""" # Copyright (c) 2025 PaddlePaddle Authors. All Rights Reserved. # # Licensed under the Apache License, Version 2.0 (the "License" # you may not use this file except in compliance with the License. # You may obtain a copy of the License at # # http://www.apache.org/licenses/LICENSE-2.0 # # Unless required by applicable law or agreed to in writing, software # distributed under the License is distributed on an "AS IS" BASIS, # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. # See the License for the specific language governing permissions and # limitations under the License. """ import argparse import concurrent.futures import gc import json import os import queue import threading import time import traceback from typing import List import numpy as np import paddle import yaml from fastdeploy import envs from fastdeploy.cache_manager.cache_data import CacheStatus from fastdeploy.cache_manager.cache_tasks import ReadStorageTask, WriteStorageTask from fastdeploy.cache_manager.ops import ( cuda_host_alloc, cuda_host_free, memory_allocated, set_data_ipc, set_device, share_external_data_, swap_cache_all_layers, swap_cache_layout, unset_data_ipc, ) from fastdeploy.cache_manager.transfer_factory import ( AttentionStore, FileStore, MooncakeStore, ) from fastdeploy.config import CacheConfig, SpeculativeConfig from fastdeploy.inter_communicator import EngineCacheQueue, IPCSignal, KVCacheStatus from fastdeploy.platforms import current_platform from fastdeploy.utils import console_logger, get_logger def parse_args(): """ 从命令行解析参数 """ parser = argparse.ArgumentParser("Cache transfer manager") parser.add_argument( "--splitwise_role", type=str, default="mixed", help="splitwise role, can be decode, prefill or mixed", ) parser.add_argument("--rank", type=int, default=0, help="local tp rank") parser.add_argument("--device_id", type=int, default=0, help="device id") parser.add_argument("--max_model_len", type=int, default=32768, help="max model length") parser.add_argument("--num_layers", type=int, default=1, help="model num layers") parser.add_argument("--mp_num", type=int, default=1, help="number of model parallel") parser.add_argument( "--cache_dtype", type=str, default="bfloat16", choices=["uint8", "bfloat16", "block_wise_fp8"], help="cache dtype", ) parser.add_argument( "--default_dtype", type=str, default="bfloat16", choices=["float16", "bfloat16", "uint8"], help="paddle default dtype, swap_cache_batch only support float16、bfloat16 and uint8 now", ) parser.add_argument("--key_cache_shape", type=str, default="", help="key cache shape") parser.add_argument("--value_cache_shape", type=str, default="", help="value cache shape") parser.add_argument("--cache_queue_port", type=int, default=9923, help="cache queue port") parser.add_argument("--enable_splitwise", type=int, default=0, help="enable splitwise ") parser.add_argument("--pod_ip", type=str, default="0.0.0.0", help="pod ip") parser.add_argument( "--engine_worker_queue_port", type=int, default=9923, help="engine worker queue port", ) parser.add_argument("--num_cpu_blocks", type=int, default=4, help="cpu cache block number") parser.add_argument("--ipc_suffix", type=str, default=None, help="engine pid") parser.add_argument( "--protocol", type=str, default="ipc", help="cache transfer protocol, only support ipc now", ) parser.add_argument("--local_data_parallel_id", type=int, default=0) parser.add_argument("--rdma_port", type=str, default="", help="rmda port") parser.add_argument( "--speculative_config", type=json.loads, default="{}", help="speculative config", ) parser.add_argument("--create_cache_tensor", action="store_true") parser.add_argument( "--kvcache_storage_backend", type=str, default=None, choices=["mooncake", "attention_store", "file"], help="The storage backend for kvcache storage. If not set, storage backend is disabled.", ) parser.add_argument( "--write_policy", type=str, choices=["write_through"], default="write_through", help="KVCache write policy", ) parser.add_argument("--model_path", type=str, help="The path of model") args = parser.parse_args() return args def get_key_prefix_from_version(version_file_path): # the format of version string is RL-STEP{xx}-{timestamp}-{uuid4} with open(version_file_path, "r", encoding="utf-8") as f: data = yaml.safe_load(f) version = data["version"] parts = version.split("-", 2) key_prefix = "-".join(parts[:2]) return key_prefix class CacheTransferManager: """ 管理CPU和GPU之间缓存的交换传输 """ def __init__(self, args): """ 初始化CacheTransferManager """ self.gpu_cache_kvs = {} self.cpu_cache_kvs = {} self.gpu_cache_k_tensors = [] self.gpu_cache_v_tensors = [] self.gpu_cache_scales_k_tensors = [] self.gpu_cache_scales_v_tensors = [] self.speculative_config = SpeculativeConfig(args.speculative_config) # parse kv cache shape self.key_cache_shape = [int(i) for i in args.key_cache_shape.split(",")] self.value_cache_shape = [] if args.value_cache_shape: self.value_cache_shape = [int(i) for i in args.value_cache_shape.split(",")] # extract kv cache shape into fields self.num_gpu_blocks = self.key_cache_shape[0] self.head_num = self.key_cache_shape[1] self.block_size = self.key_cache_shape[2] self.head_dim = self.key_cache_shape[3] # compute cache bytes self.cache_dtype = args.cache_dtype self.cache_item_bytes = CacheConfig.get_cache_bytes(self.cache_dtype) self.scale_item_bytes = CacheConfig.get_cache_bytes(paddle.get_default_dtype()) self.has_cache_scale = self.cache_dtype == "block_wise_fp8" if self.has_cache_scale: self.cache_scale_shape = [self.num_gpu_blocks, self.head_num, self.block_size] # extract other arg values self.model_id = os.path.basename(args.model_path.rstrip("/")) self.n_ranks = args.mp_num self.rank = args.rank self.device = args.device_id self.num_layers = args.num_layers self.ipc_suffix = args.ipc_suffix self.local_data_parallel_id = args.local_data_parallel_id self.num_extra_layers = self.speculative_config.num_extra_cache_layer self.num_extra_layer_gpu_blocks = int(self.num_gpu_blocks * self.speculative_config.num_gpu_block_expand_ratio) paddle.set_default_dtype(args.default_dtype) self.swap_to_cpu_thread_pool = concurrent.futures.ThreadPoolExecutor(max_workers=1) self.swap_to_gpu_thread_pool = concurrent.futures.ThreadPoolExecutor(max_workers=1) self.read_storage_thread_pool = concurrent.futures.ThreadPoolExecutor(max_workers=1) self.write_back_storage_thread_pool = concurrent.futures.ThreadPoolExecutor(max_workers=1) self.timeout_thread_pool = concurrent.futures.ThreadPoolExecutor(max_workers=2) self.transfer_task_queue = queue.Queue() # 用来接收传输任务 self.tansfer_done_queue = queue.Queue() # 用来告知任务执行完毕 address = (args.pod_ip, args.cache_queue_port) self.cache_task_queue = EngineCacheQueue( address=address, is_server=False, num_client=args.mp_num, client_id=self.rank, local_data_parallel_id=args.local_data_parallel_id, ) cache_ready_signal_data = np.zeros(shape=[args.mp_num], dtype=np.int32) self.cache_ready_signal = IPCSignal( name="cache_ready_signal", array=cache_ready_signal_data, dtype=np.int32, suffix=args.engine_worker_queue_port, create=False, ) swap_space_ready_data = np.zeros(shape=[args.mp_num], dtype=np.int32) self.swap_space_ready_signal = IPCSignal( name="swap_space_ready_signal", array=swap_space_ready_data, dtype=np.int32, suffix=args.engine_worker_queue_port, create=False, ) self.num_cpu_blocks = args.num_cpu_blocks self._init_gpu_cache(args) if self.num_cpu_blocks > 0: self._init_cpu_cache(args) self._init_storage(args) cache_task_broadcast_data = np.zeros(shape=[1], dtype=np.int32) self.cache_task_broadcast_signal = IPCSignal( name="cache_task_broadcast_signal", array=cache_task_broadcast_data, dtype=np.int32, suffix=args.engine_worker_queue_port, create=False, ) # NOTE: `cache_task_is_paused_signal` indicates if do_data_transfer thread # of the FIRST rank (rank#0) has received a pause signal self.cache_task_is_paused_signal = IPCSignal( name="cache_task_is_paused", array=np.zeros([1], dtype=np.int32), dtype=np.int32, suffix=args.engine_worker_queue_port, create=False, ) # NOTE: `cache_task_inflight_signal` indicates if do_data_transfer thread # of each rank has finished remaining tasks and finally paused self.cache_task_inflight_signal = IPCSignal( name="cache_task_inflight", array=np.zeros([self.n_ranks], dtype=np.int32), dtype=np.int32, suffix=args.engine_worker_queue_port, create=False, ) max_chips_per_node = 16 if current_platform.is_iluvatar() else 8 array_size = min(max_chips_per_node, args.mp_num) worker_healthy_live_array = np.zeros(shape=[array_size], dtype=np.int32) self.worker_healthy_live_signal = IPCSignal( name="worker_healthy_live_signal", array=worker_healthy_live_array, dtype=np.int32, suffix=args.engine_worker_queue_port, create=False, ) # Initialize update/clear signals for RL self.kv_cache_status_signal = IPCSignal( name="kv_cache_status", array=np.zeros([1], dtype=np.int32), dtype=np.int32, suffix=args.engine_worker_queue_port, create=False, ) threading.Thread(target=self.check_cache_status, args=[args], daemon=True).start() self.is_paused = False # transfer manager state self.inflight = 0 # number of inflight transfer tasks cache_transfer_inited_signal_data = np.zeros(shape=[args.mp_num], dtype=np.int32) self.cache_transfer_inited_signal = IPCSignal( name="cache_transfer_inited_signal", array=cache_transfer_inited_signal_data, dtype=np.int32, suffix=args.engine_worker_queue_port, create=False, ) self.cache_transfer_inited_signal.value[self.rank] = 1 def _init_storage(self, args): self.storage_backend_type = args.kvcache_storage_backend try: # TODO: support cache scale for other backend if self.has_cache_scale: if self.storage_backend_type not in ["mooncake"]: raise ValueError( f"Unsupported storage backend ({self.storage_backend_type}) " "when cache quantization is block_wise_fp8" ) if self.storage_backend_type is None: self.storage_backend = None elif self.storage_backend_type == "mooncake": logger.info("Start initialize mooncake store...") self.storage_backend = MooncakeStore(tp_rank=self.rank) self._init_storage_buffer(args) logger.info("Initialized mooncake store successfully") elif self.storage_backend_type == "attention_store": logger.info("Start initialize attention store...") # TODO: support different model version in rl self.storage_backend = AttentionStore( namespace=self.model_id, shard_id=self.rank, shard_num=self.n_ranks, layer_num=self.num_layers + self.num_extra_layers, block_token_size=self.block_size, bytes_per_shard_layer_per_block=self.head_num * self.block_size * self.head_dim * self.cache_item_bytes, device_id=self.device, dp_id=self.local_data_parallel_id, ) logger.info("Initialized attention store successfully!") elif args.kvcache_storage_backend == "file": logger.info("Start initialize file store...") self.storage_backend = FileStore( namespace=self.model_id, tp_rank=self.rank, tp_size=self.n_ranks, ) self._init_storage_buffer(args) logger.info("Initialized file store successfully") else: raise NotImplementedError(f"Unsupported storage backend: {self.storage_backend_type}") except Exception as e: err_msg = f"Fail to initialize storage backend, {e}, traceback: {traceback.format_exc()}" logger.error(err_msg) console_logger.error(err_msg) # print error message to console raise if args.write_policy not in ["write_through"]: raise ValueError(f"Invalid write policy: {args.write_policy}") self.write_policy = args.write_policy self.key_prefix = "" version_file_path = os.path.join(args.model_path, "version.yaml") if os.path.exists(version_file_path): self.key_prefix = get_key_prefix_from_version(version_file_path) logger.info(f"The key_prefix of cache storage is {self.key_prefix}") logger.info("Initialize cache storage successfully") def _init_storage_buffer(self, args): """ Initialize pinned memory buffer that can hold the cache for a longest request cache layout: layer_num * [block_num, head_num, block_size, head_dim] scale layout: layer_num * [block_num, head_num, block_size] cache buffer layout: [block_num, layer_num, head_num, block_size, head_dim] scale