# SPDX-License-Identifier: Apache-2.0 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project import os import queue import signal import threading import time from collections import deque from collections.abc import Callable, Generator from concurrent.futures import Future from contextlib import ExitStack, contextmanager from inspect import isclass, signature from logging import DEBUG from typing import Any, TypeVar, cast import msgspec import zmq from vllm.config import ParallelConfig, VllmConfig from vllm.distributed import stateless_destroy_torch_distributed_process_group from vllm.envs import enable_envs_cache from vllm.logger import init_logger from vllm.logging_utils.dump_input import dump_engine_exception from vllm.lora.request import LoRARequest from vllm.multimodal import MULTIMODAL_REGISTRY from vllm.multimodal.cache import engine_receiver_cache_from_config from vllm.tasks import POOLING_TASKS, SupportedTask from vllm.transformers_utils.config import maybe_register_config_serialize_by_value from vllm.utils.gc_utils import ( freeze_gc_heap, maybe_attach_gc_debug_callback, ) from vllm.utils.hashing import get_hash_fn_by_name from vllm.utils.network_utils import make_zmq_socket from vllm.utils.system_utils import decorate_logs, set_process_title from vllm.v1.core.kv_cache_utils import ( BlockHash, generate_scheduler_kv_cache_config, get_kv_cache_configs, get_request_block_hasher, init_none_hash, ) from vllm.v1.core.sched.interface import SchedulerInterface from vllm.v1.core.sched.output import SchedulerOutput from vllm.v1.engine import ( EngineCoreOutputs, EngineCoreRequest, EngineCoreRequestType, ReconfigureDistributedRequest, ReconfigureRankType, UtilityOutput, UtilityResult, ) from vllm.v1.engine.utils import ( EngineHandshakeMetadata, EngineZmqAddresses, get_device_indices, ) from vllm.v1.executor import Executor from vllm.v1.kv_cache_interface import KVCacheConfig from vllm.v1.metrics.stats import SchedulerStats from vllm.v1.outputs import ModelRunnerOutput from vllm.v1.request import Request, RequestStatus from vllm.v1.serial_utils import MsgpackDecoder, MsgpackEncoder from vllm.v1.structured_output import StructuredOutputManager from vllm.version import __version__ as VLLM_VERSION logger = init_logger(__name__) POLLING_TIMEOUT_S = 2.5 HANDSHAKE_TIMEOUT_MINS = 5 _R = TypeVar("_R") # Return type for collective_rpc class EngineCore: """Inner loop of vLLM's Engine.""" def __init__( self, vllm_config: VllmConfig, executor_class: type[Executor], log_stats: bool, executor_fail_callback: Callable | None = None, ): # plugins need to be loaded at the engine/scheduler level too from vllm.plugins import load_general_plugins load_general_plugins() self.vllm_config = vllm_config if vllm_config.parallel_config.data_parallel_rank == 0: logger.info( "Initializing a V1 LLM engine (v%s) with config: %s", VLLM_VERSION, vllm_config, ) self.log_stats = log_stats # Setup Model. self.model_executor = executor_class(vllm_config) if executor_fail_callback is not None: self.model_executor.register_failure_callback(executor_fail_callback) self.available_gpu_memory_for_kv_cache = -1 # Setup KV Caches and update CacheConfig after profiling. num_gpu_blocks, num_cpu_blocks, kv_cache_config = self._initialize_kv_caches( vllm_config ) vllm_config.cache_config.num_gpu_blocks = num_gpu_blocks vllm_config.cache_config.num_cpu_blocks = num_cpu_blocks self.collective_rpc("initialize_cache", args=(num_gpu_blocks, num_cpu_blocks)) self.structured_output_manager = StructuredOutputManager(vllm_config) # Setup scheduler. Scheduler = vllm_config.scheduler_config.get_scheduler_cls() if len(kv_cache_config.kv_cache_groups) == 0: # Encoder models without KV cache don't support # chunked prefill. But do SSM models? logger.info("Disabling chunked prefill for model without KVCache") vllm_config.scheduler_config.enable_chunked_prefill = False scheduler_block_size = ( vllm_config.cache_config.block_size * vllm_config.parallel_config.decode_context_parallel_size ) self.scheduler: SchedulerInterface = Scheduler( vllm_config=vllm_config, kv_cache_config=kv_cache_config, structured_output_manager=self.structured_output_manager, include_finished_set=vllm_config.parallel_config.data_parallel_size > 1, log_stats=self.log_stats, block_size=scheduler_block_size, ) self.use_spec_decode = vllm_config.speculative_config is not None if self.scheduler.connector is not None: # type: ignore self.model_executor.init_kv_output_aggregator(self.scheduler.connector) # type: ignore self.mm_registry = mm_registry = MULTIMODAL_REGISTRY self.mm_receiver_cache = engine_receiver_cache_from_config( vllm_config, mm_registry ) # If a KV connector is initialized for scheduler, we want to collect # handshake metadata from all workers so the connector in the scheduler # will have the full context kv_connector = self.scheduler.get_kv_connector() if kv_connector is not None: # Collect and store KV connector xfer metadata from workers # (after KV cache registration) xfer_handshake_metadata = ( self.model_executor.get_kv_connector_handshake_metadata() ) if xfer_handshake_metadata: # xfer_handshake_metadata is list of dicts from workers # Each dict already has structure {tp_rank: metadata} # Merge all worker dicts into a single dict content: dict[int, Any] = {} for worker_dict in xfer_handshake_metadata: if worker_dict is not None: content.update(worker_dict) kv_connector.set_xfer_handshake_metadata(content) # Setup