[Fluidstack] Improve fluidstack provisioner for corner cases

sky/provision/fluidstack/fluidstack_utils.py

@@ -153,7 +153,9 @@ def create_instance(
                                  auth=(self.api_key, self.api_token),
                                  json=body)
         raise_fluidstack_error(response)
-        return response.json().get('multiple')
+        instance_ids = response.json().get('multiple')
+        assert all(id is not None for id in instance_ids), instance_ids
+        return instance_ids
 
     def list_ssh_keys(self):
         response = requests.get(ENDPOINT + 'ssh',
@@ -231,11 +233,20 @@ def status(self, instance_id: str):
         response = self.info(instance_id)
         return response['status']
 
-    def add_tags(self, instance_id: str, tags: Dict[str, str]):
+    def add_tags(self, instance_id: str, tags: Dict[str, str]) -> str:
         response = requests.patch(
             ENDPOINT + f'server/{instance_id}/tag',
             auth=(self.api_key, self.api_token),
             json=dict(tags=json.dumps(tags)),
         )
         raise_fluidstack_error(response)
         return response.json()
+
+    def rename(self, instance_id: str, hostname: str) -> str:
+        response = requests.patch(
+            ENDPOINT + f'server/{instance_id}/rename',
+            auth=(self.api_key, self.api_token),
+            json=dict(name=hostname),
+        )
+        raise_fluidstack_error(response)
+        return response.json()

sky/provision/fluidstack/instance.py

@@ -43,8 +43,10 @@ def get_internal_ip(node_info: Dict[str, Any]) -> None:
         node_info['internal_ip'] = result[1].strip()
 
 
-def _filter_instances(cluster_name_on_cloud: str,
-                      status_filters: Optional[List[str]]) -> Dict[str, Any]:
+def _filter_instances(
+        cluster_name_on_cloud: str,
+        status_filters: Optional[List[str]],
+        include_instances: Optional[List[str]] = None) -> Dict[str, Any]:
 
     instances = utils.FluidstackClient().list_instances()
     possible_names = [
@@ -56,6 +58,9 @@ def _filter_instances(cluster_name_on_cloud: str,
         if (status_filters is not None and
                 instance['status'] not in status_filters):
             continue
+        if (include_instances is not None and
+                instance['id'] not in include_instances):
+            continue
         if instance.get('hostname') in possible_names:
             filtered_instances[instance['id']] = instance
     return filtered_instances
@@ -86,21 +91,63 @@ def run_instances(region: str, cluster_name_on_cloud: str,
         'reboot',
         'rebooting',
     ]
-
     while True:
         instances = _filter_instances(cluster_name_on_cloud, pending_status)
+        if len(instances) > config.count:
+            raise RuntimeError(
+                f'Cluster {cluster_name_on_cloud} already has '
+                f'{len(instances)} nodes, but {config.count} are '
+                'required. Please try terminate the cluster and retry.')
         if not instances:
             break
-        logger.info(f'Waiting for {len(instances)} instances to be ready.')
+        instance_statuses = [
+            instance['status'] for instance in instances.values()
+        ]
+        logger.info(f'Waiting for {len(instances)} instances to be ready: '
+                    f'{instance_statuses}')
         time.sleep(POLL_INTERVAL)
     exist_instances = _filter_instances(cluster_name_on_cloud, ['running'])
     head_instance_id = _get_head_instance_id(exist_instances)
 
+    def rename(instance_id: str, new_name: str) -> None:
+        try:
+            utils.FluidstackClient().rename(instance_id, new_name)
+        except Exception as e:
+            logger.warning(f'run_instances error: {e}')
+            raise
+
+    for instance_id, instance in exist_instances.items():
+        if head_instance_id is None:
+            # It is possible that head instance does not exist because the
+            # worker instance provisioning succeeded, but failed for the head
+            # instance in a previous launch.
+            head_instance_id = instance_id
+            instance_name = f'{cluster_name_on_cloud}-head'
+            logger.info(f'Renaming head node {head_instance_id} to '
+                        f'{instance_name}')
+            rename(instance_id, instance_name)
+        if (instance_id != head_instance_id and
+                instance['hostname'].endswith('-head')):
+            # Multiple head instances exist.
+            # This is a rare case when the instance name was manually modified
+            # on the cloud or some unexpected behavior happened.
+            # TODO(zhwu): This may not be necessary. An althernative can be
+            # terminating those head instances.
+            instance_name = f'{cluster_name_on_cloud}-worker'
+            logger.info(f'Renaming worker node {instance_id} to '
+                        f'{instance_name}.')
+            try:
+                utils.FluidstackClient().rename(instance_id, instance_name)
+            except Exception as e:  # pylint: disable=broad-except
+                logger.warning(f'run_instances error: {e}')
+                raise
+
     to_start_count = config.count - len(exist_instances)
     if to_start_count < 0:
         raise RuntimeError(
             f'Cluster {cluster_name_on_cloud} already has '
-            f'{len(exist_instances)} nodes, but {config.count} are required.')
+            f'{len(exist_instances)} nodes, but {config.count} are '
+            'required. Please try terminate the cluster and retry.')
     if to_start_count == 0:
         if head_instance_id is None:
             raise RuntimeError(
@@ -134,20 +181,43 @@ def run_instances(region: str, cluster_name_on_cloud: str,
 
