EngineX/xc-llm-ascend
3
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

[bugfix] adapt_remote_request_id (#6051)

Browse Source 
This PR addresses a request ID mismatch issue in the PD
(Prefill-Decoding) separation deployment scenario for vllm-ascend.
Upstream vLLM recently mitigated request ID collisions by appending a
random suffix to each request_id (e.g., req-123 → req-123-abc), refer to
[PR-27987](https://github.com/vllm-project/vllm/pull/27987 ) &
[PR-29665](https://github.com/vllm-project/vllm/pull/29665). While this
works in single-node deployments, it breaks compatibility in
PD-separated setups: the Producer (Prefill node) and Consumer (Decoding
node) end up with different request_id values, preventing the Consumer
from correctly retrieving the KV cache generated by the Producer.
To resolve this, this PR introduces a new field remote_request_id in the
metadata passed via mooncake_connector. The Producer preserves and
forwards the original (unmodified) request_id as remote_request_id. The
Consumer then uses this remote_request_id—instead of its locally
generated suffixed ID—to fetch the correct KV cache from the Prefill
node.
This ensures consistent request identification across PD nodes while
maintaining compatibility with upstream vLLM’s request ID deduplication
mechanism.
<img width="1279" height="781" alt="image"
src="https://github.com/user-attachments/assets/274238c1-dab6-4d3a-9ee4-6e578679b762"
/>

- vLLM version: v0.13.0
- vLLM main:
d68209402d

Signed-off-by: ghphotoframe <854746559@qq.com>
Co-authored-by: jiangweixiang <jwx02384838@antgroup.com>
This commit is contained in:
JiangWeixiang
2026-01-22 10:48:40 +08:00
committed by GitHub
parent ef9d8367f5
commit cef04b3555
1 changed files with 17 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
vllm_ascend/distributed/kv_transfer/kv_p2p/mooncake_connector.py
View File

