[distributed][misc] error on same VLLM_HOST_IP setting (#7756)
This commit is contained in:
@@ -218,6 +218,19 @@ class RayGPUExecutor(DistributedGPUExecutor):
|
||||
for node_id, gpu_ids in node_gpus.items():
|
||||
node_gpus[node_id] = sorted(gpu_ids)
|
||||
|
||||
all_ips = set(worker_ips + [driver_ip])
|
||||
n_ips = len(all_ips)
|
||||
n_nodes = len(node_workers)
|
||||
|
||||
if n_nodes != n_ips:
|
||||
raise RuntimeError(
|
||||
f"Every node should have a unique IP address. Got {n_nodes}"
|
||||
f" nodes with node ids {list(node_workers.keys())} and "
|
||||
f"{n_ips} unique IP addresses {all_ips}. Please check your"
|
||||
" network configuration. If you set `VLLM_HOST_IP` or "
|
||||
"`HOST_IP` environment variable, make sure it is unique for"
|
||||
" each node.")
|
||||
|
||||
VLLM_INSTANCE_ID = get_vllm_instance_id()
|
||||
|
||||
# Set environment variables for the driver and workers.
|
||||
|
||||
Reference in New Issue
Block a user