     # Wait for instances to be ready.
     while True:
-        instances = _filter_instances(cluster_name_on_cloud, ['running'])
-        ready_instance_cnt = len(instances)
+        instances = _filter_instances(cluster_name_on_cloud,
+                                      pending_status + ['running'])
+        if len(instances) < config.count:
+            # Some of pending instances have been convert to a state that will
+            # not convert to `running` status. This can be due to resource
+            # availability issue.
+            all_instances = _filter_instances(
+                cluster_name_on_cloud,
+                status_filters=None,
+                include_instances=created_instance_ids)
+            all_statuses = [
+                instance['status'] for instance in all_instances.values()
+            ]
+            failed_instance_cnt = config.count - len(instances)
+            logger.error(f'Failed to create {failed_instance_cnt} '
+                         f'instances for cluster {cluster_name_on_cloud}, '
+                         f'with statuses: {all_statuses}')
+            raise RuntimeError(
+                f'Failed to create {failed_instance_cnt} instances, '
+                f'with statuses: {all_statuses}')
+
+        ready_instances = []
+        pending_instances = []
+        for instance in instances.values():
+            if instance['status'] == 'running':
+                ready_instances.append(instance)
+            else:
+                pending_instances.append(instance)
+        ready_instance_cnt = len(ready_instances)
+        pending_statuses = [
+            instance['status'] for instance in pending_instances
+        ]
         logger.info('Waiting for instances to be ready: '
-                    f'({ready_instance_cnt}/{config.count}).')
+                    f'({ready_instance_cnt}/{config.count}).\n'
+                    f'  Pending instance statuses: {pending_statuses}')
         if ready_instance_cnt == config.count:
             break
-        failed_instances = _filter_instances(
-            cluster_name_on_cloud,
-            ['timeout error', 'failed to create', 'out of stock'])
-        if failed_instances:
-            logger.error(f'Failed to create {len(failed_instances)}'
-                         f'instances for cluster {cluster_name_on_cloud}')
-            raise RuntimeError(
-                f'Failed to create {len(failed_instances)} instances.')
 
         time.sleep(POLL_INTERVAL)
     assert head_instance_id is not None, 'head_instance_id should not be None'
@@ -254,6 +324,7 @@ def query_instances(
         'failed to create': status_lib.ClusterStatus.INIT,
         'timeout error': status_lib.ClusterStatus.INIT,
         'out of stock': status_lib.ClusterStatus.INIT,
+        'terminating': None,
         'terminated': None,
     }
     statuses: Dict[str, Optional[status_lib.ClusterStatus]] = {}

sky/provision/instance_setup.py

@@ -24,7 +24,7 @@
 _START_TITLE = '\n' + '-' * 20 + 'Start: {} ' + '-' * 20
 _END_TITLE = '-' * 20 + 'End:   {} ' + '-' * 20 + '\n'
 
-_MAX_RETRY = 5
+_MAX_RETRY = 6
 
 # 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
@@ -200,6 +200,23 @@ def _setup_node(runner: command_runner.SSHCommandRunner,
                                                     stream_logs=False,
                                                     log_path=log_path,
                                                     require_outputs=True)
+            retry_cnt = 0
+            while returncode == 255 and retry_cnt < _MAX_RETRY:
+                # Got network connection issue occur during setup. This could
+                # happen when a setup step requires a reboot, e.g. nvidia-driver
+                # installation (happens for fluidstack). We should retry for it.
+                logger.info('Network connection issue during setup, this is '
+                            'likely due to the reboot of the instance. '
+                            'Retrying setup in 10 seconds.')
+                time.sleep(10)
+                retry_cnt += 1
+                returncode, stdout, stderr = runner.run(cmd,
+                                                        stream_logs=False,
+                                                        log_path=log_path,
+                                                        require_outputs=True)
+                if not returncode:
+                    break
+
             if returncode:
                 raise RuntimeError(
                     'Failed to run setup commands on an instance. '
@@ -283,7 +300,7 @@ def start_ray_on_head_node(cluster_name: str, custom_resource: Optional[str],
                                                 require_outputs=True)
     if returncode:
         raise RuntimeError('Failed to start ray on the head node '
-                           f'(exit code {returncode}). Error: '
+                           f'(exit code {returncode}). Error: \n'
                            f'===== stdout ===== \n{stdout}\n'
                            f'===== stderr ====={stderr}')