@@ -76,6 +76,7 @@ class ReqMeta:
remote_host: str
remote_port: int
remote_engine_id: str
remote_request_id: str
remote_pcp_size: int
remote_dcp_size: int
remote_multi_nodes_meta_mapping: dict[str, dict[str, Any]]
@@ -375,6 +376,7 @@ class KVCacheRecvingThread(threading.Thread):
def add_request(self,
request_id: str,
remote_request_id: str,
local_block_ids: list[int],
remote_block_ids: list[int],
remote_engine_id: str,
@@ -391,6 +393,7 @@ class KVCacheRecvingThread(threading.Thread):
"local_block_ids": local_block_ids,
"remote_block_ids": remote_block_ids,
"remote_engine_id": remote_engine_id,
"remote_request_id": remote_request_id,
"remote_host": remote_host,
"remote_handshake_port": remote_handshake_port,
"offset": offset,
@@ -423,6 +426,7 @@ class KVCacheRecvingThread(threading.Thread):
def _handle_request(self, req_meta: dict[str, Any]):
request_id = req_meta["request_id"]
remote_request_id = req_meta["remote_request_id"]
remote_host = req_meta["remote_host"]
remote_handshake_port = req_meta["remote_handshake_port"]
remote_port_send_num = req_meta["remote_port_send_num"]
@@ -430,14 +434,14 @@ class KVCacheRecvingThread(threading.Thread):
try:
logger.debug(
f"Starting to transfer KV cache for request {request_id}.")
f"Starting to transfer KV cache for request {remote_request_id}.")
self._transfer_kv_cache(req_meta)
logger.debug(
f"Finished transferring KV cache for request {request_id}.")
f"Finished transferring KV cache for request {remote_request_id}.")
except Exception as e:
logger.error(
"Failed to transfer KV cache for request "
f"{request_id}: {e}",
f"{remote_request_id}: {e}",
exc_info=True)
finally:
if all_task_done:
@@ -448,10 +452,10 @@ class KVCacheRecvingThread(threading.Thread):
# Always send the done signal to the remote host to ensure proper
# resource cleanup. Failing to do so may cause a memory leak on the
# remote host.
self._send_done_recv_signal(request_id, remote_host,
self._send_done_recv_signal(remote_request_id, remote_host,
remote_handshake_port,
remote_port_send_num)
self._send_done_signal_to_free_remote_port(request_id, remote_host,
self._send_done_signal_to_free_remote_port(remote_request_id, remote_host,
remote_port_send_num)
def _send_done_signal_to_free_remote_port(self, request_id, remote_host,
@@ -472,7 +476,7 @@ class KVCacheRecvingThread(threading.Thread):
def _transfer_kv_cache(self, req_meta: dict[str, Any]):
"""Handle a KV cache transfer request."""
request_id = req_meta["request_id"]
remote_request_id = req_meta["remote_request_id"]
remote_block_ids = req_meta["remote_block_ids"]
local_block_ids = req_meta["local_block_ids"]
remote_engine_id = req_meta["remote_engine_id"]
@@ -558,7 +562,7 @@ class KVCacheRecvingThread(threading.Thread):
dst_list, length_list)
if ret < 0:
logger.error("Mooncake transfer failed for request %s",
req_meta["request_id"])
req_meta["remote_request_id"])
raise RuntimeError(f"Mooncake transfer failed, ret: {ret}")
req_end_time = time.perf_counter()
@@ -566,7 +570,7 @@ class KVCacheRecvingThread(threading.Thread):
logger.info(
"KV cache transfer for request %s took %.2f ms (%d groups,"
" %d blocks). local_ip %s local_device_id %s remote_session_id %s",
request_id, req_transfer_elapsed, num_transfer_groups, num_blocks,
remote_request_id, req_transfer_elapsed, num_transfer_groups, num_blocks,
get_ip(), self.tp_rank, session_id)
# Determine if the current position is the offset position at the end of
@@ -791,6 +795,7 @@ class MooncakeConnectorMetadata(KVConnectorMetadata):
num_external_tokens=num_external_tokens,
remote_block_ids=kv_transfer_params["remote_block_ids"],
remote_engine_id=kv_transfer_params["remote_engine_id"],
remote_request_id=kv_transfer_params["remote_request_id"],
remote_host=kv_transfer_params["remote_host"],
remote_port=kv_transfer_params["remote_port"],
remote_pcp_size=kv_transfer_params.get("remote_pcp_size", 1),
@@ -996,7 +1001,7 @@ class MooncakeConnectorScheduler:
if params is not None and params.get("do_remote_prefill"):
if params.get("remote_block_ids"):
if all(p in params for p in ("remote_engine_id", "remote_host",
"remote_port")):
"remote_port", "remote_request_id")):
local_block_ids = (blocks.get_unhashed_block_ids()
if num_external_tokens > 0 else [])
# Get unhashed blocks to pull from remote.
@@ -1074,6 +1079,7 @@ class MooncakeConnectorScheduler:
do_remote_decode=False,
remote_block_ids=computed_block_ids,
remote_engine_id=self.engine_id,
remote_request_id=request.request_id,
remote_host=self.side_channel_host,
remote_port=self.side_channel_port,
remote_pcp_size=self.pcp_size,
@@ -1583,6 +1589,7 @@ class MooncakeConnectorWorker:
meta.remote_multi_nodes_meta_mapping)
self.kv_recv_thread.add_request(
request_id=req_id,
remote_request_id=meta.remote_request_id,
local_block_ids=local_block_ids_list[pcp_dcp_rank],
remote_block_ids=remote_block_ids_list[
pcp_dcp_rank],
@@ -1610,6 +1617,7 @@ class MooncakeConnectorWorker:
meta.remote_multi_nodes_meta_mapping)
self.kv_recv_thread.add_request(
request_id=req_id,
remote_request_id=meta.remote_request_id,
local_block_ids=meta.local_block_ids,
remote_block_ids=meta.remote_block_ids,
remote_engine_id=remote_engine_id,