batch queue for pipeline parallelism. # Batch queue for scheduled batches. This enables us to asynchronously # schedule and execute batches, and is required by pipeline parallelism # to eliminate pipeline bubbles. self.batch_queue_size = self.model_executor.max_concurrent_batches self.batch_queue: ( deque[tuple[Future[ModelRunnerOutput], SchedulerOutput]] | None ) = None if self.batch_queue_size > 1: logger.info("Batch queue is enabled with size %d", self.batch_queue_size) self.batch_queue = deque(maxlen=self.batch_queue_size) self.ec_producer = ( vllm_config.ec_transfer_config is not None and vllm_config.ec_transfer_config.is_ec_producer ) self.is_pooling_model = vllm_config.model_config.runner_type == "pooling" self.request_block_hasher: Callable[[Request], list[BlockHash]] | None = None if vllm_config.cache_config.enable_prefix_caching or kv_connector is not None: caching_hash_fn = get_hash_fn_by_name( vllm_config.cache_config.prefix_caching_hash_algo ) init_none_hash(caching_hash_fn) self.request_block_hasher = get_request_block_hasher( scheduler_block_size, caching_hash_fn ) self.step_fn = ( self.step if self.batch_queue is None else self.step_with_batch_queue ) self.async_scheduling = vllm_config.scheduler_config.async_scheduling # Mark the startup heap as static so that it's ignored by GC. # Reduces pause times of oldest generation collections. freeze_gc_heap() def _initialize_kv_caches( self, vllm_config: VllmConfig ) -> tuple[int, int, KVCacheConfig]: start = time.time() # Get all kv cache needed by the model kv_cache_specs = self.model_executor.get_kv_cache_specs() has_kv_cache = any(kv_cache_spec for kv_cache_spec in kv_cache_specs) if has_kv_cache: if os.environ.get("VLLM_ELASTIC_EP_SCALE_UP_LAUNCH") == "1": dp_group = getattr(self, "dp_group", None) assert dp_group is not None self.available_gpu_memory_for_kv_cache = ( ParallelConfig.sync_kv_cache_memory_size(dp_group, -1) ) available_gpu_memory = [self.available_gpu_memory_for_kv_cache] * len( kv_cache_specs ) else: # Profiles the peak memory usage of the model to determine how # much memory can be allocated for kv cache. available_gpu_memory = self.model_executor.determine_available_memory() self.available_gpu_memory_for_kv_cache = available_gpu_memory[0] else: # Attention free models don't need memory for kv cache available_gpu_memory = [0] * len(kv_cache_specs) assert len(kv_cache_specs) == len(available_gpu_memory) kv_cache_configs = get_kv_cache_configs( vllm_config, kv_cache_specs, available_gpu_memory ) scheduler_kv_cache_config = generate_scheduler_kv_cache_config(kv_cache_configs) num_gpu_blocks = scheduler_kv_cache_config.num_blocks num_cpu_blocks = 0 # Initialize kv cache and warmup the execution self.model_executor.initialize_from_config(kv_cache_configs) elapsed = time.time() - start logger.info_once( "init engine (profile, create kv cache, warmup model) took %.2f seconds", elapsed, scope="local", ) return num_gpu_blocks, num_cpu_blocks, scheduler_kv_cache_config def get_supported_tasks(self) -> tuple[SupportedTask, ...]: return self.model_executor.supported_tasks def add_request(self, request: Request, request_wave: int = 0): """Add request to the scheduler. `request_wave`: indicate which wave of requests this is expected to belong to in DP case """ # Validate the request_id type. if not isinstance(request.request_id, str): raise TypeError( f"request_id must be a string, got {type(request.request_id)}" ) if pooling_params := request.pooling_params: supported_pooling_tasks = [ task for task in self.get_supported_tasks() if task in POOLING_TASKS ] if pooling_params.task not in supported_pooling_tasks: raise ValueError( f"Unsupported task: {pooling_params.task!r} " f"Supported tasks: {supported_pooling_tasks}" ) if request.kv_transfer_params is not None and ( not self.scheduler.get_kv_connector() ): logger.warning( "Got kv_transfer_params, but no KVConnector found. " "Disabling KVTransfer for this request." ) self.scheduler.add_request(request) def abort_requests(self, request_ids: list[str]): """Abort requests from the scheduler.""" # TODO: The scheduler doesn't really need to know the # specific finish reason, TBD whether we propagate that # (i.e. client-aborted vs stop criteria met). self.scheduler.finish_requests(request_ids, RequestStatus.FINISHED_ABORTED) @contextmanager def log_error_detail(self, scheduler_output: SchedulerOutput): """Execute the model and log detailed info on failure.""" try: yield except Exception as err: # We do not want to catch BaseException here since we're only # interested in dumping info when the exception is due to an # error from execute_model itself. # NOTE: This method is exception-free dump_engine_exception( self.vllm_config, scheduler_output, self.scheduler.make_stats() ) raise err def _log_err_callback(self, scheduler_output: SchedulerOutput): """Log error details of a future that's not expected to return a result.""" def callback(f, sched_output=scheduler_output): with self.log_error_detail(sched_output): result = f.result() assert result is None return callback def step(self) -> tuple[dict[int, EngineCoreOutputs], bool]: """Schedule, execute, and make output. Returns tuple of outputs and a flag indicating whether the model was executed. """ # Check for any requests remaining in the scheduler - unfinished, # or finished and not yet removed from the batch. if not self.scheduler.has_requests(): return {}, False scheduler_output = self.scheduler.schedule() future = self.model_executor.execute_model(scheduler_output, non_block=True) grammar_output = self.scheduler.get_grammar_bitmask(scheduler_output) with self.log_error_detail(scheduler_output): model_output = future.result() if model_output is None: model_output = self.model_executor.sample_tokens(grammar_output) engine_core_outputs = self.scheduler.update_from_output( scheduler_output, model_output ) return engine_core_outputs, scheduler_output.total_num_scheduled_tokens > 0 def post_step(self, model_executed: bool) -> None: # When using async scheduling we can't get draft token ids in advance, # so we update draft token ids in the worker process and don't # need to update draft token ids here. if not self.async_scheduling and self.use_spec_decode and model_executed: # Take the draft token ids. draft_token_ids = self.model_executor.take_draft_token_ids() if draft_token_ids is not None: self.scheduler.update_draft_token_ids(draft_token_ids) def step_with_batch_queue( self, ) -> tuple[dict[int, EngineCoreOutputs] | None, bool]: """Schedule and execute batches with the batch queue. Note that if nothing to output in this step, None is returned. The execution flow is as follows: 1. Try to schedule a new batch if the batch queue is not full. If a new batch is scheduled, directly return an empty engine core output. In other words, fulfilling the batch queue has a higher priority than getting model outputs. 2. If there is no new scheduled batch, meaning that the batch queue is full or no other requests can be scheduled, we block until the first batch in the job queue is finished. 3. Update the scheduler from the output. """ batch_queue = self.batch_queue assert batch_queue is not None # Try to schedule a new batch if the batch queue is not full, but # the scheduler may return an empty batch if all requests are scheduled. # Note that this is not blocking. assert len(batch_queue) < self.batch_queue_size model_executed = False deferred_scheduler_output = None if self.scheduler.has_requests(): scheduler_output = self.scheduler.schedule() exec_future = self.model_executor.execute_model( scheduler_output, non_block=True ) if not self.ec_producer: model_executed = scheduler_output.total_num_scheduled_tokens > 0 if self.is_pooling_model or not model_executed: # No sampling required (no requests scheduled). future = cast(Future[ModelRunnerOutput], exec_future) else: exec_future.add_done_callback(self._log_err_callback(scheduler_output)) if not scheduler_output.pending_structured_output_tokens: # We aren't waiting for any tokens, get any grammar output # and sample immediately. grammar_output = self.scheduler.get_grammar_bitmask( scheduler_output ) future = self.model_executor.sample_tokens( grammar_output, non_block=True ) else: # We need to defer sampling until we have processed the model output # from the prior step. deferred_scheduler_output = scheduler_output if not deferred_scheduler_output: # Add this step's future to the queue. batch_queue.appendleft((future, scheduler_output)) if ( model_executed and len(batch_queue) < self.batch_queue_size and not batch_queue[-1][0].done() ): # Don't block on next worker response unless the queue is full # or there are no more requests to schedule. return None, True elif not batch_queue: # Queue is empty. We should not reach here since this method should # only be called when the scheduler contains requests or the queue # is non-empty. return None, False # Block until the next result is available. future, scheduler_output = batch_queue.pop() with self.log_error_detail(scheduler_output): model_output = future.result() engine_core_outputs = self.scheduler.update_from_output( scheduler_output, model_output ) # NOTE(nick): We can either handle the deferred tasks here or save # in a field and do it immediately once step_with_batch_queue is # re-called. The latter slightly favors TTFT over TPOT/throughput. if deferred_scheduler_output: # We now have the tokens needed to compute the bitmask for the # deferred request. Get the bitmask and call sample tokens. grammar_output = self.scheduler.get_grammar_bitmask( deferred_scheduler_output ) future = self.model_executor.sample_tokens(grammar_output, non_block=True) batch_queue.appendleft((future, deferred_scheduler_output)) return engine_core_outputs, model_executed def shutdown(self): self.structured_output_manager.clear_backend() if self.model_executor: self.model_executor.shutdown() if self.scheduler: self.scheduler.shutdown() def profile(self, is_start: bool = True): self.model_executor.profile(is_start) def reset_mm_cache(self): # NOTE: Since this is mainly for debugging, we don't attempt to # re-sync the internal caches (P0 sender, P1 receiver) if self.scheduler.has_unfinished_requests(): logger.warning( "Resetting the multi-modal cache when requests are " "in progress may lead to desynced internal caches." ) # The cache either exists in EngineCore or WorkerWrapperBase if self.mm_receiver_cache is not None: self.mm_receiver_cache.clear_cache() self.model_executor.reset_mm_cache() def reset_prefix_cache(self): self.scheduler.reset_prefix_cache() def sleep(self, level: int = 1): self.model_executor.sleep(level) def wake_up(self, tags: list[str] | None = None): self.model_executor.wake_up(tags) def is_sleeping(self) -> bool: return self.model_executor.is_sleeping def execute_dummy_batch(self): self.model_executor.execute_dummy_batch() def add_lora(self, lora_request: LoRARequest) -> bool: return self.model_executor.add_lora(lora_request) def remove_lora(self, lora_id: int) -> bool: return self.model_executor.remove_lora(lora_id) def list_loras(self) -> set[int]: return self.model_executor.list_loras() def pin_lora(self, lora_id: int) -> bool: return self.model_executor.pin_lora(lora_id) def save_sharded_state( self, path: str, pattern: str | None = None, max_size: int | None = None, ) -> None: self.model_executor.save_sharded_state( path=path, pattern=pattern, max_size=max_size ) def collective_rpc( self, method: str | Callable[..., _R], timeout: float | None = None, args: tuple = (), kwargs: dict[str, Any] | None = None, ) -> list[_R]: return self.model_executor.collective_rpc(method, timeout, args, kwargs) def preprocess_add_request(self, request: EngineCoreRequest) -> tuple[Request, int]: """Preprocess the request. This function could be directly used in input processing thread to allow request initialization running in parallel with Model forward """ # Note on thread safety: no race condition. # `mm_receiver_cache` is reset at the end of LLMEngine init, # and will only be accessed in the input processing thread afterwards. if self.mm_receiver_cache is not None and request.mm_features: request.mm_features = self.mm_receiver_cache.get_and_update_features( request.mm_features ) req = Request.from_engine_core_request(request, self.request_block_hasher) if req.use_structured_output: # Note on thread safety: no race condition. # `grammar_init` is only invoked in input processing thread. For # `structured_output_manager`, each request is independent and # grammar compilation is async. Scheduler always checks grammar # compilation status before scheduling request. self.structured_output_manager.grammar_init(req) return req, request.current_wave class EngineCoreProc(EngineCore): """ZMQ-wrapper for running EngineCore in background process.""" ENGINE_CORE_DEAD = b"ENGINE_CORE_DEAD" def __init__( self, vllm_config: VllmConfig, local_client: bool, handshake_address: str, executor_class: type[Executor], log_stats: bool, client_handshake_address: str | None = None, engine_index: int = 0, ): self.input_queue = queue.Queue[tuple[EngineCoreRequestType, Any]]() self.output_queue = queue.Queue[tuple[int, EngineCoreOutputs] | bytes]() executor_fail_callback = lambda: self.input_queue.put_nowait( (EngineCoreRequestType.EXECUTOR_FAILED, b"") ) self.engine_index = engine_index identity = self.engine_index.to_bytes(length=2, byteorder="little") self.engines_running = False with self._perform_handshakes( handshake_address, identity, local_client, vllm_config, client_handshake_address, ) as addresses: self.client_count = len(addresses.outputs) # Set up data parallel environment. self.has_coordinator = addresses.coordinator_output is not None self.frontend_stats_publish_address = ( addresses.frontend_stats_publish_address ) logger.debug( "Has DP Coordinator: %s, stats publish address: %s", self.has_coordinator, self.frontend_stats_publish_address, ) # Only publish request queue stats to coordinator for "internal" # and "hybrid" LB modes . self.publish_dp_lb_stats = ( self.has_coordinator and not vllm_config.parallel_config.data_parallel_external_lb ) self._init_data_parallel(vllm_config) super().__init__( vllm_config, executor_class, log_stats, executor_fail_callback ) # Background Threads and Queues for IO. These enable us to # overlap ZMQ socket IO with GPU since they release the GIL, # and to overlap some serialization/deserialization with the # model forward pass. # Threads handle Socket <-> Queues and core_busy_loop uses Queue. ready_event = threading.Event() input_thread = threading.Thread( target=self.process_input_sockets, args=( addresses.inputs, addresses.coordinator_input, identity, ready_event, ), daemon=True, ) input_thread.start() self.output_thread = threading.Thread( target=self.process_output_sockets, args=( addresses.outputs, addresses.coordinator_output, self.engine_index, ), daemon=True, ) self.output_thread.start() # Don't complete handshake until DP coordinator ready message is # received. while not ready_event.wait(timeout=10): if not input_thread.is_alive(): raise RuntimeError("Input socket thread died during startup") assert addresses.coordinator_input is not None logger.info("Waiting for READY message from DP Coordinator...") # If enable, attach GC debugger after static variable freeze. maybe_attach_gc_debug_callback() # Enable environment variable cache (e.g. assume no more # environment variable overrides after this point) enable_envs_cache() @contextmanager def _perform_handshakes( self, handshake_address: str, identity: bytes, local_client: bool, vllm_config: VllmConfig, client_handshake_address: str | None, ) -> Generator[EngineZmqAddresses, None, None]: """ Perform startup handshakes. For DP=1 or offline mode, this is with the colocated front-end process. For DP>1 with internal load-balancing this is with the shared front-end process which may reside on a different node. For DP>1 with external or hybrid load-balancing, two handshakes are performed: - With the rank 0 front-end process which retrieves the DP Coordinator ZMQ addresses and DP process group address. - With the colocated front-end process which retrieves the client input/output socket addresses. with the exception of the rank 0 and colocated engines themselves which don't require the second handshake. Here, "front-end" process can mean