buffer layout: [block_num, layer_num, head_num, block_size] """ layer_num = self.num_layers + self.num_extra_layers block_num = (args.max_model_len + self.block_size - 1) // self.block_size logger.info( f"Creating cache buffer for storage with shape: " f"[{block_num}, {layer_num}, {self.head_num}, {self.block_size}, {self.head_dim}]" ) self.cache_buffer_stride_bytes = ( layer_num * self.head_num * self.block_size * self.head_dim * self.cache_item_bytes ) cache_buffer_total_bytes = block_num * self.cache_buffer_stride_bytes * 2 # key and value logger.info(f"Creating cache cpu buffer for all layers: {cache_buffer_total_bytes / 1024 ** 3:.2f}GB") read_buffer = cuda_host_alloc(cache_buffer_total_bytes) self.storage_key_read_buffer = read_buffer self.storage_value_read_buffer = read_buffer + cache_buffer_total_bytes // 2 self.storage_backend.register_buffer(read_buffer, cache_buffer_total_bytes) write_buffer = cuda_host_alloc(cache_buffer_total_bytes) self.storage_key_write_buffer = write_buffer self.storage_value_write_buffer = write_buffer + cache_buffer_total_bytes // 2 self.storage_backend.register_buffer(write_buffer, cache_buffer_total_bytes) if self.has_cache_scale: self.scale_buffer_stride_bytes = layer_num * self.head_num * self.block_size * self.scale_item_bytes scale_buffer_total_bytes = block_num * self.scale_buffer_stride_bytes * 2 logger.info( f"Creating scale cpu buffer cache for all layers: {scale_buffer_total_bytes / 1024 ** 3:.2f}GB" ) read_buffer = cuda_host_alloc(scale_buffer_total_bytes) self.storage_key_scale_read_buffer = read_buffer self.storage_value_scale_read_buffer = read_buffer + scale_buffer_total_bytes // 2 self.storage_backend.register_buffer(read_buffer, scale_buffer_total_bytes) write_buffer = cuda_host_alloc(scale_buffer_total_bytes) self.storage_key_scale_write_buffer = write_buffer self.storage_value_scale_write_buffer = write_buffer + scale_buffer_total_bytes // 2 self.storage_backend.register_buffer(write_buffer, scale_buffer_total_bytes) def _init_gpu_cache(self, args): if not args.create_cache_tensor: logger.info(f"[rank {self.rank}/{self.n_ranks}] Waiting for runners or messagers to create kv cache.") while self.cache_ready_signal.value[self.rank] != 1: time.sleep(0.1) logger.info(f"[rank {self.rank}/{self.n_ranks}] OK! Stop waiting.") if args.cache_dtype == "block_wise_fp8": cache_type = "uint8" else: cache_type = args.cache_dtype logger.info(f"[rank {self.rank}/{self.n_ranks}] Initializing kv cache for all layers.") set_device(self.device) for i in range(self.num_layers + self.num_extra_layers): # NOTE: num_extra_layer_gpu_blocks is usually equal to num_gpu_blocks num_gpu_blocks = self.num_gpu_blocks if i < self.num_layers else self.num_extra_layer_gpu_blocks key_name = f"key_caches_{i}_rank{self.rank}.device{self.device}" val_name = f"value_caches_{i}_rank{self.rank}.device{self.device}" key_cache_scales_name = f"key_cache_scales_{i}_rank{self.rank}.device{self.device}" value_cache_scales_name = f"value_cache_scales_{i}_rank{self.rank}.device{self.device}" key_cache_shape = [ num_gpu_blocks, self.key_cache_shape[1], self.key_cache_shape[2], self.key_cache_shape[3], ] value_cache_shape = [] if self.value_cache_shape: value_cache_shape = [ num_gpu_blocks, self.value_cache_shape[1], self.value_cache_shape[2], self.value_cache_shape[3], ] if args.create_cache_tensor: logger.info( f"[rank {self.rank}/{self.n_ranks}] ..creating kv cache for layer {i}: {key_cache_shape} {value_cache_shape}" ) key_cache = paddle.full(shape=key_cache_shape, fill_value=0, dtype=cache_type) set_data_ipc(key_cache, key_name) if args.cache_dtype == "block_wise_fp8": key_cache_scales = paddle.full( shape=[num_gpu_blocks, self.key_cache_shape[1], self.key_cache_shape[2]], fill_value=0, dtype=paddle.get_default_dtype(), ) set_data_ipc(key_cache_scales, key_cache_scales_name) if self.value_cache_shape: val_cache = paddle.full(shape=value_cache_shape, fill_value=0, dtype=cache_type) set_data_ipc(val_cache, val_name) if args.cache_dtype == "block_wise_fp8": value_cache_scales = paddle.full( shape=[num_gpu_blocks, self.value_cache_shape[1], self.value_cache_shape[2]], fill_value=0, dtype=paddle.get_default_dtype(), ) set_data_ipc(value_cache_scales, value_cache_scales_name) else: logger.info( f"[rank {self.rank}/{self.n_ranks}] ..attaching kv cache for layer {i}: {key_cache_shape} {value_cache_shape}" ) key_cache = paddle.empty(shape=[], dtype=cache_type) val_cache = paddle.empty(shape=[], dtype=cache_type) key_cache = share_external_data_(key_cache, key_name, key_cache_shape, True) if args.cache_dtype == "block_wise_fp8": key_cache_scales = paddle.empty(shape=[], dtype=paddle.get_default_dtype()) key_cache_scales = share_external_data_( key_cache_scales, key_cache_scales_name, [num_gpu_blocks, self.key_cache_shape[1], self.key_cache_shape[2]], True, ) if self.value_cache_shape: val_cache = share_external_data_(val_cache, val_name, value_cache_shape, True) if args.cache_dtype == "block_wise_fp8": value_cache_scales = paddle.empty(shape=[], dtype=paddle.get_default_dtype()) value_cache_scales = share_external_data_( value_cache_scales, value_cache_scales_name, [num_gpu_blocks, self.value_cache_shape[1], self.value_cache_shape[2]], True, ) self.gpu_cache_kvs[key_name] = key_cache self.gpu_cache_k_tensors.append(self.gpu_cache_kvs[key_name]) if args.cache_dtype == "block_wise_fp8": self.gpu_cache_kvs[key_cache_scales_name] = key_cache_scales self.gpu_cache_scales_k_tensors.append(self.gpu_cache_kvs[key_cache_scales_name]) if args.value_cache_shape: self.gpu_cache_kvs[val_name] = val_cache self.gpu_cache_v_tensors.append(self.gpu_cache_kvs[val_name]) if args.cache_dtype == "block_wise_fp8": self.gpu_cache_kvs[value_cache_scales_name] = value_cache_scales self.gpu_cache_scales_v_tensors.append(self.gpu_cache_kvs[value_cache_scales_name]) if args.create_cache_tensor: logger.info(f"[rank {self.rank}/{self.n_ranks}] ✅ kv cache is ready!") self.cache_ready_signal.value[self.rank] = 