skypilot-nightly 1.0.0.dev20250522__py3-none-any.whl → 1.0.0.dev20250523__py3-none-any.whl

sky/templates/nebius-ray.yml.j2

@@ -46,6 +46,13 @@ available_node_types:
       InstanceType: {{instance_type}}
       ImageId: {{image_id}}
       DiskSize: {{disk_size}}
+      filesystems:
+        {%- for fs in filesystems %}
+        - filesystem_id: {{ fs.filesystem_id }}
+          filesystem_mount_tag: {{ fs.filesystem_mount_tag }}
+          filesystem_attach_mode: {{ fs.filesystem_attach_mode }}
+          filesystem_mount_path: {{ fs.filesystem_mount_path }}
+        {%- endfor %}
       UserData: |
         runcmd:
           - sudo sed -i 's/^#\?AllowTcpForwarding.*/AllowTcpForwarding yes/' /etc/ssh/sshd_config
@@ -130,6 +137,11 @@ setup_commands:
   - {%- for initial_setup_command in initial_setup_commands %}
     {{ initial_setup_command }}
     {%- endfor %}
+    {%- for fs in filesystems %}
+    sudo mkdir {{ fs.filesystem_mount_path }};
+    sudo mount -t virtiofs {{ fs.filesystem_mount_tag }} {{ fs.filesystem_mount_path }};
+    sudo chmod a+w {{ fs.filesystem_mount_path }};
+    {%- endfor %}
     sudo systemctl stop unattended-upgrades || true;
     sudo systemctl disable unattended-upgrades || true;
     sudo sed -i 's/Unattended-Upgrade "1"/Unattended-Upgrade "0"/g' /etc/apt/apt.conf.d/20auto-upgrades || true;

sky/utils/infra_utils.py

@@ -86,6 +86,16 @@ class InfraInfo:
             cloud_name = 'kubernetes'  # Normalize k8s to kubernetes
             region = '/'.join(parts[1:]) if len(parts) >= 2 else None
             zone = None
+        elif cloud_name == 'ssh':
+            # For SSH, the entire string after "ssh/" is the
+            # node pool name. We prepend 'ssh-' for the internal implementation
+            # which reuses the context name.
+            # TODO(romilb): This is a workaround while we use the global
+            # kubeconfig to store the ssh contexts.
+            region = '/'.join(parts[1:]) if len(parts) >= 2 else None
+            if region:
+                region = f'ssh-{region}'
+            zone = None
         else:
             # For non-Kubernetes clouds, continue with regular parsing
             # but be careful to only split into max 3 parts
@@ -133,6 +143,12 @@ class InfraInfo:
         if zone is None:
             zone = '*'
 
+        # If the cloud is ssh, we remove the ssh- prefix from the region
+        # TODO(romilb): This is a workaround while we use the global
+        # kubeconfig to store the ssh contexts.
+        if region and region.startswith('ssh-'):
+            region = region[4:]
+
         # Build the parts list and filter out trailing wildcards
         parts = [cloud.lower(), region, zone]
         while parts and parts[-1] == '*':
@@ -160,7 +176,11 @@ class InfraInfo:
         if self.zone is not None and self.zone != '*':
             region_or_zone = self.zone
         elif self.region is not None and self.region != '*':
-            region_or_zone = self.region
+            # If using region, we remove the ssh- prefix if it exists for SSH
+            # Node Pools.
+            # TODO(romilb): This is a workaround while we use the global
+            # kubeconfig to store the ssh contexts.
+            region_or_zone = self.region.lstrip('ssh-')
 
         if region_or_zone is not None and truncate:
             region_or_zone = common_utils.truncate_long_string(

sky/utils/kubernetes/cleanup-tunnel.sh

@@ -0,0 +1,62 @@
+#!/bin/bash
+# cleanup-tunnel.sh - Script to clean up SSH tunnels for a Kubernetes context
+
+# Usage: cleanup-tunnel.sh CONTEXT_NAME
+
+CONTEXT="${1:-default}"
+TUNNEL_DIR="$HOME/.sky/ssh_node_pools_info"
+PID_FILE="$TUNNEL_DIR/$CONTEXT-tunnel.pid"
+LOG_FILE="$TUNNEL_DIR/$CONTEXT-tunnel.log"
+LOCK_FILE="$TUNNEL_DIR/$CONTEXT-tunnel.lock"
+
+# Get the port from kubeconfig if available
+KUBE_PORT=$(kubectl config view --minify --context="$CONTEXT" -o jsonpath='{.clusters[0].cluster.server}' 2>/dev/null | grep -o ":[0-9]\+" | tr -d ":" || echo "")
+
+if [[ -z "$KUBE_PORT" ]]; then
+    # Default to 6443 if we can't determine the port
+    KUBE_PORT=6443
+    echo "$(date): Could not determine port from kubeconfig, using default port $KUBE_PORT" >> "$LOG_FILE"
+else
+    echo "$(date): Found port $KUBE_PORT in kubeconfig for context $CONTEXT" >> "$LOG_FILE"
+fi
+
+# Check if PID file exists
+if [[ -f "$PID_FILE" ]]; then
+    OLD_PID=$(cat "$PID_FILE")
+    
+    # Log the cleanup attempt
+    echo "$(date): Attempting to clean up tunnel for context $CONTEXT (PID: $OLD_PID, Port: $KUBE_PORT)" >> "$LOG_FILE"
+    
+    # Try to kill the process
+    if kill -0 "$OLD_PID" 2>/dev/null; then
+        # Process exists, kill it
+        kill "$OLD_PID" 2>/dev/null
+        
+        # Wait a moment and check if it's really gone
+        sleep 1
+        if kill -0 "$OLD_PID" 2>/dev/null; then
+            # Still running, force kill
+            kill -9 "$OLD_PID" 2>/dev/null
+            echo "$(date): Forcefully terminated tunnel process $OLD_PID" >> "$LOG_FILE"
+        else
+            echo "$(date): Successfully terminated tunnel process $OLD_PID" >> "$LOG_FILE"
+        fi
+    else
+        echo "$(date): No running process found with PID $OLD_PID" >> "$LOG_FILE"
+    fi
+    
+    # Remove PID file
+    rm -f "$PID_FILE"
+else
+    echo "$(date): No PID file found for context $CONTEXT. Nothing to clean up." >> "$LOG_FILE"
+fi
+
+# Clean up lock file if it exists
+rm -f "$LOCK_FILE"
+
+# Check if port is still in use
+if nc -z localhost "$KUBE_PORT" 2>/dev/null; then
+    echo "$(date): Warning: Port $KUBE_PORT is still in use after cleanup. Another process might be using it." >> "$LOG_FILE"
+fi
+
+echo "$(date): Cleanup complete for context $CONTEXT" >> "$LOG_FILE" 

sky/utils/kubernetes/create_cluster.sh

@@ -85,6 +85,7 @@ fi
 if kind get clusters | grep -q skypilot; then
     echo "Local cluster already exists. Exiting."
     # Switch context to the local cluster
+    kind export kubeconfig --name skypilot
     kubectl config use-context kind-skypilot
     exit 100
 fi