[Cudo] Allow opening ports for cudo (#3717)

* Allow opening ports for cudo * fix logging * format * Avoid host controller for cudo * install cudoctl on controller * fix cudoctl installation * update cudo controller message
skypilot-org · Jul 4, 2024 · f7cd5ad · f7cd5ad
1 parent f0f4de8
commit f7cd5ad
Show file tree

Hide file tree

Showing 4 changed files with 25 additions and 10 deletions.
diff --git a/sky/clouds/cudo.py b/sky/clouds/cudo.py
@@ -66,6 +66,10 @@ class Cudo(clouds.Cloud):
         clouds.CloudImplementationFeatures.DOCKER_IMAGE:
             ('Docker image is currently not supported on Cudo. You can try '
              'running docker command inside the `run` section in task.yaml.'),
+        clouds.CloudImplementationFeatures.HOST_CONTROLLERS: (
+            'Cudo Compute cannot host a controller as it does not '
+            'autostopping, which will leave the controller to run indefinitely.'
+        ),
     }
     _MAX_CLUSTER_NAME_LEN_LIMIT = 60
 

diff --git a/sky/provision/cudo/__init__.py b/sky/provision/cudo/__init__.py
@@ -3,6 +3,7 @@
 from sky.provision.cudo.config import bootstrap_instances
 from sky.provision.cudo.instance import cleanup_ports
 from sky.provision.cudo.instance import get_cluster_info
+from sky.provision.cudo.instance import open_ports
 from sky.provision.cudo.instance import query_instances
 from sky.provision.cudo.instance import run_instances
 from sky.provision.cudo.instance import stop_instances
@@ -11,4 +12,4 @@
 
 __all__ = ('bootstrap_instances', 'run_instances', 'stop_instances',
            'terminate_instances', 'wait_instances', 'get_cluster_info',
-           'cleanup_ports', 'query_instances')
+           'cleanup_ports', 'query_instances', 'open_ports')
diff --git a/sky/provision/cudo/instance.py b/sky/provision/cudo/instance.py
@@ -157,11 +157,10 @@ def terminate_instances(
     del provider_config
     instances = _filter_instances(cluster_name_on_cloud, None)
     for inst_id, inst in instances.items():
-        logger.info(f'Terminating instance {inst_id}.'
-                    f'{inst}')
         if worker_only and inst['name'].endswith('-head'):
             continue
-        logger.info(f'Removing {inst_id}: {inst}')
+        logger.debug(f'Terminating Cudo instance {inst_id}.'
+                     f'{inst}')
         cudo_wrapper.remove(inst_id)
 
 
@@ -220,6 +219,16 @@ def query_instances(
     return statuses
 
 
+def open_ports(
+    cluster_name_on_cloud: str,
+    ports: List[str],
+    provider_config: Optional[Dict[str, Any]] = None,
+) -> None:
+    del cluster_name_on_cloud, ports, provider_config
+    # Cudo has all ports open by default. Nothing to do here.
+    return
+
+
 def cleanup_ports(
     cluster_name_on_cloud: str,
     ports: List[str],

diff --git a/sky/utils/controller_utils.py b/sky/utils/controller_utils.py
@@ -247,6 +247,13 @@ def _get_cloud_dependencies_installation_commands(
                 '/bin/linux/amd64/kubectl" && '
                 'sudo install -o root -g root -m 0755 '
                 'kubectl /usr/local/bin/kubectl))')
+        elif isinstance(cloud, clouds.Cudo):
+            commands.append(
+                f'echo -en "\\r{prefix_str}Cudo{empty_str}" && '
+                'pip list | grep cudo-compute > /dev/null 2>&1 || '
+                'pip install "cudo-compute>=0.1.10" > /dev/null 2>&1 && '
+                'wget https://download.cudo.org/compute/cudoctl-0.3.2-amd64.deb -O ~/cudoctl.deb > /dev/null 2>&1 && '  # pylint: disable=line-too-long
+                'sudo dpkg -i ~/cudoctl.deb > /dev/null 2>&1')
         if controller == Controllers.JOBS_CONTROLLER:
             if isinstance(cloud, clouds.IBM):
                 commands.append(
@@ -263,12 +270,6 @@ def _get_cloud_dependencies_installation_commands(
                     f'echo -en "\\r{prefix_str}RunPod{empty_str}" && '
                     'pip list | grep runpod > /dev/null 2>&1 || '
                     'pip install "runpod>=1.5.1" > /dev/null 2>&1')
-            elif isinstance(cloud, clouds.Cudo):
-                # cudo doesn't support open port
-                commands.append(
-                    f'echo -en "\\r{prefix_str}Cudo{empty_str}" && '
-                    'pip list | grep cudo-compute > /dev/null 2>&1 || '
-                    'pip install "cudo-compute>=0.1.8" > /dev/null 2>&1')
     if (cloudflare.NAME
             in storage_lib.get_cached_enabled_storage_clouds_or_refresh()):
         commands.append(f'echo -en "\\r{prefix_str}Cloudflare{empty_str}" && ' +