation, provider_id), wait
)
except ApplicationUnavailableError:
if generation is self._current:
raise
continue
if generation is self._current:
return self.catalog_snapshot()
async def _wait_for_catalog_work[T](
self,
generation: _ProviderGeneration,
task: asyncio.Task[T],
wait: InitializationWait,
) -> None:
if generation is not self._current:
return
if task.done():
await wait.wait(task)
return
replaced = asyncio.create_task(generation.replaced.wait())
settled = asyncio.create_task(
asyncio.wait((task, replaced), return_when=asyncio.FIRST_COMPLETED)
)
try:
await wait.wait(settled)
if generation is self._current:
await wait.wait(task)
finally:
# These waits belong to the caller; discovery belongs to its generation.
settled.cancel()
replaced.cancel()
await asyncio.gather(settled, replaced, return_exceptions=True)
async def wait_for_catalog_file(self, wait: InitializationWait) -> tuple[int, int]:
while True:
self._ensure_open()
generation = self._current
version = (generation.generation_id, self._catalog_revision)
if self._published_version != version:
try:
await self._wait_for_catalog_work(
generation, self._request_publication(), wait
)
except ApplicationUnavailableError:
if generation is self._current:
raise
continue
if generation is not self._current:
continue
if generation.file_state == "failed":
raise ApplicationUnavailableError(
"Could not publish the Codex model catalog. Check file permissions and retry."
)
if self._published_version == (
generation.generation_id,
self._catalog_revision,
):
return self._published_version
def catalog_status(self) -> JsonObject:
generation = self._current
providers: JsonObject = {}
errors: JsonObject = {}
for provider_id in dict.fromkeys(
(*generation.initial_ids, *generation.catalog_tasks)
):
task = generation.catalog_tasks.get(provider_id)
if task is None or not task.done():
state = "starting"
elif task.cancelled() or task.exception() is not None:
state = "failed"
else:
result = task.result()
message = result.failure_messages.get(provider_id)
state = "failed" if result.failed_provider_ids else "ready"
if is_optional_local_failure(generation.settings, provider_id, message):
state = "unavailable"
elif message:
errors[provider_id] = message
providers[provider_id] = state
return {
"generation_id": generation.generation_id,
"catalog_revision": self._catalog_revision,
"catalog": "ready" if generation.initial_complete else "starting",
"catalog_file": generation.file_state,
"providers": providers,
"provider_errors": errors,
}
async def connected_provider_changed(
self, provider_id: str, *, connected: bool
) -> ProviderModelRefreshResult:
async with self._replace_lock, self._publication_lock:
self._ensure_open()
generation = self._current
self._start_pass(generation)
generation.initial_ids = tuple(
p for p in generation.initial_ids if p != provider_id
)
task = generation.catalog_tasks.pop(provider_id, None)
if task is not None and not task.done():
task.cancel()
await asyncio.gather(task, return_exceptions=True)
generation.initialized.discard(provider_id)
generation.cache.remove_provider(provider_id)
if connected:
generation.cache.add_provider(provider_id)
generation.initial_ids += (provider_id,)
self._catalog_task(generation, provider_id)
generation.initial_complete = False
generation.file_state = "starting"
self._catalog_revision += 1
self._start_pass(generation)
self._catalog_updated()
return ProviderModelRefreshResult()
def _catalog_updated(self) -> None:
if self._closing or not self._current.catalog_started:
return
self._request_publication()
if self._catalog_changed is not None and not self._closing:
self._catalog_changed()
def _request_publication(self) -> asyncio.Task[None]:
self._current.file_state = "starting"
if self._publication_task is None or self._publication_task.done():
self._publication_task = asyncio.create_task(
self._publish_catalog(), name="fcc-model-catalog"
)
return self._publication_task
async def _publish_catalog(self) -> None:
while not self._closing:
async with self._publication_lock:
if self._closing:
return
generation = self._current
version = (generation.generation_id, self._catalog_revision)
try:
publisher = self._model_catalog_publisher
if publisher is not None and self._published_version != version:
snapshot = generation.snapshot()
await run_sync_owned(partial(publisher.publish, snapshot))
except asyncio.CancelledError:
raise
except Exception as exc:
generation.file_state = "failed"
logger.warning(
"Model catalog publication failed: exc_type={}",
type(exc).__name__,
)
if version[1] == self._catalog_revision:
return
else:
self._published_version = version
if version[1] == self._catalog_revision:
generation.file_state = "ready"
return
generation.file_state = "starting"
async def replace(
self, settings: Settings, *, commit: CommitConfig, reason: str = "admin_apply"
) -> int:
async with self._replace_lock:
self._ensure_open()
await self._retry_unpublished_cleanup()
limits = ProviderAdmissionLimits.from_settings(settings)
candidate_id = self._next_generation_id
candidate_runtime: ProviderRuntime | None = None
try:
candidate_runtime = self._runtime_factory(
settings, self._admission_registry, self._history_cache
)
await commit()
except BaseException as exc:
trace_event(
stage="runtime",
event="provider_generation.replace_failed",
source="runtime",
current_generation_id=self._current.generation_id,
candidate_generation_id=candidate_id,
reason=reason,
exc_type=type(exc).__name__,
)
if candidate_runtime is not None:
await self._cleanup_unpublished(candidate_runtime)
raise
self._next_generation_id += 1
assert candidate_runtime is not None
async with self._publication_lock:
previous = self._current
cache = previous.cache.copy(
model_cache_provider_ids_for_settings(
settings, self._connected_provider_ids()
)
)
for provider_id in cache.cached_model_ids():
descriptor = PROVIDER_CATALOG.get(provider_id)
if (
