Address comments

sky/clouds/runpod.py

@@ -178,8 +178,6 @@ def _get_feasible_launchable_resources(
         self, resources: 'resources_lib.Resources'
     ) -> Tuple[List['resources_lib.Resources'], List[str]]:
         """Returns a list of feasible resources for the given resources."""
-        if resources.use_spot:
-            return ([], [])
         if resources.instance_type is not None:
             assert resources.is_launchable(), resources
             resources = resources.copy(accelerators=None)
@@ -244,7 +242,7 @@ def check_credentials(cls) -> Tuple[bool, Optional[str]]:
             return True, None
 
         except ImportError:
-            return False, ('Failed to import runpod.'
+            return False, ('Failed to import runpod. '
                            'To install, run: pip install skypilot[runpod]')
 
     def get_credential_file_mounts(self) -> Dict[str, str]:

sky/provision/runpod/instance.py

@@ -159,18 +159,18 @@ def get_cluster_info(
     running_instances = _filter_instances(cluster_name_on_cloud, ['RUNNING'])
     instances: Dict[str, List[common.InstanceInfo]] = {}
     head_instance_id = None
-    for node_id, node_info in running_instances.items():
-        instances[node_id] = [
+    for instance_id, instance_info in running_instances.items():
+        instances[instance_id] = [
             common.InstanceInfo(
-                instance_id=node_id,
-                internal_ip=node_info['internal_ip'],
-                external_ip=node_info['external_ip'],
-                ssh_port=node_info['ssh_port'],
+                instance_id=instance_id,
+                internal_ip=instance_info['internal_ip'],
+                external_ip=instance_info['external_ip'],
+                ssh_port=instance_info['ssh_port'],
                 tags={},
             )
         ]
-        if node_info['name'].endswith('-head'):
-            head_instance_id = node_id
+        if instance_info['name'].endswith('-head'):
+            head_instance_id = instance_id
 
     return common.ClusterInfo(
         instances=instances,

sky/provision/runpod/utils.py

@@ -1,7 +1,7 @@
 """RunPod library wrapper for SkyPilot."""
 
 import time
-from typing import Dict, List
+from typing import Any, Dict, List
 
 from sky import sky_logging
 from sky.adaptors import runpod
@@ -64,11 +64,11 @@ def wrapper(*args, **kwargs):
     return wrapper
 
 
-def list_instances() -> Dict[str, dict]:
+def list_instances() -> Dict[str, Dict[str, Any]]:
     """Lists instances associated with API key."""
     instances = runpod.runpod().get_pods()
 
-    instance_dict: Dict[str, dict] = {}
+    instance_dict: Dict[str, Dict[str, Any]] = {}
     for instance in instances:
         info = {}
 

sky/templates/runpod-ray.yml.j2

@@ -72,37 +72,5 @@ setup_commands:
     python3 -c "from sky.skylet.ray_patches import patch; patch()" || exit 1;
     [ -f /etc/fuse.conf ] && sudo sed -i 's/#user_allow_other/user_allow_other/g' /etc/fuse.conf || (sudo sh -c 'echo "user_allow_other" > /etc/fuse.conf');
 
-# Command to start ray on the head node. You don't need to change this.
-# NOTE: these are very performance-sensitive. Each new item opens/closes an SSH
-# connection, which is expensive. Try your best to co-locate commands into fewer
-# items! The same comment applies for worker_start_ray_commands.
-#
-# Increment the following for catching performance bugs easier:
-#   current num items (num SSH connections): 1
-head_start_ray_commands:
-  # NOTE: --disable-usage-stats in `ray start` saves 10 seconds of idle wait.
-  # Line "which prlimit ..": increase the limit of the number of open files for the raylet process, as the `ulimit` may not take effect at this point, because it requires
-  # all the sessions to be reloaded. This is a workaround.
-  - export SKYPILOT_NUM_GPUS=0 && which nvidia-smi > /dev/null && SKYPILOT_NUM_GPUS=$(nvidia-smi --query-gpu=index,name --format=csv,noheader | wc -l);
-    ray stop; RAY_SCHEDULER_EVENTS=0 RAY_DEDUP_LOGS=0 ray start --disable-usage-stats --head --port={{ray_port}} --dashboard-port={{ray_dashboard_port}} --object-manager-port=8076 --autoscaling-config=~/ray_bootstrap_config.yaml {{"--resources='%s'" % custom_resources if custom_resources}} --num-gpus=$SKYPILOT_NUM_GPUS --temp-dir {{ray_temp_dir}} || exit 1;
-    which prlimit && for id in $(pgrep -f raylet/raylet); do sudo prlimit --nofile=1048576:1048576 --pid=$id || true; done;
-    {{dump_port_command}};
-
-# Worker commands are needed for TPU VM Pods
-{%- if num_nodes > 1 or tpu_vm %}
-worker_start_ray_commands:
-  - SKYPILOT_NUM_GPUS=0 && which nvidia-smi > /dev/null && SKYPILOT_NUM_GPUS=$(nvidia-smi --query-gpu=index,name --format=csv,noheader | wc -l);
-    ray stop; RAY_SCHEDULER_EVENTS=0 RAY_DEDUP_LOGS=0 ray start --disable-usage-stats --address=$RAY_HEAD_IP:{{ray_port}} --object-manager-port=8076 {{"--resources='%s'" % custom_resources if custom_resources}} --num-gpus=$SKYPILOT_NUM_GPUS --temp-dir {{ray_temp_dir}} || exit 1;
-    which prlimit && for id in $(pgrep -f raylet/raylet); do sudo prlimit --nofile=1048576:1048576 --pid=$id || true; done;
-{%- else %}
-worker_start_ray_commands: []
-{%- endif %}
-
-head_node: {}
-worker_nodes: {}
-
-# These fields are required for external cloud providers.
-head_setup_commands: []
-worker_setup_commands: []
-cluster_synced_files: []
-file_mounts_sync_continuously: False
+# Command to start ray clusters are now placed in `sky.provision.instance_setup`.
+# We do not need to list it here anymore.