refactor(disagg): hoist staging helper imports out of the bootstrap loops (#35980)
This commit is contained in:
@@ -29,6 +29,8 @@ from sglang.srt.disaggregation.common.staging_handler import (
|
|||||||
PrefillStagingContext,
|
PrefillStagingContext,
|
||||||
StagingManagerMixin,
|
StagingManagerMixin,
|
||||||
StagingTransferInfo,
|
StagingTransferInfo,
|
||||||
|
handle_staging_rsp,
|
||||||
|
handle_watermark_msg,
|
||||||
)
|
)
|
||||||
from sglang.srt.disaggregation.common.utils import (
|
from sglang.srt.disaggregation.common.utils import (
|
||||||
AuxDataCodec,
|
AuxDataCodec,
|
||||||
@@ -1944,18 +1946,10 @@ class MooncakeKVManager(StagingManagerMixin, CommonKVManager):
|
|||||||
room = waiting_req_bytes[0].decode("ascii")
|
room = waiting_req_bytes[0].decode("ascii")
|
||||||
# Staging: decode reports consumption watermark back to prefill
|
# Staging: decode reports consumption watermark back to prefill
|
||||||
if room == "WATERMARK":
|
if room == "WATERMARK":
|
||||||
from sglang.srt.disaggregation.common.staging_handler import (
|
|
||||||
handle_watermark_msg,
|
|
||||||
)
|
|
||||||
|
|
||||||
handle_watermark_msg(self._staging_ctx, waiting_req_bytes)
|
handle_watermark_msg(self._staging_ctx, waiting_req_bytes)
|
||||||
continue
|
continue
|
||||||
# Staging: decode replies with allocated staging offset
|
# Staging: decode replies with allocated staging offset
|
||||||
if room == "STAGING_RSP":
|
if room == "STAGING_RSP":
|
||||||
from sglang.srt.disaggregation.common.staging_handler import (
|
|
||||||
handle_staging_rsp,
|
|
||||||
)
|
|
||||||
|
|
||||||
handle_staging_rsp(waiting_req_bytes, self.transfer_infos)
|
handle_staging_rsp(waiting_req_bytes, self.transfer_infos)
|
||||||
continue
|
continue
|
||||||
# Decode-side abort notification: mark room as failed and ACK
|
# Decode-side abort notification: mark room as failed and ACK
|
||||||
|
|||||||
@@ -28,6 +28,8 @@ from sglang.srt.disaggregation.common.conn import (
|
|||||||
from sglang.srt.disaggregation.common.staging_handler import (
|
from sglang.srt.disaggregation.common.staging_handler import (
|
||||||
STAGING_WATERMARK_WAIT_S,
|
STAGING_WATERMARK_WAIT_S,
|
||||||
StagingManagerMixin,
|
StagingManagerMixin,
|
||||||
|
handle_staging_rsp,
|
||||||
|
handle_watermark_msg,
|
||||||
)
|
)
|
||||||
from sglang.srt.disaggregation.common.utils import (
|
from sglang.srt.disaggregation.common.utils import (
|
||||||
FastQueue,
|
FastQueue,
|
||||||
@@ -2674,20 +2676,12 @@ class NixlKVManager(StagingManagerMixin, CommonKVManager):
|
|||||||
# Staging: decode reports consumption watermark back to prefill
|
# Staging: decode reports consumption watermark back to prefill
|
||||||
if waiting_req_bytes[0] == b"WATERMARK":
|
if waiting_req_bytes[0] == b"WATERMARK":
|
||||||
if self.enable_staging:
|
if self.enable_staging:
|
||||||
from sglang.srt.disaggregation.common.staging_handler import (
|
|
||||||
handle_watermark_msg,
|
|
||||||
)
|
|
||||||
|
|
||||||
handle_watermark_msg(self._staging_ctx, waiting_req_bytes)
|
handle_watermark_msg(self._staging_ctx, waiting_req_bytes)
|
||||||
continue
|
continue
|
||||||
|
|
||||||
# Staging: decode replies with allocated staging offset
|
# Staging: decode replies with allocated staging offset
|
||||||
if waiting_req_bytes[0] == b"STAGING_RSP":
|
if waiting_req_bytes[0] == b"STAGING_RSP":
|
||||||
if self.enable_staging:
|
if self.enable_staging:
|
||||||
from sglang.srt.disaggregation.common.staging_handler import (
|
|
||||||
handle_staging_rsp,
|
|
||||||
)
|
|
||||||
|
|
||||||
handle_staging_rsp(waiting_req_bytes, self.transfer_infos)
|
handle_staging_rsp(waiting_req_bytes, self.transfer_infos)
|
||||||
continue
|
continue
|
||||||
|
|
||||||
|
|||||||
Reference in New Issue
Block a user