the process containing the engine core client (which is the API server process in the case the API server is not scaled out), OR the launcher process running the run_multi_api_server() function in serve.py. """ input_ctx = zmq.Context() is_local = local_client and client_handshake_address is None headless = not local_client handshake = self._perform_handshake( input_ctx, handshake_address, identity, is_local, headless, vllm_config, vllm_config.parallel_config, ) if client_handshake_address is None: with handshake as addresses: yield addresses else: assert local_client local_handshake = self._perform_handshake( input_ctx, client_handshake_address, identity, True, False, vllm_config ) with handshake as addresses, local_handshake as client_addresses: addresses.inputs = client_addresses.inputs addresses.outputs = client_addresses.outputs yield addresses # Update config which may have changed from the handshake vllm_config.__post_init__() @contextmanager def _perform_handshake( self, ctx: zmq.Context, handshake_address: str, identity: bytes, local_client: bool, headless: bool, vllm_config: VllmConfig, parallel_config_to_update: ParallelConfig | None = None, ) -> Generator[EngineZmqAddresses, None, None]: with make_zmq_socket( ctx, handshake_address, zmq.DEALER, identity=identity, linger=5000, bind=False, ) as handshake_socket: # Register engine with front-end. addresses = self.startup_handshake( handshake_socket, local_client, headless, parallel_config_to_update ) yield addresses # Send ready message. num_gpu_blocks = vllm_config.cache_config.num_gpu_blocks # We pass back the coordinator stats update address here for the # external LB case for our colocated front-end to use (coordinator # only runs with rank 0). dp_stats_address = self.frontend_stats_publish_address # Include config hash for DP configuration validation ready_msg = { "status": "READY", "local": local_client, "headless": headless, "num_gpu_blocks": num_gpu_blocks, "dp_stats_address": dp_stats_address, } if vllm_config.parallel_config.data_parallel_size > 1: ready_msg["parallel_config_hash"] = ( vllm_config.parallel_config.compute_hash() ) handshake_socket.send(msgspec.msgpack.encode(ready_msg)) @staticmethod def startup_handshake( handshake_socket: zmq.Socket, local_client: bool, headless: bool, parallel_config: ParallelConfig | None = None, ) -> EngineZmqAddresses: # Send registration message. handshake_socket.send( msgspec.msgpack.encode( { "status": "HELLO", "local": local_client, "headless": headless, } ) ) # Receive initialization message. logger.debug("Waiting for init message from front-end.") if not handshake_socket.poll(timeout=HANDSHAKE_TIMEOUT_MINS * 60_000): raise RuntimeError( "Did not receive response from front-end " f"process within {HANDSHAKE_TIMEOUT_MINS} " f"minutes" ) init_bytes = handshake_socket.recv() init_message: EngineHandshakeMetadata = msgspec.msgpack.decode( init_bytes, type=EngineHandshakeMetadata ) logger.debug("Received init message: %s", init_message) if parallel_config is not None: for key, value in init_message.parallel_config.items(): setattr(parallel_config, key, value) return init_message.addresses @staticmethod def run_engine_core(*args, dp_rank: int = 0, local_dp_rank: int = 0, **kwargs): """Launch EngineCore busy loop in background process.""" # Signal handler used for graceful termination. # SystemExit exception is only raised once to allow this and worker # processes to terminate without error shutdown_requested = False # Ensure we can serialize transformer config after spawning maybe_register_config_serialize_by_value() def signal_handler(signum, frame): nonlocal shutdown_requested if not shutdown_requested: shutdown_requested = True raise SystemExit() # Either SIGTERM or SIGINT will terminate the engine_core signal.signal(signal.SIGTERM, signal_handler) signal.signal(signal.SIGINT, signal_handler) engine_core: EngineCoreProc | None = None try: parallel_config: ParallelConfig = kwargs["vllm_config"].parallel_config if parallel_config.data_parallel_size > 1 or dp_rank > 0: set_process_title("EngineCore", f"DP{dp_rank}") decorate_logs() # Set data parallel rank for this engine process. parallel_config.data_parallel_rank = dp_rank parallel_config.data_parallel_rank_local = local_dp_rank engine_core = DPEngineCoreProc(*args, **kwargs) else: set_process_title("EngineCore") decorate_logs() engine_core = EngineCoreProc(*args, **kwargs) engine_core.run_busy_loop() except SystemExit: logger.debug("EngineCore exiting.") raise except Exception as e: if engine_core is None: logger.exception("EngineCore failed to start.") else: logger.exception("EngineCore encountered a fatal error.") engine_core._send_engine_dead() raise e finally: if engine_core is not None: engine_core.shutdown() def _init_data_parallel(self, vllm_config: VllmConfig): pass def run_busy_loop(self): """Core busy loop of the EngineCore.""" # Loop until process is sent a SIGINT or SIGTERM while True: # 1) Poll the input queue until there is work to do. self._process_input_queue() # 2) Step the engine core and return the outputs. self._process_engine_step() def _process_input_queue(self): """Exits when an engine step needs to be performed.""" waited = False while ( not self.engines_running and not self.scheduler.has_requests() and not self.batch_queue ): if logger.isEnabledFor(DEBUG) and self.input_queue.empty(): logger.debug("EngineCore waiting for work.") waited = True req = self.input_queue.get() self._handle_client_request(*req) if waited: logger.debug("EngineCore loop active.") # Handle any more client requests. while not self.input_queue.empty(): req = self.input_queue.get_nowait() self._handle_client_request(*req) def _process_engine_step(self) -> bool: """Called only when there are unfinished local requests.""" # Step the engine core. outputs, model_executed = self.step_fn() # Put EngineCoreOutputs into the output queue. for output in outputs.items() if outputs else (): self.output_queue.put_nowait(output) # Post-step hook. self.post_step(model_executed) return model_executed def _handle_client_request( self, request_type: EngineCoreRequestType, request: Any ) -> None: """Dispatch request from client.""" if request_type == EngineCoreRequestType.ADD: req, request_wave = request self.add_request(req, request_wave) elif request_type == EngineCoreRequestType.ABORT: self.abort_requests(request) elif request_type == EngineCoreRequestType.UTILITY: client_idx, call_id, method_name, args = request output = UtilityOutput(call_id) try: method = getattr(self, method_name) result = method(*self._convert_msgspec_args(method, args)) output.result = UtilityResult(result) except BaseException as e: logger.exception("Invocation of %s method failed", method_name) output.failure_message = ( f"Call to {method_name} method failed: {str(e)}" ) self.output_queue.put_nowait( (client_idx, EngineCoreOutputs(utility_output=output)) ) elif request_type == EngineCoreRequestType.EXECUTOR_FAILED: raise RuntimeError("Executor failed.") else: logger.error( "Unrecognized input request type encountered: %s", request_type ) @staticmethod def _convert_msgspec_args(method, args): """If a provided arg type doesn't match corresponding target method arg type, try converting to msgspec object.""" if not args: return args arg_types = signature(method).parameters.values() assert len(args) <= len(arg_types) return tuple( msgspec.convert(v, type=p.annotation) if isclass(p.annotation) and issubclass(p.annotation, msgspec.Struct) and not isinstance(v, p.annotation) else v for v, p in zip(args, arg_types) ) def _send_engine_dead(self): """Send EngineDead status to the EngineCoreClient.""" # Put ENGINE_CORE_DEAD in the queue. self.output_queue.put_nowait(EngineCoreProc.ENGINE_CORE_DEAD) # Wait until msg sent by the daemon before shutdown. self.output_thread.join(timeout=5.0) if self.output_thread.is_alive(): logger.fatal( "vLLM shutdown signal from EngineCore failed " "to send. Please report this issue." ) def process_input_sockets( self, input_addresses: list[str], coord_input_address: str | None, identity: bytes, ready_event: threading.Event, ): """Input socket IO thread.""" # Msgpack serialization decoding. add_request_decoder = MsgpackDecoder(EngineCoreRequest) generic_decoder = MsgpackDecoder() with ExitStack() as stack, zmq.Context() as ctx: input_sockets = [ stack.enter_context( make_zmq_socket( ctx, input_address, zmq.DEALER, identity=identity, bind=False ) ) for input_address in input_addresses ] if coord_input_address is None: coord_socket = None else: coord_socket = stack.enter_context( make_zmq_socket( ctx, coord_input_address, zmq.XSUB, identity=identity, bind=False, ) ) # Send subscription message to coordinator. coord_socket.send(b"\x01") # Register sockets with poller. poller = zmq.Poller() for input_socket in input_sockets: # Send initial message to each input socket - this is required # before the front-end ROUTER socket can send input messages # back to us. input_socket.send(b"") poller.register(input_socket, zmq.POLLIN) if coord_socket is not None: # Wait for ready message from coordinator. assert coord_socket.recv() == b"READY" poller.register(coord_socket, zmq.POLLIN) ready_event.set() del ready_event while True: for input_socket, _ in poller.poll(): # (RequestType, RequestData) type_frame, *data_frames = input_socket.recv_multipart(copy=False) request_type = EngineCoreRequestType(bytes(type_frame.buffer)) # Deserialize the request data. if request_type == EngineCoreRequestType.ADD: request = add_request_decoder.decode(data_frames) request = self.preprocess_add_request(request) else: request = generic_decoder.decode(data_frames) # Push to input queue for core busy loop. self.input_queue.put_nowait((request_type, request)) def process_output_sockets( self, output_paths: list[str], coord_output_path: str | None, engine_index: int, ): """Output socket IO thread.""" # Msgpack serialization encoding. encoder = MsgpackEncoder() # Send buffers to reuse. reuse_buffers: list[bytearray] = [] # Keep references to outputs and buffers until zmq is finished # with them (outputs may contain tensors/np arrays whose # backing buffers were extracted for zero-copy send). pending = deque[tuple[zmq.MessageTracker, Any, bytearray]]() # We must set linger to ensure the ENGINE_CORE_DEAD # message is sent prior to closing the socket. with ExitStack() as stack, zmq.Context() as ctx: sockets = [ stack.enter_context( make_zmq_socket(ctx, output_path, zmq.PUSH, linger=4000) ) for output_path in output_paths ] coord_socket = ( stack.enter_context( make_zmq_socket( ctx, coord_output_path, zmq.PUSH, bind=False, linger=4000 ) ) if coord_output_path is not None else None ) max_reuse_bufs = len(sockets) + 1 while True: output = self.output_queue.get() if output == EngineCoreProc.ENGINE_CORE_DEAD: for socket in sockets: socket.send(output) break assert not isinstance(output, bytes) client_index, outputs = output outputs.engine_index = engine_index if client_index == -1: # Don't reuse buffer for coordinator message # which will be very small. assert coord_socket is not None coord_socket.send_multipart(encoder.encode(outputs)) continue # Reclaim buffers that zmq is finished with. while pending and pending[-1][0].done: reuse_buffers.append(pending.pop()[2]) buffer = reuse_buffers.pop() if reuse_buffers else bytearray() buffers = encoder.encode_into(outputs, buffer) tracker = sockets[client_index].send_multipart( buffers, copy=False, track=True ) if not tracker.done: ref = outputs if len(buffers) > 1 else None pending.appendleft((tracker, ref, buffer)) elif len(reuse_buffers) < max_reuse_bufs: # Limit the number of buffers to reuse. reuse_buffers.append(buffer) class DPEngineCoreProc(EngineCoreProc): """ZMQ-wrapper for running EngineCore in background process in a data parallel context.""" def __init__( self, vllm_config: VllmConfig, local_client: bool, handshake_address: str, executor_class: type[Executor], log_stats: bool, client_handshake_address: str | None = None, ): # Counts forward-passes of the model so that we can synchronize # finished with DP peers every N steps. self.step_counter = 0 self.current_wave = 0 self.last_counts = (0, 0) # Initialize the engine. dp_rank = vllm_config.parallel_config.data_parallel_rank super().__init__( vllm_config, local_client, handshake_address, executor_class, log_stats, client_handshake_address, dp_rank, ) def _init_data_parallel(self, vllm_config: VllmConfig): # Configure GPUs and stateless process group for data parallel. dp_rank = vllm_config.parallel_config.data_parallel_rank dp_size = vllm_config.parallel_config.data_parallel_size local_dp_rank = vllm_config.parallel_config.data_parallel_rank_local assert dp_size > 1 assert local_dp_rank is not None assert 0 <= local_dp_rank <= dp_rank < dp_size if vllm_config.kv_transfer_config is not None: # modify the engine_id and append the local_dp_rank to it to ensure # that the kv_transfer_config is unique for each DP rank. vllm_config.kv_transfer_config.engine_id = ( f"{vllm_config.kv_transfer_config.engine_id}_dp{local_dp_rank}" ) logger.debug( "Setting kv_transfer_config.engine_id to %s", vllm_config.kv_transfer_config.engine_id, ) self.dp_rank = dp_rank self.dp_group = vllm_config.parallel_config.stateless_init_dp_group() def shutdown(self): super().shutdown() if dp_group := getattr(self, "dp_group", None): stateless_destroy_torch_distributed_process_group(dp_group) def add_request(self, request: Request, request_wave: int = 0): if self.has_coordinator and request_wave != self.current_wave: if request_wave > self.current_wave: self.current_wave = request_wave elif not self.engines_running: # Request received for an already-completed wave, notify # front-end that we need to start the next one. self.output_queue.put_nowait( (-1, EngineCoreOutputs(start_wave=self.current_wave)) ) super().add_request(request, request_wave) def _handle_client_request( self, request_type: EngineCoreRequestType, request: Any ) -> None: if request_type == EngineCoreRequestType.START_DP_WAVE: new_wave, exclude_eng_index = request if exclude_eng_index != self.engine_index and ( new_wave >= self.current_wave ): self.current_wave = new_wave if not self.engines_running: logger.debug("EngineCore starting idle loop for wave %d.", new_wave) self.engines_running = True else: super()._handle_client_request(request_type, request) def _maybe_publish_request_counts(self): if not self.publish_dp_lb_stats: return # Publish our request counts (if they've changed). counts = self.scheduler.get_request_counts() if counts != self.last_counts: self.last_counts = counts stats = SchedulerStats( *counts, step_counter=self.step_counter, current_wave=self.current_wave ) self.output_queue.put_nowait((-1, EngineCoreOutputs(scheduler_stats=stats))) def run_busy_loop(self): """Core busy loop of the EngineCore for data parallel case.""" # Loop until process is sent a SIGINT or SIGTERM while True: # 1) Poll the input queue until there is work to do. self._process_input_queue() # 2) Step the engine core. executed = self._process_engine_step() self._maybe_publish_request_counts() local_unfinished_reqs = self.scheduler.has_unfinished_requests() if not executed: if not local_unfinished_reqs and not self.engines_running: # All engines are idle. continue # We are in a running state and so must execute a dummy pass # if the model didn't execute any ready requests. self.execute_dummy_batch() # 3) All-reduce operation to determine global unfinished reqs. self.engines_running = self._has_global_unfinished_reqs( local_unfinished_reqs ) if not self.engines_running: if self.dp_rank == 0 or not self.has_coordinator: # Notify client that we are pausing the loop. logger.debug( "Wave %d finished, pausing engine loop.", self.current_wave ) # In the coordinator case, dp rank 0 sends updates to the # coordinator. Otherwise (offline spmd case), each rank # sends the update to its colocated front-end process. client_index = -1 if self.has_coordinator else 0 self.output_queue.put_nowait( ( client_index, EngineCoreOutputs(wave_complete=self.current_wave), ) ) # Increment wave count and reset step counter. self.current_wave += 1 self.step_counter = 0 def _has_global_unfinished_reqs(self, local_unfinished: bool) -> bool: # Optimization - only perform finish-sync all-reduce every 32 steps. self.step_counter += 1 if self.step_counter % 32 != 0: return True return ParallelConfig.has_unfinished_dp(self.dp_group, local_unfinished) def reinitialize_distributed( self, reconfig_request: ReconfigureDistributedRequest ) -> None: stateless_destroy_torch_distributed_process_group(self.dp_group) self.shutdown() parallel_config = self.vllm_config.parallel_config old_dp_size = parallel_config.data_parallel_size parallel_config.data_parallel_size = reconfig_request.new_data_parallel_size if reconfig_request.new_data_parallel_rank != -1: parallel_config.data_parallel_rank = reconfig_request.new_data_parallel_rank # local rank specifies device visibility, it should not be changed assert ( reconfig_request.new_data_parallel_rank_local == ReconfigureRankType.KEEP_CURRENT_RANK ) parallel_config.data_parallel_master_ip = ( reconfig_request.new_data_parallel_master_ip ) parallel_config.data_parallel_master_port = ( reconfig_request.new_data_parallel_master_port ) if reconfig_request.new_data_parallel_rank != -2: self.dp_rank = parallel_config.data_parallel_rank self.dp_group = parallel_config.stateless_init_dp_group() reconfig_request.new_data_parallel_master_port = ( parallel_config.data_parallel_master_port ) self.model_executor.reinitialize_distributed(reconfig_request) if reconfig_request.new_data_parallel_size > old_dp_size: assert self.available_gpu_memory_for_kv_cache > 0 # pass available_gpu_memory_for_kv_cache from existing # engine-cores to new engine-cores so they can directly # use it in _initialize_kv_caches() rather than profiling. ParallelConfig.sync_kv_cache_memory_size( self.dp_group, self.available_gpu_memory_for_kv_cache ) # NOTE(yongji): newly joined workers require dummy_run even # CUDA graph is not used self.model_executor.collective_rpc("compile_or_warm_up_model") if ( reconfig_request.new_data_parallel_rank == ReconfigureRankType.SHUTDOWN_CURRENT_RANK ): self.shutdown() logger.info("DPEngineCoreProc %s shutdown", self.dp_rank) else: logger.info( "Distributed environment reinitialized for DP rank %s", self.dp_rank ) class DPEngineCoreActor(DPEngineCoreProc): """ Ray actor for running EngineCore in a data parallel context """ def __init__( self, vllm_config: VllmConfig, local_client: bool, addresses: EngineZmqAddresses, executor_class: type[Executor], log_stats: bool, dp_rank: int = 0, local_dp_rank: int = 0, ): self.addresses = addresses vllm_config.parallel_config.data_parallel_rank = dp_rank vllm_config.parallel_config.data_parallel_rank_local = local_dp_rank # Set CUDA_VISIBLE_DEVICES as early as possible in actor life cycle # NOTE: in MP we set CUDA_VISIBLE_DEVICES at process creation time, # and this cannot be done in the same way for Ray because: # 1) Ray manages life cycle of all ray workers (including # DPEngineCoreActor) # 2) Ray sets CUDA_VISIBLE_DEVICES based on num_gpus configuration # To bypass 2, we need to also set # RAY_EXPERIMENTAL_NOSET_CUDA_VISIBLE_DEVICES, but vLLM workers created # thereafter would have CUDA_VISIBLE_DEVICES set, which is sticky: # https://github.com/ray-project/ray/blob/e752fc319ddedd9779a0989b6d3613909bad75c9/python/ray/_private/worker.py#L456 # noqa: E501 # This is problematic because when the vLLM worker (a Ray actor) # executes a task, it indexes into the sticky CUDA_VISIBLE_DEVICES # rather than directly using the GPU ID, potentially resulting in # index out of bounds error. See: # https://github.com/ray-project/ray/pull/40461/files#diff-31e8159767361e4bc259b6d9883d9c0d5e5db780fcea4a52ead4ee3ee4a59a78R1860 # noqa: E501 # and get_accelerator_ids_for_accelerator_resource() in worker.py # of ray. self._set_visible_devices(vllm_config, local_dp_rank) super().__init__(vllm_config, local_client, "", executor_class, log_stats) def _set_visible_devices(self, vllm_config: VllmConfig, local_dp_rank: int): from vllm.platforms import current_platform if current_platform.is_xpu(): pass else: device_control_env_var = current_platform.device_control_env_var self._set_cuda_visible_devices( vllm_config, local_dp_rank, device_control_env_var ) def _set_cuda_visible_devices( self, vllm_config: VllmConfig, local_dp_rank: int, device_control_env_var: str ): world_size = vllm_config.parallel_config.world_size # Set CUDA_VISIBLE_DEVICES or equivalent. try: value = get_device_indices( device_control_env_var, local_dp_rank, world_size ) os.environ[device_control_env_var] = value except IndexError as e: raise Exception( f"Error setting {device_control_env_var}: " f"local range: [{local_dp_rank * world_size}, " f"{(local_dp_rank + 1) * world_size}) " f'base value: "{os.getenv(device_control_env_var)}"' ) from e @contextmanager def _perform_handshakes( self, handshake_address: str, identity: bytes, local_client: bool, vllm_config: VllmConfig, client_handshake_address: str | None, ): """ For Ray, we don't need to actually perform handshake. All addresses information is known before the actor creation. Therefore, we simply yield these addresses. """ yield self.addresses def wait_for_init(self): """ Wait until the engine core is initialized. This is just an empty method. When ray.get() on this method (or any other method of the actor) returns, it is guaranteed that actor creation (i.e., __init__) is complete. """ pass def run(self): """ Run the engine core busy loop. """ try: self.run_busy_loop() except SystemExit: logger.debug("EngineCore exiting.") raise except Exception: logger.exception("EngineCore encountered a fatal error.") raise finally: self.shutdown()