@tduncan / Nvidia-Optimization-Stack-VLLM Mirror · read-only / commits / 2410e320b3
Fork 0

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 17, 04:15 UTC

fix `get_ip` error in pure ipv6 environment (#2931)

Jingru committed Feb 27, 2024 at 11:22 UTC 2410e320b35cd704059b7c6ba8d8ba7643fe46ee
1 file changed +19 -5
vllm/utils.py
+19 -5
@@ -162,9 +162,16 @@ def make_async(func: Callable[..., T]) -> Callable[..., Awaitable[T]]:
162
163
164 def get_ip() -> str:
165 + # try ipv4
166 s = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
166 - s.connect(("8.8.8.8", 80)) # Doesn't need to be reachable
167 - return s.getsockname()[0]
167 + try:
168 + s.connect(("dns.google", 80)) # Doesn't need to be reachable
169 + return s.getsockname()[0]
170 + except OSError:
171 + # try ipv6
172 + s = socket.socket(socket.AF_INET6, socket.SOCK_DGRAM)
173 + s.connect(("dns.google", 80))
174 + return s.getsockname()[0]
175
176
177 def get_distributed_init_method(ip: str, port: int) -> str:
@@ -172,9 +179,16 @@ def get_distributed_init_method(ip: str, port: int) -> str:
179
180
181 def get_open_port() -> int:
175 - with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
176 - s.bind(("", 0))
177 - return s.getsockname()[1]
182 + # try ipv4
183 + try:
184 + with socket.socket(socket.AF_INET, socket.SOCK_STREAM) as s:
185 + s.bind(("", 0))
186 + return s.getsockname()[1]
187 + except OSError:
188 + # try ipv6
189 + with socket.socket(socket.AF_INET6, socket.SOCK_STREAM) as s:
190 + s.bind(("", 0))
191 + return s.getsockname()[1]
192
193
194 def set_cuda_visible_devices(device_ids: List[int]) -> None: