A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 16, 05:45 UTC
[misc] do not read HOST_IP (#10644)
Signed-off-by: youkaichao <youkaichao@gmail.com>
youkaichao committed
Nov 25, 2024 at 17:04 UTC
6e9ff050c8e83ad6d5e5eab621e83549e35933a1
4 files changed
+12
-5
vllm/envs.py
+1
-1
@@ -153,7 +153,7 @@ environment_variables: Dict[str, Callable[[], Any]] = {
153
# If you are using multi-node inference, you should set this differently
154
# on each node.
155
'VLLM_HOST_IP':
156
- lambda: os.getenv('VLLM_HOST_IP', "") or os.getenv("HOST_IP", ""),
156
+ lambda: os.getenv('VLLM_HOST_IP', ""),
157
158
# used in distributed environment to manually set the communication port
159
# Note: if VLLM_PORT is set, and some code asks for multiple ports, the
vllm/executor/ray_gpu_executor.py
+2
-2
@@ -216,8 +216,8 @@ class RayGPUExecutor(DistributedGPUExecutor):
216
f"Every node should have a unique IP address. Got {n_nodes}"
217
f" nodes with node ids {list(node_workers.keys())} and "
218
f"{n_ips} unique IP addresses {all_ips}. Please check your"
219
- " network configuration. If you set `VLLM_HOST_IP` or "
220
- "`HOST_IP` environment variable, make sure it is unique for"
219
+ " network configuration. If you set `VLLM_HOST_IP`"
220
+ " environment variable, make sure it is unique for"
221
" each node.")
222
223
VLLM_INSTANCE_ID = get_vllm_instance_id()
vllm/executor/ray_hpu_executor.py
+2
-2
@@ -192,8 +192,8 @@ class RayHPUExecutor(DistributedGPUExecutor):
192
f"Every node should have a unique IP address. Got {n_nodes}"
193
f" nodes with node ids {list(node_workers.keys())} and "
194
f"{n_ips} unique IP addresses {all_ips}. Please check your"
195
- " network configuration. If you set `VLLM_HOST_IP` or "
196
- "`HOST_IP` environment variable, make sure it is unique for"
195
+ " network configuration. If you set `VLLM_HOST_IP` "
196
+ "environment variable, make sure it is unique for"
197
" each node.")
198
199
VLLM_INSTANCE_ID = get_vllm_instance_id()
vllm/utils.py
+7
@@ -467,6 +467,13 @@ async def collect_from_async_generator(
467
468
def get_ip() -> str:
469
host_ip = envs.VLLM_HOST_IP
470
+ if "HOST_IP" in os.environ and "VLLM_HOST_IP" not in os.environ:
471
+ logger.warning(
472
+ "The environment variable HOST_IP is deprecated and ignored, as"
473
+ " it is often used by Docker and other software to"
474
+ "interact with the container's network stack. Please"
475
+ "use VLLM_HOST_IP instead to set the IP address for vLLM processes"
476
+ " to communicate with each other.")
477
if host_ip:
478
return host_ip
479