EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Super tiny rename environment variable (#6648)

Browse Source
This commit is contained in:
fzyzcjy
2025-05-27 12:01:16 +08:00
committed by GitHub
parent 32cd707002
commit 1a8f5f6836
1 changed files with 11 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

19
python/sglang/srt/disaggregation/mooncake/conn.py
View File

@@ -31,7 +31,12 @@ from sglang.srt.disaggregation.base.conn import (
from sglang.srt.disaggregation.mooncake.transfer_engine import MooncakeTransferEngine from sglang.srt.disaggregation.mooncake.transfer_engine import MooncakeTransferEngine
from sglang.srt.disaggregation.utils import DisaggregationMode from sglang.srt.disaggregation.utils import DisaggregationMode
from sglang.srt.server_args import ServerArgs from sglang.srt.server_args import ServerArgs
from sglang.srt.utils import get_free_port, get_ip, get_local_ip_by_remote from sglang.srt.utils import (
get_free_port,
get_int_env_var,
get_ip,
get_local_ip_by_remote,
)
logger = logging.getLogger(__name__) logger = logging.getLogger(__name__)
@@ -172,11 +177,9 @@ class MooncakeKVManager(BaseKVManager):
# Determine the number of threads to use for kv sender # Determine the number of threads to use for kv sender
cpu_count = os.cpu_count() cpu_count = os.cpu_count()
self.executor = concurrent.futures.ThreadPoolExecutor( self.executor = concurrent.futures.ThreadPoolExecutor(
int( get_int_env_var(
os.getenv( "SGLANG_DISAGGREGATION_THREAD_POOL_SIZE",
"DISAGGREGATION_THREAD_POOL_SIZE", min(max(1, cpu_count // 8), 8),
min(max(1, cpu_count // 8), 8),
)
) )
) )
elif self.disaggregation_mode == DisaggregationMode.DECODE: elif self.disaggregation_mode == DisaggregationMode.DECODE:
@@ -187,11 +190,11 @@ class MooncakeKVManager(BaseKVManager):
self.connection_lock = threading.Lock() self.connection_lock = threading.Lock()
# Heartbeat interval should be at least 2 seconds # Heartbeat interval should be at least 2 seconds
self.heartbeat_interval = max( self.heartbeat_interval = max(
float(os.getenv("DISAGGREGATION_HEARTBEAT_INTERVAL", 5.0)), 2.0 float(os.getenv("SGLANG_DISAGGREGATION_HEARTBEAT_INTERVAL", 5.0)), 2.0
) )
# Heartbeat failure should be at least 1 # Heartbeat failure should be at least 1
self.max_failures = max( self.max_failures = max(
int(os.getenv("DISAGGREGATION_HEARTBEAT_MAX_FAILURE", 2)), 1 int(os.getenv("SGLANG_DISAGGREGATION_HEARTBEAT_MAX_FAILURE", 2)), 1
) )
self.start_decode_thread() self.start_decode_thread()
self.connection_pool: Dict[str, Dict[str, Union[str, int]]] = {} self.connection_pool: Dict[str, Dict[str, Union[str, int]]] = {}