1 cache_kv_size_byte = sum([tmp.numel() * 1 for key, tmp in self.gpu_cache_kvs.items()]) logger.info(f"[rank {self.rank}/{self.n_ranks}] device :{self.device}") logger.info(f"[rank {self.rank}/{self.n_ranks}] cache_kv_size_byte : {cache_kv_size_byte}") logger.info(f"[rank {self.rank}/{self.n_ranks}] done init cache (full) gmem alloc : {memory_allocated()}") def _init_cpu_cache(self, args): key_cache_size = self.key_cache_shape[1] * self.key_cache_shape[2] * self.key_cache_shape[3] if args.value_cache_shape: value_cache_size = self.value_cache_shape[1] * self.value_cache_shape[2] * self.value_cache_shape[3] else: value_cache_size = 0 cache_item_bytes = CacheConfig.get_cache_bytes(self.cache_dtype) key_need_to_allocate_bytes = args.num_cpu_blocks * cache_item_bytes * key_cache_size value_need_to_allocate_bytes = args.num_cpu_blocks * cache_item_bytes * value_cache_size if args.cache_dtype == "block_wise_fp8": cache_scales = paddle.empty(shape=[], dtype=paddle.get_default_dtype()) cache_scales_size = self.key_cache_shape[1] * self.key_cache_shape[2] scales_key_need_to_allocate_bytes = args.num_cpu_blocks * cache_scales.element_size() * cache_scales_size scales_value_need_to_allocate_bytes = args.num_cpu_blocks * cache_scales.element_size() * cache_scales_size logger.info( f"[rank {self.rank}/{self.n_ranks}] ..swap space size : {(key_need_to_allocate_bytes + value_need_to_allocate_bytes) / 1024 ** 3:.2f}GB" ) if args.num_cpu_blocks == 0: logger.info(f"[rank {self.rank}/{self.n_ranks}] 💡 no swap space (cpu cache) is specified.") self.swap_space_ready_signal.value[self.rank] = 1 return logger.info(f"[rank {self.rank}/{self.n_ranks}] Initializing swap space (cpu cache) for all layers.") paddle.set_device("cpu") self.k_dst_ptrs = [] self.v_dst_ptrs = [] self.k_scales_ptrs = [] self.v_scales_ptrs = [] for i in range(self.num_layers + self.num_extra_layers): key_name = f"key_caches_{i}_rank{self.rank}" val_name = f"value_caches_{i}_rank{self.rank}" key_cache_scales_name = f"key_cache_scales_{i}_rank{self.rank}" value_cache_scales_name = f"value_cache_scales_{i}_rank{self.rank}" logger.info( f"[rank {self.rank}/{self.n_ranks}] ..creating cpu cache for layer {i}: {(key_need_to_allocate_bytes + value_need_to_allocate_bytes) / 1024 ** 3:.2f}GB" ) self.cpu_cache_kvs[key_name] = cuda_host_alloc(key_need_to_allocate_bytes) self.k_dst_ptrs.append(self.cpu_cache_kvs[key_name]) if args.cache_dtype == "block_wise_fp8": self.cpu_cache_kvs[key_cache_scales_name] = cuda_host_alloc(scales_key_need_to_allocate_bytes) self.k_scales_ptrs.append(self.cpu_cache_kvs[key_cache_scales_name]) if value_need_to_allocate_bytes > 0: self.cpu_cache_kvs[val_name] = cuda_host_alloc(value_need_to_allocate_bytes) self.v_dst_ptrs.append(self.cpu_cache_kvs[val_name]) if args.cache_dtype == "block_wise_fp8": self.cpu_cache_kvs[value_cache_scales_name] = cuda_host_alloc(scales_value_need_to_allocate_bytes) self.v_scales_ptrs.append(self.cpu_cache_kvs[value_cache_scales_name]) logger.info(f"[rank {self.rank}/{self.n_ranks}] ✅ swap space (cpu cache) is ready!") self.swap_space_ready_signal.value[self.rank] = 1 def _run_read_storage( self, task_id: str, token_ids: List[int], start_read_block_idx: int, k_cache_keys: List[str], v_cache_keys: List[str], k_scale_keys: List[str], v_scale_keys: List[str], gpu_block_ids: List[int], cpu_block_ids: List[int], timeout: float, ): """ Read storage data from the given blocks to the corresponding cache tensors on the current rank's GPU. """ try: if self.storage_backend_type in ("mooncake", "file"): block_num = len(gpu_block_ids) keys = k_cache_keys + v_cache_keys k_cache_ptrs = [ self.storage_key_read_buffer + i * self.cache_buffer_stride_bytes for i in cpu_block_ids ] v_cache_ptrs = [ self.storage_value_read_buffer + i * self.cache_buffer_stride_bytes for i in cpu_block_ids ] target_locations = k_cache_ptrs + v_cache_ptrs target_sizes = [self.cache_buffer_stride_bytes] * block_num * 2 # key and value if k_scale_keys and v_scale_keys: keys.extend(k_scale_keys + v_scale_keys) k_scale_ptrs = [ self.storage_key_scale_read_buffer + i * self.scale_buffer_stride_bytes for i in cpu_block_ids ] v_scale_ptrs = [ self.storage_value_scale_read_buffer + i * self.scale_buffer_stride_bytes for i in cpu_block_ids ] target_locations.extend(k_scale_ptrs + v_scale_ptrs) target_sizes.extend([self.scale_buffer_stride_bytes] * block_num * 2) start_time = time.time() result = self.storage_backend.batch_get( keys=keys, target_locations=target_locations, target_sizes=target_sizes ) read_cost_time = time.time() - start_time if k_scale_keys and v_scale_keys: k_result, v_result = result[:block_num], result[block_num : 2 * block_num] k_scale_result, v_scale_result = result[2 * block_num : 3 * block_num], result[3 * block_num :] success_block_num = 0 for k, v, k_scale, v_scale in zip(k_result, v_result, k_scale_result, v_scale_result): if not (k > 0 and v > 0 and k_scale > 0 and v_scale > 0): break success_block_num += 1 else: k_result, v_result = result[:block_num], result[block_num : 2 * block_num] success_block_num = 0 for k, v in zip(k_result, v_result): if not (k > 0 and v > 0): break success_block_num += 1 logger.debug(f"_run_read_storage, success_block_num: {success_block_num}") valid_gpu_block_ids = gpu_block_ids[:success_block_num] valid_cpu_block_ids = cpu_block_ids[:success_block_num] mode = 1 # cpu ==> gpu start_time = time.time() swap_cache_layout( self.gpu_cache_k_tensors, self.storage_key_read_buffer, self.key_cache_shape, valid_gpu_block_ids, valid_cpu_block_ids, self.device, mode, ) swap_cache_layout( self.gpu_cache_v_tensors, self.storage_value_read_buffer, self.value_cache_shape, valid_gpu_block_ids, valid_cpu_block_ids, self.device, mode, ) if k_scale_keys and v_scale_keys: swap_cache_layout( self.gpu_cache_scales_k_tensors, self.storage_key_scale_read_buffer, self.cache_scale_shape, valid_gpu_block_ids, valid_cpu_block_ids, self.device, mode, ) swap_cache_layout( self.gpu_cache_scales_v_tensors, self.storage_value_scale_read_buffer, self.cache_scale_shape, valid_gpu_block_ids, valid_cpu_block_ids, self.device, mode, ) swap_cost_time = time.time() - start_time logger.debug( f"_run_read_storage, swap_cost_time: {swap_cost_time:.6f}s, read_cost_time: {read_cost_time:.6f}s" ) elif self.storage_backend_type == "attention_store": key_cache = [] val_cache = [] for i in range(self.num_layers + self.num_extra_layers): key_cache.append(self.gpu_cache_kvs[f"key_caches_{i}_rank{self.rank}.device{self.device}"]) val_cache.append(self.gpu_cache_kvs[f"value_caches_{i}_rank{self.rank}.device{self.device}"]) start_time = time.time() read_block_num = self.storage_backend.read( task_id, key_cache, val_cache, token_ids, gpu_block_ids, start_read_block_idx, timeout ) read_cost_time = time.time() - start_time valid_gpu_block_ids = gpu_block_ids[:read_block_num] logger.debug(f"_run_read_storage, read_cost_time: {read_cost_time:.6f}s") return valid_gpu_block_ids except Exception as e: logger.error( f"An error occurred in _run_read_storage, " f"error: {e}, traceback:\n{traceback.format_exc()}" ) raise def read_storage_task(self, task: ReadStorageTask): """Read cache from the storage backend to the GPU memory.""" assert ( self.storage_backend ), f"storage_backend not initialized, storage_backend_type: {self.storage_backend_type}" try: gpu_block_ids = task.gpu_block_ids.copy() cpu_block_ids = [i for i in range(len(gpu_block_ids))] k_cache_keys = [f"prefix{self.key_prefix}_{key}_{self.rank}_key" for key in task.keys] v_cache_keys = [f"prefix{self.key_prefix}_{key}_{self.rank}_value" for key in task.keys] if not self.has_cache_scale: k_scale_keys = None v_scale_keys = None else: k_scale_keys = [f"prefix{self.key_prefix}_{key}_{self.rank}_key_scale" for key in task.keys] v_scale_keys = [f"prefix{self.key_prefix}_{key}_{self.rank}_value_scale" for key in task.keys] match_block_num = 0 if self.storage_backend_type in ("mooncake", "file"): match_block_num = self.storage_backend.query( k_cache_keys, v_cache_keys, k_scale_keys, v_scale_keys, task.timeout ) elif self.storage_backend_type == "attention_store": match_block_num = self.storage_backend.query( task.task_id, task.token_ids, task.start_read_block_idx, task.timeout ) logger.info(f"Matched {match_block_num} blocks in cache storage for read task {task.task_id}") k_cache_keys = k_cache_keys[:match_block_num] v_cache_keys = v_cache_keys[:match_block_num] k_scale_keys = k_scale_keys[:match_block_num] if k_scale_keys else None v_scale_keys = v_scale_keys[:match_block_num] if v_scale_keys else None gpu_block_ids = gpu_block_ids[:match_block_num] cpu_block_ids = cpu_block_ids[:match_block_num] valid_gpu_block_ids = [] if match_block_num > 0: # TODO: support timeout with actual block count try: valid_gpu_block_ids = self._run_read_storage( task.task_id, task.token_ids[: match_block_num * self.block_size], task.start_read_block_idx, k_cache_keys, v_cache_keys, k_scale_keys, v_scale_keys, gpu_block_ids, cpu_block_ids, task.timeout, ) logger.info( f"Successfully read {len(valid_gpu_block_ids)} blocks from cache storage for task {task.task_id}" ) except Exception as e: logger.error( f"Failed to read cache for task {task.task_id}, error: {e}, traceback: {traceback.format_exc()}" ) valid_gpu_block_ids = [] finally: try: if (self.rank == 0) and self.storage_backend_type == "attention_store": self.storage_backend.flush_token_index(task.task_id, task.token_ids, 0, True) logger.info(f"Report cache index in HBM to cache storage for task {task.task_id}") except Exception as e: logger.info( f"Failed to report cache index in HBM to cache storage for task {task.task_id}, error: {e}" ) result = (CacheStatus.STORAGE2GPU, task.task_id, task.keys, valid_gpu_block_ids) self.cache_task_queue.swap_storage_to_gpu_barrier.wait() self.cache_task_queue.swap_storage_to_gpu_barrier.reset() self.cache_task_queue.put_transfer_done_signal(result) logger.debug(f"read_storage_task: put transfer done signal for {task.task_id}") except Exception as e: logger.error( f"An error occurred in read_storage_task: " f"task_id: {task.task_id}, error:{e}, {traceback.format_exc()}" ) def _run_write_back_storage( self, task_id, token_ids, start_write_block_idx, k_cache_keys, v_cache_keys, k_scale_keys, v_scale_keys, gpu_block_ids, cpu_block_ids, timeout, ): try: if self.storage_backend_type in ("mooncake", "file"): mode = 0 # gpu ==> cpu start_time = time.time() swap_cache_layout( self.gpu_cache_k_tensors, self.storage_key_write_buffer, self.key_cache_shape, gpu_block_ids, cpu_block_ids, self.device, mode, ) swap_cache_layout( self.gpu_cache_v_tensors, self.storage_value_write_buffer, self.key_cache_shape, gpu_block_ids, cpu_block_ids, self.device, mode, ) if k_scale_keys and v_scale_keys: swap_cache_layout( self.gpu_cache_scales_k_tensors, self.storage_key_scale_write_buffer, self.cache_scale_shape, gpu_block_ids, cpu_block_ids, self.device, mode, ) swap_cache_layout( self.gpu_cache_scales_v_tensors, self.storage_value_scale_write_buffer, self.cache_scale_shape, gpu_block_ids, cpu_block_ids, self.device, mode, ) swap_cost_time = time.time() - start_time block_num = len(gpu_block_ids) keys = k_cache_keys + v_cache_keys k_cache_ptrs = [ self.storage_key_write_buffer + i * self.cache_buffer_stride_bytes for i in cpu_block_ids ] v_cache_ptrs = [ self.storage_value_write_buffer + i * self.cache_buffer_stride_bytes for i in cpu_block_ids ] target_locations = k_cache_ptrs + v_cache_ptrs target_sizes = [self.cache_buffer_stride_bytes] * block_num * 2 # key and value if k_scale_keys and v_scale_keys: keys.extend(k_scale_keys + v_scale_keys) k_scale_ptrs = [ self.storage_key_scale_write_buffer + i * self.scale_buffer_stride_bytes for i in cpu_block_ids ] v_scale_ptrs = [ self.storage_value_scale_write_buffer + i * self.scale_buffer_stride_bytes for i in cpu_block_ids ] target_locations.extend(k_scale_ptrs + v_scale_ptrs) target_sizes.extend([self.scale_buffer_stride_bytes] * block_num * 2) start_time = time.time() self.storage_backend.batch_set(keys=keys, target_locations=target_locations, target_sizes=target_sizes) write_cost_time = time.time() - start_time logger.debug( f"_run_write_back_storage, swap_cost_time: {swap_cost_time:.6f}s, write_cost_time: {write_cost_time:.6f}s" ) return block_num elif self.storage_backend_type == "attention_store": key_cache = [] val_cache = [] for i in range(self.num_layers + self.num_extra_layers): key_cache.append(self.gpu_cache_kvs[f"key_caches_{i}_rank{self.rank}.device{self.device}"]) val_cache.append(self.gpu_cache_kvs[f"value_caches_{i}_rank{self.rank}.device{self.device}"]) start_time = time.time() write_block_num = self.storage_backend.write( task_id, key_cache, val_cache, token_ids, gpu_block_ids, start_write_block_idx, timeout ) write_cost_time = time.time() - start_time logger.debug(f"_run_write_back_storage, write_cost_time: {write_cost_time:.6f}s") return write_block_num except Exception as e: logger.error( f"An error occurred in _run_write_back_storage, " f"error: {e}, traceback:\n{traceback.format_exc()}" ) return 0 def write_back_storage_task(self, task: WriteStorageTask): """ Write cache to the storage backend from the GPU memory. """ assert ( self.storage_backend ), f"storage_backend not initialized, storage_backend_type: {self.storage_backend_type}" try: gpu_block_ids = task.gpu_block_ids.copy() cpu_block_ids = [i for i in range(len(gpu_block_ids))] k_cache_keys = [f"prefix{self.key_prefix}_{key}_{self.rank}_key" for key in task.keys] v_cache_keys = [f"prefix{self.key_prefix}_{key}_{self.rank}_value" for key in task.keys] if not self.has_cache_scale: k_scale_keys = None v_scale_keys = None else: k_scale_keys = [f"prefix{self.key_prefix}_{key}_{self.rank}_key_scale" for key in task.keys] v_scale_keys = [f"prefix{self.key_prefix}_{key}_{self.rank}_value_scale" for key in task.keys] match_block_num = 0 if self.storage_backend_type == ("mooncake", "file"): match_block_num = self.storage_backend.query( k_cache_keys, v_cache_keys, k_scale_keys, v_scale_keys, task.timeout ) elif self.storage_backend_type == "attention_store": match_block_num = self.storage_backend.query(task.task_id, task.token_ids, 0, task.timeout) logger.info(f"Matched {match_block_num} blocks in cache storage for write task {task.task_id}") if match_block_num >= len(k_cache_keys): logger.info(f"No uncached keys found for task {task.task_id}") gpu_block_ids = [] else: try: k_cache_keys = k_cache_keys[match_block_num:] v_cache_keys = v_cache_keys[match_block_num:] k_scale_keys = k_scale_keys[match_block_num:] if k_scale_keys else None v_scale_keys = v_scale_keys[match_block_num:] if v_scale_keys else None gpu_block_ids = gpu_block_ids[match_block_num:] cpu_block_ids = cpu_block_ids[match_block_num:] # TODO: support timeout with actual block count write_block_num = self._run_write_back_storage( task.task_id, task.token_ids, match_block_num, k_cache_keys, v_cache_keys, k_scale_keys, v_scale_keys, gpu_block_ids, cpu_block_ids, task.timeout, ) logger.info( f"Successfully wrote {write_block_num} blocks to cache storage for task {task.task_id}" ) except Exception as e: logger.error(f"Error in write back storage task: {e}, traceback:{traceback.format_exc()}") gpu_block_ids = [] finally: try: if (self.rank == 0) and self.storage_backend_type == "attention_store": self.storage_backend.flush_token_index(task.task_id, task.token_ids, 0, False) logger.info(f"Report cache index out HBM to cache storage for task {task.task_id}") except Exception as e: logger.info( f"Failed to report cache index out HBM to cache storage for task {task.task_id}, error: {e}" ) result = (CacheStatus.GPU2STORAGE, task.task_id, task.keys, gpu_block_ids) self.cache_task_queue.swap_to_storage_barrier.wait() if self.rank == 0: # 只有当rank为0时执行同步操作 self.cache_task_queue.swap_to_storage_barrier.reset() self.cache_task_queue.put_transfer_done_signal(result) # 发送传输完成信号 logger.debug(f"write_back_storage_task: put_transfer_done_signal {result}") except Exception as e: logger.error( f"An error occurred in write_back_storage_task, " f"error: {e}, traceback:\n{traceback.format_exc()}" ) def _do_swap_to_cpu_task( self, swap_node_ids, gpu_block_id, cpu_block_id, event_type, transfer_task_id, ): """ swap cache GPU->CPU """ self.cache_task_queue.swap_to_cpu_barrier1.wait() if self.rank == 0: self.cache_task_queue.swap_to_cpu_barrier1.reset() result = self._transfer_data( swap_node_ids, gpu_block_id, cpu_block_id, event_type, transfer_task_id, ) self.cache_task_queue.swap_to_cpu_barrier2.wait() if self.rank == 0: self.cache_task_queue.swap_to_cpu_barrier2.reset() self.cache_task_queue.put_transfer_done_signal(result) logger.debug(f"_do_swap_to_cpu_task: put_transfer_done_signal {result}") logger.info(f"_do_swap_to_cpu_task: put_transfer_done_signal for transfer_task_id {transfer_task_id}") def _do_swap_to_gpu_task( self, swap_node_ids, gpu_block_id, cpu_block_id, event_type, transfer_task_id, ): """ swap cache CPU->GPU """ self.cache_task_queue.swap_to_gpu_barrier1.wait() if self.rank == 0: self.cache_task_queue.swap_to_gpu_barrier1.reset() result = self._transfer_data( swap_node_ids, gpu_block_id, cpu_block_id, event_type, transfer_task_id, ) self.cache_task_queue.swap_to_gpu_barrier2.wait() if self.rank == 0: self.cache_task_queue.swap_to_gpu_barrier2.reset() self.cache_task_queue.put_transfer_done_signal(result) logger.debug(f"_do_swap_to_gpu_task: put_transfer_done_signal {result}") logger.info(f"_do_swap_to_gpu_task: put_transfer_done_signal for transfer_task_id {transfer_task_id}") def check_work_status(self, time_interval_threashold=envs.FD_CACHE_PROC_EXIT_TIMEOUT): """ Check the health of the model server by checking whether all workers are alive. """ if self.worker_healthy_live_signal.value[0]: elapsed_time = time.time() - self.worker_healthy_live_signal.value[0] if elapsed_time > time_interval_threashold: return False, "Worker Service Not Healthy" return True, "" def submit_task(self, thread_pool: concurrent.futures.ThreadPoolExecutor, task_fn, *args): def