descriptor is not None
and descriptor.local
and any(
attr is not None
and getattr(previous.settings, attr)
!= getattr(settings, attr)
for attr in (
descriptor.base_url_attr,
descriptor.proxy_attr,
)
)
):
cache.remove_provider(provider_id)
cache.add_provider(provider_id)
for definition in settings.custom_providers:
old = previous.settings.custom_provider(definition.provider_id)
if old is None or not definition.same_inventory(old):
cache.remove_provider(definition.provider_id)
cache.add_provider(definition.provider_id)
candidate = _ProviderGeneration(
generation_id=candidate_id,
settings=settings,
runtime=candidate_runtime,
cache=cache,
)
self._admission_registry.reconfigure(limits)
self._current = candidate
self._catalog_revision += 1
previous.retired = True
previous.replaced.set()
self._retired[previous.generation_id] = previous
self._trace_published(candidate, previous=previous, reason=reason)
self._trace_retired(previous, reason=reason)
self._start_pass(candidate)
if previous.active_leases == 0:
await self._close_generation(previous, forced=False)
return candidate.generation_id
def begin_shutdown(self) -> None:
self._closing = True
for generation in (self._current, *self._retired.values()):
if generation.refresh_task is not None:
generation.refresh_task.cancel()
for task in generation.catalog_tasks.values():
task.cancel()
async def _cancel_generation_work(self, generation: _ProviderGeneration) -> None:
tasks = [*generation.catalog_tasks.values()]
if generation.refresh_task is not None:
tasks.append(generation.refresh_task)
for task in tasks:
if not task.done():
task.cancel()
await asyncio.gather(*tasks, return_exceptions=True)
async def close(self) -> None:
async with self._close_lock:
if self._closed:
return
self.begin_shutdown()
async with self._replace_lock:
current = self._current
if not current.retired:
current.retired = True
self._retired[current.generation_id] = current
self._trace_retired(current, reason="shutdown")
generations = tuple(self._retired.values())
await asyncio.gather(
*(self._cancel_generation_work(g) for g in generations)
)
if self._publication_task is not None:
await asyncio.gather(self._publication_task, return_exceptions=True)
if self._encoder_task is not None:
self._encoder_task.cancel()
await asyncio.gather(self._encoder_task, return_exceptions=True)
await asyncio.gather(*(g.drained.wait() for g in generations))
generation_results = await asyncio.gather(
*(self._close_generation(g, forced=False) for g in generations)
)
unpublished_closed = await self._retry_unpublished_cleanup()
if not all(generation_results) or not unpublished_closed:
raise RuntimeError("One or more provider runtimes failed to close.")
self._current.cache.clear()
self._admission_registry.close()
self._history_cache.clear()
self._closed = True
def _prune_admission(self) -> None:
runtimes = (
self._current.runtime,
*(generation.runtime for generation in self._retired.values()),
*self._unpublished,
)
self._admission_registry.retain_custom(
definition.provider_id
for runtime in runtimes
for definition in runtime.settings.custom_providers
)
async def _release(self, generation: _ProviderGeneration) -> None:
if generation.active_leases <= 0:
return
generation.active_leases -= 1
if generation.active_leases == 0:
generation.drained.set()
if generation.retired and not self._closing:
await self._close_generation(generation, forced=False)
async def _cleanup_unpublished(self, runtime: ProviderRuntime) -> bool:
self._unpublished.add(runtime)
try:
await runtime.cleanup()
except asyncio.CancelledError:
raise
except Exception as exc:
logger.warning(
"Unpublished provider generation cleanup failed: exc_type={}",
type(exc).__name__,
)
return False
self._unpublished.discard(runtime)
self._prune_admission()
return True
async def _retry_unpublished_cleanup(self) -> bool:
all_closed = True
for runtime in tuple(self._unpublished):
if not await self._cleanup_unpublished(runtime):
all_closed = False
return all_closed
async def _close_generation(
self,
generation: _ProviderGeneration,
*,
forced: bool,
) -> bool:
if generation.closed:
return True
if generation.active_leases != 0:
return False
task = generation.cleanup_task
if task is None:
task = asyncio.create_task(
self._run_generation_cleanup(generation, forced=forced),
name=f"provider-generation-cleanup-{generation.generation_id}",
)
generation.cleanup_task = task
return await asyncio.shield(task)
async def _run_generation_cleanup(
self,
generation: _ProviderGeneration,
*,
forced: bool,
) -> bool:
task = asyncio.current_task()
try:
try:
await self._cancel_generation_work(generation)
await generation.runtime.cleanup()
except asyncio.CancelledError:
raise
except Exception as exc:
logger.warning(
"Provider generation cleanup failed: generation_id={} exc_type={}",
generation.generation_id,
type(exc).__name__,
)
return False
generation.closed = True
self._retired.pop(generation.generation_id, None)
self._prune_admission()
trace_event(
stage="runtime",
event="provider_generation.closed",
source="runtime",
generation_id=generation.generation_id,
active_leases=generation.active_leases,
forced=forced,
outcome="ok",
)
return True
finally:
if not generation.closed and generation.cleanup_task is task:
generation.cleanup_task = None
@staticmethod
def _trace_published(
generation: _ProviderGeneration,
*,
previous: _ProviderGeneration | None,
reason: str,
) -> None:
trace_event(
stage="runtime",
event="provider_generation.published",
source="runtime",
generation_id=generation.generation_id,
previous_generation_id=(
previous.generation_id if previous is not None else None
),
reason=reason,
)
@staticmethod
def _trace_retired(generation: _ProviderGeneration, *, reason: str) -> None:
trace_event(
stage="runtime",
event="provider_generation.retired",
source="runtime",
generation_id=generation.generation_id,
active_leases=generation.active_leases,
reason=reason,
)
|