feat: Add RAY_DISABLE_METRICS=1 to disable Ray metrics across cluster configurations and scripts.

rdma_cluster/troubleshooting_rccl.md

@@ -117,6 +117,7 @@ export GLOO_SOCKET_IFNAME=$RDMA_IFACE
 export NCCL_SOCKET_IFNAME=$RDMA_IFACE
 
 # Prevent Ray from masking the APU (Strix Halo Requirement)
+export RAY_DISABLE_METRICS=1
 export RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1
 
 # Start Head
@@ -132,6 +133,7 @@ ray stop --force
 export RDMA_IFACE=$(ip -o addr show to 192.168.100.0/24 | awk '{print $2}' | head -n1)
 export GLOO_SOCKET_IFNAME=$RDMA_IFACE
 export NCCL_SOCKET_IFNAME=$RDMA_IFACE
+export RAY_DISABLE_METRICS=1
 export RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1
 
 ray start --address='192.168.100.1:6379' --num-gpus=1 --num-cpus=8 --disable-usage-stats

scripts/configure_cluster.sh

@@ -54,6 +54,7 @@ setup_head() {
     ray stop --force
 
     # Critical Config
+    export RAY_DISABLE_METRICS=1
     export RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1
     export RAY_memory_monitor_refresh_ms=0
     export VLLM_HOST_IP=$HEAD_IP
@@ -78,6 +79,7 @@ setup_worker() {
     ray stop --force
 
     # Critical Config
+    export RAY_DISABLE_METRICS=1
     export RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1
     export RAY_memory_monitor_refresh_ms=0
     export VLLM_HOST_IP=$WORKER_IP

scripts/start_vllm_cluster.py

@@ -141,6 +141,7 @@ def setup_worker_node(worker_ip, head_ip):
     source /etc/profile
     # Silece the kill command
     ray stop --force > /dev/null 2>&1 || true
+    export RAY_DISABLE_METRICS=1
     export RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1
     export RAY_memory_monitor_refresh_ms=0
     export VLLM_HOST_IP={worker_ip}
@@ -175,6 +176,7 @@ def setup_head_node(head_ip):
     script = f"""
     # Silence the kill command
     ray stop --force > /dev/null 2>&1 || true
+    export RAY_DISABLE_METRICS=1
     export RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1
     export RAY_memory_monitor_refresh_ms=0
     export VLLM_HOST_IP={head_ip}