chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

[HotFix]: Add missing SGLANG_EPLB_HEATMAP_COLLECTION_INTERVAL env var (#12776)

Browse Source
This commit is contained in:
hzh0425
2025-11-07 10:58:44 +08:00
committed by GitHub
parent d9c812d881
commit ed5e905c89
2 changed files with 4 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
python/sglang/srt/environ.py
View File

@@ -215,6 +215,7 @@ class Envs:
SGLANG_EXPERT_LOCATION_UPDATER_LOG_METRICS = EnvBool(False)
SGLANG_LOG_EXPERT_LOCATION_METADATA = EnvBool(False)
SGLANG_EXPERT_DISTRIBUTION_RECORDER_DIR = EnvStr("/tmp")
SGLANG_EPLB_HEATMAP_COLLECTION_INTERVAL = EnvInt(0)
# TBO
SGLANG_TBO_DEBUG = EnvBool(False)

9
python/sglang/srt/eplb/expert_distribution.py
View File

@@ -31,7 +31,7 @@ from sglang.srt.environ import envs
from sglang.srt.metrics.collector import ExpertDispatchCollector
from sglang.srt.model_executor.forward_batch_info import ForwardBatch
from sglang.srt.server_args import ServerArgs
from sglang.srt.utils import Withable, is_npu
from sglang.srt.utils import Withable, get_int_env_var, is_npu
_is_npu = is_npu()
@@ -716,11 +716,8 @@ class _UtilizationRateAccumulatorMixin(_Accumulator):
def _collect_metrics_if_needed(self, gpu_physical_count: torch.Tensor):
# sglang:eplb_gpu_physical_count metric is disabled if SGLANG_EPLB_HEATMAP_COLLECTION_INTERVAL <= 0
if (
envs.SGLANG_EPLB_HEATMAP_COLLECTION_INTERVAL > 0
and self._collection_counter % envs.SGLANG_EPLB_HEATMAP_COLLECTION_INTERVAL
== 0
):
interval = get_int_env_var("SGLANG_EPLB_HEATMAP_COLLECTION_INTERVAL", 0)
if interval > 0 and self._collection_counter % interval == 0:
for layer_idx in range(self._expert_location_metadata.num_layers):
count_of_layer = (
self._expert_dispatch_collector.eplb_gpu_physical_count.labels(