feat: Fix nsight profiling file sync for multi-node jobs

docs/nsys-profiling.md

@@ -91,7 +91,7 @@ If you are not using model parallelism in Vllm, you should directly refer to `vl
 
 3. **File Location**: Profile files are saved in `/tmp/ray/session*/logs/nsight/` directory on each worker node. Ensure you check both `ls /tmp/ray/session_[0-9]*/logs/nsight` and `ls /tmp/ray/session_latest/logs/nsight` for the profiles, since the "latest" pointer may be stale.
 
-**Note for SLURM users with `ray.sub`**: When using `ray.sub` on SLURM, set `RAY_LOG_SYNC_FREQUENCY=$NUM_SEC` (e.g., `RAY_LOG_SYNC_FREQUENCY=30`) to ensure that the nsight profile files get copied from the container's ephemeral filesystem (`/tmp/ray`) to the persistent `$SLURM_JOB_ID-logs/ray` directory.
+**Note for SLURM users with `ray.sub`**: When using `ray.sub` on SLURM, set `RAY_LOG_SYNC_FREQUENCY=$NUM_SEC` (e.g., `RAY_LOG_SYNC_FREQUENCY=30`) to ensure that the nsight profile files get copied from the container's ephemeral filesystem (`/tmp/ray`) to the persistent directory. The header node's files will be synced to ``$SLURM_JOB_ID-logs/ray`, and other nodes' files will be synced to `$SLURM_JOB_ID-logs/ray/$node_ip/` where `$node_ip` is the IP address of the node.
 
 ## Analyze Profile Files
 

ray.sub

@@ -266,6 +266,37 @@ monitor-sidecar() {
 }
 monitor-sidecar &
 
+# Background process to sync ray logs every $RAY_LOG_SYNC_FREQUENCY seconds
+log-sync-sidecar() {
+  set +x
+  if [[ -z "$RAY_LOG_SYNC_FREQUENCY" ]]; then
+    echo "RAY_LOG_SYNC_FREQUENCY is not set, skipping log sync sidecar"
+    return
+  fi
+  mkdir -p $LOG_DIR/ray/$node_i
+  while true; do
+    sleep $RAY_LOG_SYNC_FREQUENCY
+    if ls /tmp/ray/session_[0-9]* > /dev/null 2>&1; then
+      for session_dir in /tmp/ray/session_[0-9]*/; do
+        if [[ -d "\$session_dir/logs" ]]; then
+          session_name=\$(basename "\$session_dir")
+          mkdir -p "$LOG_DIR/ray/$node_i/\$session_name"
+          if command -v rsync > /dev/null 2>&1; then
+            rsync -ahP "\$session_dir/logs/" $LOG_DIR/ray/$node_i/\$session_name/logs/ 2>/dev/null || true
+          else
+            cp -r "\$session_dir/logs" $LOG_DIR/ray/$node_i/\$session_name/
+          fi
+        fi
+      done
+    fi
+    if [[ -f "$LOG_DIR/ENDED" ]]; then
+      echo "Log sync sidecar terminating..."
+      break
+    fi
+  done
+}
+log-sync-sidecar &
+
 # Patch nsight.py before starting Ray worker
 sed -i 's/context\.py_executable = " "\.join(self\.nsight_cmd) + " python"/context.py_executable = " ".join(self.nsight_cmd) + f" {context.py_executable}"/g' /opt/nemo_rl_venv/lib64/python*/site-packages/ray/_private/runtime_env/nsight.py