[KV Transfer] Enable HMA by default for connectors that support it (#41847)

Signed-off-by: Ethan Feng <ethan.fengch@gmail.com>
This commit is contained in:
Ethan Feng
2026-05-26 12:28:51 +02:00
committed by GitHub
parent 97e4022c6c
commit 755043cf3c
7 changed files with 161 additions and 41 deletions
@@ -5,6 +5,7 @@ import importlib
from collections.abc import Callable
from typing import TYPE_CHECKING, cast
from vllm.config.kv_transfer import KVTransferConfig
from vllm.distributed.kv_transfer.kv_connector.base import (
KVConnectorBase,
KVConnectorBaseType,
@@ -18,7 +19,6 @@ from vllm.utils.func_utils import supports_kw
if TYPE_CHECKING:
from vllm.config import VllmConfig
from vllm.config.kv_transfer import KVTransferConfig
from vllm.v1.kv_cache_interface import KVCacheConfig
logger = init_logger(__name__)
@@ -53,7 +53,7 @@ class KVConnectorFactory:
# check if the connector supports HMA
hma_enabled = not config.scheduler_config.disable_hybrid_kv_cache_manager
if hma_enabled and not supports_hma(connector_cls):
if hma_enabled and not cls.supports_hma_config(kv_transfer_config):
raise ValueError(
f"Connector {connector_cls.__name__} does not support HMA but "
f"HMA is enabled. Please set `--disable-hybrid-kv-cache-manager`."
@@ -127,6 +127,23 @@ class KVConnectorFactory:
raise ValueError(f"Unsupported connector type: {connector_name}")
return connector_cls
@classmethod
def supports_hma_config(cls, kv_transfer_config: "KVTransferConfig") -> bool:
"""Return whether this KV transfer config supports HMA.
MultiConnector is a special case: the wrapper class implements
SupportsHMA, but effective support depends on every configured child.
"""
connector_cls = cls.get_connector_class(kv_transfer_config)
if kv_transfer_config.kv_connector != "MultiConnector":
return supports_hma(connector_cls)
from vllm.distributed.kv_transfer.kv_connector.v1.multi_connector import (
MultiConnector,
)
return MultiConnector.all_children_support_hma(kv_transfer_config)
# Register various connectors here.
# The registration should not be done in each individual file, as we want to
@@ -19,7 +19,6 @@ from vllm.distributed.kv_transfer.kv_connector.v1.base import (
KVConnectorRole,
KVConnectorWorkerMetadata,
SupportsHMA,
supports_hma,
)
from vllm.distributed.kv_transfer.kv_connector.v1.metrics import (
KVConnectorPromMetrics,
@@ -151,6 +150,22 @@ class MultiConnector(KVConnectorBase_V1, SupportsHMA):
return True
return False
@classmethod
def all_children_support_hma(cls, kv_transfer_config: "KVTransferConfig") -> bool:
"""Return True only if every configured child connector supports HMA."""
connectors_config = kv_transfer_config.kv_connector_extra_config.get(
"connectors", []
)
if not connectors_config:
return False
for conn_config in connectors_config:
child_config = KVTransferConfig(
**{"engine_id": kv_transfer_config.engine_id, **conn_config}
)
if not KVConnectorFactory.supports_hma_config(child_config):
return False
return True
def __init__(
self,
vllm_config: "VllmConfig",
@@ -169,7 +184,10 @@ class MultiConnector(KVConnectorBase_V1, SupportsHMA):
self._connectors.append(connector_cls(temp_config, role, kv_cache_config))
self._ktc_kv_transfer_config.append(temp_config.kv_transfer_config)
self._all_support_hma = all(supports_hma(c) for c in self._connectors)
assert vllm_config.kv_transfer_config is not None
self._all_support_hma = MultiConnector.all_children_support_hma(
vllm_config.kv_transfer_config
)
assert (
vllm_config.scheduler_config.disable_hybrid_kv_cache_manager
or self._all_support_hma