Files
sglang/python/sglang/srt/distributed/bootstrap.py
T

381 lines
13 KiB
Python

import logging
import os
import time
from typing import List, Optional
import msgspec
import torch
import torch.distributed as dist
from sglang.srt.configs.model_config import ModelConfig
from sglang.srt.distributed import (
get_default_distributed_backend,
get_pp_group,
get_tp_group,
get_world_group,
init_distributed_environment,
initialize_model_parallel,
set_custom_all_reduce,
set_flashinfer_allreduce_only,
set_mscclpp_all_reduce,
set_torch_symm_mem_all_reduce,
)
from sglang.srt.distributed.gated_launch import maybe_wait_for_gated_launch
from sglang.srt.distributed.parallel_state import (
_tag_groups_for_flashinfer_allreduce_only,
)
from sglang.srt.distributed.parallel_state_wrapper import ParallelState
from sglang.srt.environ import envs
from sglang.srt.layers.dp_attention import initialize_dp_attention
from sglang.srt.layers.layernorm_sp import initialize_layernorm_sp
from sglang.srt.platforms import current_platform
from sglang.srt.runtime_context import (
get_disagg,
get_exec,
get_parallel,
get_serving,
)
from sglang.srt.server_args import ServerArgs
from sglang.srt.utils import (
cpu_has_amx_support,
get_available_gpu_memory,
is_hip,
is_host_cpu_arm64,
is_npu,
monkey_patch_p2p_access_check,
)
from sglang.srt.utils.network import NetworkAddress
from sglang.srt.utils.patch_torch import register_sgl_tp_rank
logger = logging.getLogger(__name__)
_is_cpu_amx_available = cpu_has_amx_support()
_is_cpu_arm64 = is_host_cpu_arm64()
# A representative per-peer payload for materializing the PyNCCL P2P
# connections used by TP LM-head all-to-all. The input/output tensors are
# temporary; NCCL owns the transport resources retained after the warmup.
# In dsv4-pro, assume bs per dp is 120 and the vocab_size is 129280.
# Therefore, the chunk size that each peer sends is 120*129280/8=1.849MB.
# The total warmup bytes per peer should be 1.849*2 = 4MB
_TP_ALL_TO_ALL_WARMUP_BYTES_PER_PEER = 4 << 20
class TorchDistributedResult(msgspec.Struct, frozen=True, kw_only=True):
tp_group: object
pp_group: object
attention_tp_group: object
pre_model_load_memory: float
def init_torch_distributed(
*,
server_args: ServerArgs,
model_config: ModelConfig,
device: str,
ps: ParallelState,
dist_port: int,
is_draft_worker: bool,
local_omp_cpuid: Optional[List[int]],
):
tic = time.perf_counter()
logger.info("Init torch distributed begin.")
backend = _resolve_backend(device=device, server_args=server_args)
before_avail_memory = get_available_gpu_memory(device, ps.gpu_id)
if not get_parallel().enable_p2p_check:
monkey_patch_p2p_access_check()
dist_init_method = _resolve_dist_init_method(dist_port=dist_port)
_set_all_reduce_flags(server_args=server_args)
if not is_draft_worker:
if device == "cpu":
_init_cpu_threads_env(
tp_size=ps.tp_size,
tp_rank=ps.tp_rank,
local_omp_cpuid=local_omp_cpuid,
dist_init_method=dist_init_method,
)
# Only initialize the distributed environment on the target model worker.
_init_parallel_groups(
backend=backend,
dist_init_method=dist_init_method,
server_args=server_args,
model_config=model_config,
gpu_id=ps.gpu_id,
tp_rank=ps.tp_rank,
tp_size=ps.tp_size,
pp_rank=ps.pp_rank,
pp_size=ps.pp_size,
attn_dp_size=ps.attn_dp_size,
attn_cp_size=ps.attn_cp_size,
moe_ep_size=ps.moe_ep_size,
moe_dp_size=ps.moe_dp_size,
dcp_size=ps.attn_dcp_size,
)
# Pre-warm NCCL/RCCL/HCCL to eliminate cold-start latency in first request
# Controlled by --pre-warm-nccl flag (default: enabled on AMD GPUs)
if get_exec().comm.pre_warm_nccl and (
ps.tp_size > 1 or ps.pp_size > 1 or ps.moe_ep_size > 1
):
_prewarm_nccl(
tp_size=ps.tp_size, pp_size=ps.pp_size, moe_ep_size=ps.moe_ep_size
)