inflight_task(fn, *args): try: return fn(*args) finally: self.inflight -= 1 thread_pool.submit(inflight_task, task_fn, *args) def do_data_transfer(self): """ do data transfer task """ consecutive_error_count = 0 max_errors = ( envs.FD_CACHE_PROC_ERROR_COUNT ) # After this many consecutive errors, check if the worker process exists. while True: try: if self.rank == 0: self.cache_task_is_paused_signal.value[0] = 1 if self.is_paused else 0 if self.n_ranks > 1: self.cache_task_queue.barrier0.wait() if self.rank == 0: self.cache_task_queue.barrier0.reset() # Ensure all ranks synchronically do one of the following things: # (1) If rank#0 is paused, wait for a short time and check out rank#0 status again; # (2) otherwise, all ranks are allowed to pull tasks from cache task queue if self.cache_task_is_paused_signal.value[0] == 1: # wait for inflight tasks to finish first while self.inflight != 0: time.sleep(0.1) # mark the current rank as not having inflight tasks self.cache_task_inflight_signal.value[self.rank] = 0 time.sleep(1) continue else: self.cache_task_inflight_signal.value[self.rank] = 1 if self.rank == 0: if not self.cache_task_queue.empty(): self.cache_task_broadcast_signal.value[0] = 1 if self.n_ranks > 1: self.cache_task_queue.barrier1.wait() if self.rank == 0: self.cache_task_queue.barrier1.reset() if self.cache_task_broadcast_signal.value[0] == 1: self.inflight += 1 data, read_finish = self.cache_task_queue.get_transfer_task() logger.debug(f"do_data_transfer: {data}") if read_finish: self.cache_task_broadcast_signal.value[0] = 0 event_type, event_args = data[0], data[1:] if event_type.value == CacheStatus.SWAP2CPU.value: transfer_task_id, swap_node_ids, gpu_block_id, cpu_block_id = event_args self.submit_task( self.swap_to_cpu_thread_pool, self._do_swap_to_cpu_task, swap_node_ids, gpu_block_id, cpu_block_id, event_type, transfer_task_id, ) elif event_type.value == CacheStatus.SWAP2GPU.value: transfer_task_id, swap_node_ids, gpu_block_id, cpu_block_id = event_args self.submit_task( self.swap_to_gpu_thread_pool, self._do_swap_to_gpu_task, swap_node_ids, gpu_block_id, cpu_block_id, event_type, transfer_task_id, ) elif event_type.value == CacheStatus.STORAGE2GPU.value: read_storage_task = event_args[0] self.submit_task( self.read_storage_thread_pool, self.read_storage_task, read_storage_task, ) elif event_type.value == CacheStatus.GPU2STORAGE.value: write_storage_task = event_args[0] self.submit_task( self.write_back_storage_thread_pool, self.write_back_storage_task, write_storage_task, ) else: if self.n_ranks > 1: self.cache_task_queue.barrier2.wait() if self.rank == 0: self.cache_task_queue.barrier2.reset() continue if self.n_ranks > 1: self.cache_task_queue.barrier3.wait() if self.rank == 0: self.cache_task_queue.barrier3.reset() consecutive_error_count = 0 except (BrokenPipeError, EOFError, ConnectionResetError) as e: # When a cache_transfer_manager process remains, it keeps printing error logs and may exhaust disk space. # Add a check to see if the worker process is alive; if it has ended, exit the loop to stop continuous logging. logger.error(f"[CacheTransferManager] Connection broken: {e}") consecutive_error_count += 1 if consecutive_error_count > max_errors: try: status, msg = self.check_work_status() except Exception: status = True if status is False: logger.critical( f"The Worker process has been inactive for over {envs.FD_CACHE_PROC_EXIT_TIMEOUT} seconds, and the Cache process will automatically terminate (the waiting timeout can be extended via FD_CACHE_PROC_EXIT_TIMEOUT)." ) break time.sleep(1) continue except Exception as e: logger.info(f"do_data_transfer: error: {e}, {str(traceback.format_exc())}") def _transfer_data( self, swap_node_ids, task_gpu_block_id, task_cpu_block_id, event_type, transfer_task_id, ): """ transfer data task_gpu_block_id format: [[block_id0, [fold_block_id0, fold_block_id1]], [block_id1, [fold_block_id0, fold_block_id1]], ...] """ logger.debug( f"transfer data: transfer_task_id {transfer_task_id}: swap_node_ids {swap_node_ids}" + f"task_gpu_block_id {task_gpu_block_id} task_cpu_block_id {task_cpu_block_id} event_type {event_type}" ) start_time = time.time() try: # transform block id assert len(task_gpu_block_id) == len(task_cpu_block_id) gpu_block_ids = task_gpu_block_id cpu_block_ids = task_cpu_block_id if event_type.value == CacheStatus.SWAP2CPU.value: swap_cache_all_layers( self.gpu_cache_k_tensors, self.k_dst_ptrs, self.num_cpu_blocks, gpu_block_ids, cpu_block_ids, self.device, 0, ) swap_cache_all_layers( self.gpu_cache_v_tensors, self.v_dst_ptrs, self.num_cpu_blocks, gpu_block_ids, cpu_block_ids, self.device, 0, ) if self.cache_dtype == "block_wise_fp8": swap_cache_all_layers( self.gpu_cache_scales_k_tensors, self.k_scales_ptrs, self.num_cpu_blocks, gpu_block_ids, cpu_block_ids, self.device, 0, ) swap_cache_all_layers( self.gpu_cache_scales_v_tensors, self.v_scales_ptrs, self.num_cpu_blocks, gpu_block_ids, cpu_block_ids, self.device, 0, ) elif event_type.value == CacheStatus.SWAP2GPU.value: swap_cache_all_layers( self.gpu_cache_k_tensors, self.k_dst_ptrs, self.num_cpu_blocks, gpu_block_ids, cpu_block_ids, self.device, 1, ) swap_cache_all_layers( self.gpu_cache_v_tensors, self.v_dst_ptrs, self.num_cpu_blocks, gpu_block_ids, cpu_block_ids, self.device, 1, ) if self.cache_dtype == "block_wise_fp8": swap_cache_all_layers( self.gpu_cache_scales_k_tensors, self.k_scales_ptrs, self.num_cpu_blocks, gpu_block_ids, cpu_block_ids, self.device, 1, ) swap_cache_all_layers( self.gpu_cache_scales_v_tensors, self.v_scales_ptrs, self.num_cpu_blocks, gpu_block_ids, cpu_block_ids, self.device, 1, ) else: logger.warning( f"transfer data: Get unexpected event type {event_type}, only SWAP2CPU and SWAP2GPU supported" ) except Exception as e: logger.error(f"transfer data: error: {e}") raise e end_time = time.time() elasped_time = end_time - start_time logger.info( f"transfer data: transfer_task_id {transfer_task_id} event_type {event_type}: " + f"transfer {len(gpu_block_ids)} blocks done elapsed_time {elasped_time:.4f}" ) return ( event_type, transfer_task_id, swap_node_ids, task_gpu_block_id, task_cpu_block_id, ) def check_cache_status(self, args): # TODO XPU support RL if unset_data_ipc is None: return logger.info("[RL] Launch a thread to clear/restore kv cache when model weights are cleared/updated.") while True: # handle cache clearing/restoring if self.kv_cache_status_signal.value[0] == KVCacheStatus.CLEARING: assert args.splitwise_role == "mixed", "Only mixed mode supports clearing cache." try: # wait for inflight transfer tasks to finish and pause transfer manager self.pause() # clear cpu caches logger.info("[RL] start clearing caches") logger.debug("[RL] start clearing cpu caches") if self.num_cpu_blocks > 0 and envs.FD_ENABLE_SWAP_SPACE_CLEARING: paddle.set_device("cpu") for ptrs in self.k_dst_ptrs + self.v_dst_ptrs: cuda_host_free(ptrs) self.cpu_cache_kvs.clear() self.k_dst_ptrs.clear() self.v_dst_ptrs.clear() if self.cache_dtype == "block_wise_fp8": self.k_scales_ptrs.clear() self.v_scales_ptrs.clear() gc.collect() logger.debug("[RL] successfully cleared cpu caches") # reset swap_space_ready_signal self.swap_space_ready_signal.value[self.rank] = 0 while np.sum(self.swap_space_ready_signal.value) != 0: time.sleep(0.1) logger.debug("[RL] all ranks cleared cpu caches") else: logger.debug("[RL] skip clearing cpu caches") # clear gpu caches logger.debug("[RL] start clearing gpu caches") if args.create_cache_tensor: logger.info("[RL] waiting for gpu runner to unlink cuda ipc") while self.cache_ready_signal.value[self.rank] != 0: time.sleep(0.1) logger.info("[RL] stop waiting! gpu runner has unlinked cuda ipc") paddle.set_device(f"gpu:{self.device}") self.gpu_cache_kvs.clear() self.gpu_cache_k_tensors.clear() self.gpu_cache_v_tensors.clear() if self.cache_dtype == "block_wise_fp8": self.gpu_cache_scales_k_tensors.clear() self.gpu_cache_scales_v_tensors.clear() paddle.device.cuda.empty_cache() logger.debug("[RL] successfully cleared gpu caches") else: for name, tensor in self.gpu_cache_kvs.items(): unset_data_ipc(tensor, name, True, False) logger.debug("[RL] successfully unlinked gpu caches cuda ipc") self.cache_ready_signal.value[self.rank] = 0 while np.sum(self.cache_ready_signal.value) != 0: time.sleep(0.1) logger.info("[RL] all ranks cleared caches!") # reset kv_cache_status_signal self.kv_cache_status_signal.value[0] = KVCacheStatus.CLEARED self._log_memory("after clearing caches") except Exception as e: logger.error(f"[RL] failed to clear caches: {e}") elif self.kv_cache_status_signal.value[0] == KVCacheStatus.UPDATING: assert args.splitwise_role == "mixed", "Only mixed mode supports updating cache." try: # restore cpu cache logger.info("[RL] start restoring caches") logger.debug("[RL] start restoring cpu caches") if self.num_cpu_blocks > 0 and envs.FD_ENABLE_SWAP_SPACE_CLEARING: self._init_cpu_cache(args) logger.debug("[RL] successfully restored cpu caches") while np.sum(self.swap_space_ready_signal.value) != args.mp_num: time.sleep(0.1) logger.debug("[RL] all ranks restored cpu caches") else: logger.debug("[RL] skip restoring cpu caches") # restore gpu cache and set cache_ready_signal logger.debug("[RL] start restoring gpu caches") self._init_gpu_cache(args) logger.debug("[RL] successfully restored gpu caches") if self.storage_backend_type is not None: # use key_prefix to distinguish cache for different version of weight in rl version_file_path = os.path.join(args.model_path, "version.yaml") assert os.path.exists(version_file_path), f"version.yaml not found at {version_file_path}" self.key_prefix = get_key_prefix_from_version(version_file_path) logger.info(f"Update key_prefix of cache storage to {self.key_prefix}") # wait for all ranks caches to be ready while np.sum(self.cache_ready_signal.value) != args.mp_num: time.sleep(0.1) logger.info("[RL] all ranks restored caches!") # resume transfer self.resume() # set kv_cache_status_signal self.kv_cache_status_signal.value[0] = KVCacheStatus.NORMAL self._log_memory("after restoring caches") except Exception as e: logger.error(f"[RL] failed to restore caches: {e}") time.sleep(0.1) def pause(self): if self.n_ranks > 1: self.cache_task_queue.pause_barrier.wait() if self.rank == 0: self.cache_task_queue.pause_barrier.reset() logger.info("[RL] 🟠 wait for inflight transfer tasks to finish") self.is_paused = True while np.sum(self.cache_task_inflight_signal.value) != 0: time.sleep(0.1) logger.info("[RL] 🔴 pause transfer manager and stop do transfer tasks") def resume(self): if self.n_ranks > 1: self.cache_task_queue.resume_barrier.wait() if self.rank == 0: self.cache_task_queue.resume_barrier.reset() self.is_paused = False while np.sum(self.cache_task_inflight_signal.value) != self.n_ranks: time.sleep(0.1) logger.info("[RL] 🟢 resume transfer manager and start to do transfer tasks") def _log_memory(self, context: str): """Log current GPU memory usage.""" max_alloc = paddle.device.cuda.max_memory_allocated() / (1024**3) max_reserved = paddle.device.cuda.max_memory_reserved() / (1024**3) curr_alloc = paddle.device.cuda.memory_allocated() / (1024**3) curr_reserved = paddle.device.cuda.memory_reserved() / (1024**3) logger.warning( f"GPU memory usage {context}:" f"max_allocated: {max_alloc:.2f}GB " f"max_reserved: {max_reserved:.2f}GB " f"current_allocated: {curr_alloc:.2f}GB " f"current_reserved: {curr_reserved:.2f}GB" ) def main(): """ 启动cache manager """ cache_manager = CacheTransferManager(args) cache_manager.do_data_transfer() if __name__ == "__main__": args = parse_args() rank_id = args.rank + args.local_data_parallel_id * args.mp_num if args.mp_num > 1: logger = get_logger("cache_transfer", f"cache_transfer_{rank_id}.log") else: logger = get_logger("cache_transfer", "cache_transfer.log") logger.info(f"args: {vars(args)}") set_device(args.device_id) try: main() except Exception as e: logger.error(f"cache_transfer_manager failed with error: {e}, traceback: {traceback.format_exc()}") raise