[BugFix][Mooncake] Use global data_parallel_index for the DP engine index (#48061)
Signed-off-by: Yifan Qiao <yifanqiao@inferact.ai> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
co-authored by
mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent
12292d94b2
commit
ffee32460d
@@ -209,6 +209,7 @@ def _make_vllm_config(
|
||||
parallel_config=SimpleNamespace(
|
||||
pipeline_parallel_size=1,
|
||||
rank=rank,
|
||||
data_parallel_index=0,
|
||||
decode_context_parallel_size=decode_context_parallel_size,
|
||||
prefill_context_parallel_size=1,
|
||||
),
|
||||
@@ -270,7 +271,6 @@ def _patch_worker_runtime(
|
||||
# DCP groups are contiguous splits of the TP group (see
|
||||
# parallel_state.py), so dcp_rank == tp_rank % dcp_size.
|
||||
dcp_group = SimpleNamespace(world_size=dcp_size, rank_in_group=tp_rank % dcp_size)
|
||||
monkeypatch.setattr(worker, "get_mooncake_dp_engine_index", lambda _: 0)
|
||||
monkeypatch.setattr(worker, "get_tensor_model_parallel_rank", lambda: tp_rank)
|
||||
monkeypatch.setattr(worker, "get_tensor_model_parallel_world_size", lambda: tp_size)
|
||||
monkeypatch.setattr(worker, "get_pcp_group", lambda: single_rank_group)
|
||||
|
||||
@@ -8,7 +8,6 @@ import uvicorn
|
||||
from fastapi import FastAPI, HTTPException
|
||||
from pydantic import BaseModel
|
||||
|
||||
from vllm.config import ParallelConfig
|
||||
from vllm.distributed.kv_transfer.kv_connector.utils import EngineId
|
||||
from vllm.logger import init_logger
|
||||
|
||||
@@ -17,15 +16,6 @@ WorkerAddr = str
|
||||
logger = init_logger(__name__)
|
||||
|
||||
|
||||
def get_mooncake_dp_engine_index(parallel_config: ParallelConfig) -> int:
|
||||
"""Return the per-engine DP index used for Mooncake side channels."""
|
||||
if parallel_config.local_engines_only:
|
||||
assert parallel_config.data_parallel_rank_local is not None
|
||||
return parallel_config.data_parallel_rank_local
|
||||
|
||||
return parallel_config.data_parallel_index
|
||||
|
||||
|
||||
class RegisterWorkerPayload(BaseModel):
|
||||
engine_id: EngineId
|
||||
dp_rank: int
|
||||
|
||||
@@ -37,9 +37,6 @@ from vllm.distributed import (
|
||||
)
|
||||
from vllm.distributed.kv_events import BlockStored
|
||||
from vllm.distributed.kv_transfer.kv_connector.v1.mooncake import rdma_utils
|
||||
from vllm.distributed.kv_transfer.kv_connector.v1.mooncake.mooncake_utils import (
|
||||
get_mooncake_dp_engine_index,
|
||||
)
|
||||
from vllm.distributed.kv_transfer.kv_connector.v1.mooncake.store.coordinator import ( # noqa: E501
|
||||
ExternalCachedBlockPool,
|
||||
MooncakeStoreCoordinator,
|
||||
@@ -1156,7 +1153,7 @@ class MooncakeStoreWorker:
|
||||
model_config = vllm_config.model_config
|
||||
parallel_config = vllm_config.parallel_config
|
||||
|
||||
self.dp_rank = get_mooncake_dp_engine_index(parallel_config)
|
||||
self.dp_rank = parallel_config.data_parallel_index
|
||||
self.tp_rank = get_tensor_model_parallel_rank()
|
||||
self.tp_size = get_tensor_model_parallel_world_size()
|
||||
self.pp_size = parallel_config.pipeline_parallel_size
|
||||
@@ -1973,7 +1970,7 @@ class LookupKeyClient:
|
||||
def get_zmq_rpc_path_lookup(vllm_config: VllmConfig) -> str:
|
||||
"""Construct IPC path for ZMQ lookup socket."""
|
||||
assert vllm_config.kv_transfer_config is not None
|
||||
dp_rank = get_mooncake_dp_engine_index(vllm_config.parallel_config)
|
||||
dp_rank = vllm_config.parallel_config.data_parallel_index
|
||||
base_url = envs.VLLM_RPC_BASE_PATH
|
||||
rpc_port = 0
|
||||
hostname = socket.gethostname()
|
||||
|
||||
Reference in New Issue
Block a user