# CUDA graph capture enables the PyNCCL communicator for TP LM-head
# all-to-all. Exercise that exact send/recv path before measuring
# pre_model_load_memory so its persistent transport allocations are
# included in later KV-cache sizing instead of appearing during capture.
if (
device == "cuda"
and get_parallel().enable_tp_lm_head_all_to_all
and ps.tp_size > 1
):
_prewarm_tp_lm_head_all_to_all()
maybe_wait_for_gated_launch(
host=get_serving().host, port=get_parallel().gated_launch_port
)
# Draft workers reuse the target pool config and may exist on only one PP stage;
# including them in this WORLD reduction would deadlock on absent peers.
pre_model_load_memory = get_available_gpu_memory(
device,
ps.gpu_id,
distributed=get_world_group().world_size > 1 and not is_draft_worker,
cpu_group=get_world_group().cpu_group,
)
tp_group = get_tp_group()
pp_group = get_pp_group()
attention_tp_group = get_parallel().attn_tp_group
# Check memory for tensor parallelism
local_gpu_memory = get_available_gpu_memory(device, ps.gpu_id)
if ps.tp_size > 1 and not is_draft_worker:
_check_tp_memory_balance(
pre_model_load_memory=pre_model_load_memory,
local_gpu_memory=local_gpu_memory,
)
logger.info(
f"Init torch distributed ends. elapsed={time.perf_counter() - tic:.2f} s, "
f"mem usage={(before_avail_memory - local_gpu_memory):.2f} GB"
)
return TorchDistributedResult(
tp_group=tp_group,
pp_group=pp_group,
attention_tp_group=attention_tp_group,
pre_model_load_memory=pre_model_load_memory,
)
def _resolve_backend(*, device: str, server_args: ServerArgs) -> str:
backend = get_default_distributed_backend(device)
if device == "cuda" and server_args.elastic_ep_backend == "mooncake":
backend = "mooncake"
return backend
def _resolve_dist_init_method(*, dist_port: int) -> str:
# Allow external orchestrators (e.g. trainpi) to override the distributed
# init method. When set to "env://", torch uses MASTER_ADDR/MASTER_PORT
# env-vars and an externally-created TCPStore, completely avoiding port
# conflicts with intra-host collocation.
dist_init_method_override = envs.SGLANG_DISTRIBUTED_INIT_METHOD_OVERRIDE.get()
if dist_init_method_override:
dist_init_method = dist_init_method_override
elif get_parallel().dist_init_addr:
na = NetworkAddress.parse(get_parallel().dist_init_addr)
dist_init_method = na.to_tcp()
else:
dist_init_method = NetworkAddress(
get_serving().host or "127.0.0.1", dist_port
).to_tcp()
return dist_init_method
def _set_all_reduce_flags(*, server_args: ServerArgs) -> None:
set_custom_all_reduce(not get_exec().comm.disable_custom_all_reduce)
set_mscclpp_all_reduce(server_args.enable_mscclpp)
set_torch_symm_mem_all_reduce(get_exec().comm.enable_torch_symm_mem)
set_flashinfer_allreduce_only(
get_exec().comm.flashinfer_allreduce_fusion_backend is not None
)
def _set_shm_master_env(dist_init_method: Optional[str]) -> None:
# setdefault so an explicit user-provided MASTER_ADDR/MASTER_PORT wins.
prefix = "tcp://"
if (
dist_init_method
and dist_init_method.startswith(prefix)
and ":" in dist_init_method[len(prefix) :]
):
host, port = dist_init_method[len(prefix) :].rsplit(":", 1)
os.environ.setdefault("MASTER_ADDR", host)
os.environ.setdefault("MASTER_PORT", port)
def _init_cpu_threads_env(
*,
tp_size: int,
tp_rank: int,
local_omp_cpuid: Optional[List[int]],
dist_init_method: Optional[str] = None,
) -> None:
if _is_cpu_amx_available or _is_cpu_arm64:
# Bind OpenMP threads to CPU cores
torch.ops.sgl_kernel.init_cpu_threads_env(local_omp_cpuid)
# Set local size to hint SGLang to use shared memory based AllReduce
os.environ["LOCAL_SIZE"] = str(tp_size)
# shm.cpp names its /dev/shm segments from MASTER_ADDR/MASTER_PORT.
# Feed each engine's unique dist_init_method (tcp://host:port) into
# these env vars so co-located engines get distinct segment names and
# don't collide.
_set_shm_master_env(dist_init_method)
torch.ops.sgl_kernel.initialize(tp_size, tp_rank)
else:
logger.warning(
"init_cpu_threads_env and shared memory based AllReduce is disabled, only intel amx backend and arm64 are supported"
)
def _init_parallel_groups(
*,
backend: str,
dist_init_method: str,
server_args: ServerArgs,
model_config: ModelConfig,
gpu_id: int,
tp_rank: int,
tp_size: int,
pp_rank: int,
pp_size: int,
attn_dp_size: int,
attn_cp_size: int,
moe_ep_size: int,
moe_dp_size: int,
dcp_size: int,
) -> None:
is_ep_joiner = server_args.is_ep_joiner
is_scale_joiner = server_args.is_ep_scale_joiner
rank_offset = get_parallel().ep_join_rank_offset if is_scale_joiner else 0
world_size = (
rank_offset + tp_size * pp_size if is_scale_joiner else tp_size * pp_size
)
rank = rank_offset + tp_size * pp_rank + tp_rank
init_distributed_environment(
backend=backend,
world_size=world_size,
rank=rank,
local_rank=gpu_id,
distributed_init_method=dist_init_method,
timeout=get_parallel().dist_timeout,
moe_a2a_backend=get_exec().moe.moe_a2a_backend,
recovered_rank=is_ep_joiner,
max_world_size=get_parallel().max_ep_size,
)
initialize_model_parallel(
tensor_model_parallel_size=tp_size,
attention_data_parallel_size=attn_dp_size,
pipeline_model_parallel_size=pp_size,
expert_model_parallel_size=moe_ep_size,
attention_context_model_parallel_size=attn_cp_size,
moe_data_model_parallel_size=moe_dp_size,
decode_context_parallel_size=dcp_size,
duplicate_tp_group=get_disagg().enable_pdmux,
duplicate_attn_cp_group=(
is_hip()
and server_args.enable_two_batch_overlap
and get_parallel().enable_dsa_prefill_context_parallel
),
enable_symm_mem=get_exec().comm.enable_symm_mem,
recovered_rank=is_ep_joiner,
rank_offset=rank_offset,
max_world_size=get_parallel().max_ep_size,
)
_tag_groups_for_flashinfer_allreduce_only()
initialize_dp_attention(
server_args=server_args,
model_config=model_config,
)
initialize_layernorm_sp(
server_args=server_args,
model_config=model_config,
)
if is_npu():
register_sgl_tp_rank(gpu_id)
def _prewarm_nccl(*, tp_size: int, pp_size: int, moe_ep_size: int) -> None:
warmup_start = time.perf_counter()
tp_group_handle = get_tp_group().device_group
# Single warmup all_reduce to initialize NCCL/RCCL/HCCL communicator
warmup_tensor = torch.zeros(1, device=torch.cuda.current_device())
dist.all_reduce(warmup_tensor, group=tp_group_handle)
current_platform.synchronize()
warmup_elapsed = time.perf_counter() - warmup_start
logger.info(
f"NCCL/RCCL/HCCL warmup completed in {warmup_elapsed:.3f}s "
f"(tp_size={tp_size}, pp_size={pp_size}, ep_size={moe_ep_size})"
)
def _prewarm_tp_lm_head_all_to_all() -> None:
"""Materialize PyNCCL P2P resources before model-memory accounting."""
warmup_start = time.perf_counter()
tp_group = get_tp_group()
pynccl_comm = tp_group.pynccl_comm
if pynccl_comm is None or not pynccl_comm.available:
raise RuntimeError(
"--enable-tp-lm-head-all-to-all requires an available PyNCCL "
"communicator for CUDA graph capture."
)
numel = tp_group.world_size * _TP_ALL_TO_ALL_WARMUP_BYTES_PER_PEER
warmup_input = torch.empty(numel, dtype=torch.uint8, device=tp_group.device)
warmup_output = torch.empty_like(warmup_input)
# PyNCCL is disabled outside graph-capture contexts by default. Enable it
# explicitly so eager startup does not fall back to ProcessGroupNCCL and
# miss the P2P resources required by the captured all-to-all.
with pynccl_comm.change_state(enable=True):
pynccl_comm.all_to_all_single(warmup_output, warmup_input)
current_platform.synchronize()
del warmup_input, warmup_output
current_platform.empty_cache()
warmup_elapsed = time.perf_counter() - warmup_start
logger.info(
"TP LM-head PyNCCL all-to-all warmup completed in %.3fs "
"(tp_size=%d, bytes_per_peer=%d)",
warmup_elapsed,
tp_group.world_size,
_TP_ALL_TO_ALL_WARMUP_BYTES_PER_PEER,
)
def _check_tp_memory_balance(
*, pre_model_load_memory: float, local_gpu_memory: float
) -> None:
if pre_model_load_memory < local_gpu_memory * 0.9:
msg = "The memory capacity is unbalanced. Some GPUs may be occupied by other processes. "
msg += (
f"{pre_model_load_memory=}, {local_gpu_memory=}, {local_gpu_memory * 0.9=}"
)
if envs.SGLANG_ENABLE_TP_MEMORY_INBALANCE_CHECK.get():
raise RuntimeError(msg)
else:
logger.warning(msg)