|
| 1 | +"""Automatically launches and sets up a Sky onprem cluster on AWS. |
| 2 | +
|
| 3 | +The script allocates a cloud cluster and setups up the cloud cluster to run an |
| 4 | +'onprem' cluster on the cloud. It performs the following steps: |
| 5 | + 1) Launches a cluster on the cloud with `sky launch -c |
| 6 | + on-prem-cluster-[ID]`. The Sky yaml sets up another user named `test`. |
| 7 | + 2) Setups Sky dependencies for the admin user via `sky admin deploy ..`. |
| 8 | + 3) Creates the local cluster config for the regular user, stored in |
| 9 | + `~/.sky/local/...`. |
| 10 | +To clean up the local cluster and the launched cloud cluster, run `sky down |
| 11 | +[LOCAL_CLUSTER] onprem-cluster-[ID]` and remove the corresponding local |
| 12 | +cluster config file. |
| 13 | +
|
| 14 | +Usage: |
| 15 | + # Creates a Sky on-premise cluster named `my-local-cluster` |
| 16 | + python examples/local/launch_cloud_onprem.py my-local-cluster |
| 17 | +""" |
| 18 | + |
| 19 | +import argparse |
| 20 | +import os |
| 21 | +import subprocess |
| 22 | +import tempfile |
| 23 | +import textwrap |
| 24 | +import uuid |
| 25 | +import yaml |
| 26 | + |
| 27 | +from click import testing as cli_testing |
| 28 | + |
| 29 | +from sky import cli |
| 30 | +from sky import global_user_state |
| 31 | +from sky.backends import onprem_utils |
| 32 | +from sky.utils import common_utils |
| 33 | + |
| 34 | +parser = argparse.ArgumentParser() |
| 35 | +parser.add_argument('name', |
| 36 | + type=str, |
| 37 | + default='my-local-cluster', |
| 38 | + nargs='?', |
| 39 | + help='Name of the local cluster.') |
| 40 | +args = parser.parse_args() |
| 41 | + |
| 42 | +# Public and private keys (autogenerated by Sky) |
| 43 | +public_key = '~/.ssh/sky-key.pub' |
| 44 | +private_key = '~/.ssh/sky-key' |
| 45 | + |
| 46 | +local_cluster_name = args.name |
| 47 | + |
| 48 | +# Sky Task YAML to setup the user 'test' |
| 49 | +sky_yaml_config = { |
| 50 | + 'resources': { |
| 51 | + 'cloud': 'aws' |
| 52 | + }, |
| 53 | + 'file_mounts': { |
| 54 | + '/user-key': public_key |
| 55 | + }, |
| 56 | + 'setup': textwrap.dedent("""\ |
| 57 | + sudo adduser --disabled-password --gecos '' test |
| 58 | + sudo -H su test -c 'mkdir ~/.ssh' |
| 59 | + sudo -H su test -c 'chmod 700 ~/.ssh' |
| 60 | + sudo -H su test -c 'touch ~/.ssh/authorized_keys' |
| 61 | + sudo -H su test -c 'chmod 600 ~/.ssh/authorized_keys' |
| 62 | + sudo -H su test -c 'cat /user-key >> ~/.ssh/authorized_keys' |
| 63 | + """) |
| 64 | +} |
| 65 | + |
| 66 | +cli_runner = cli_testing.CliRunner() |
| 67 | +onprem_name = f'onprem-cluster-{uuid.uuid4().hex[:6]}' |
| 68 | +with tempfile.NamedTemporaryFile(suffix='.yaml', mode='w') as f: |
| 69 | + yaml.dump(sky_yaml_config, f) |
| 70 | + file_path = f.name |
| 71 | + cli_runner.invoke(cli.launch, ['-c', onprem_name, file_path]) |
| 72 | + |
| 73 | +handle = global_user_state.get_handle_from_cluster_name(onprem_name) |
| 74 | +head_ip = handle.head_ip |
| 75 | + |
| 76 | +# Cluster config YAML to setup Sky Onprem on the admin account 'ubuntu' |
| 77 | +cluster_yaml_config = { |
| 78 | + 'cluster': { |
| 79 | + 'ips': [head_ip], |
| 80 | + 'name': local_cluster_name |
| 81 | + }, |
| 82 | + 'auth': { |
| 83 | + 'ssh_user': 'ubuntu', |
| 84 | + 'ssh_private_key': private_key |
| 85 | + } |
| 86 | +} |
| 87 | + |
| 88 | +with tempfile.NamedTemporaryFile(suffix='.yaml', mode='w') as f: |
| 89 | + yaml.dump(cluster_yaml_config, f) |
| 90 | + file_path = f.name |
| 91 | + subprocess.run(f'sky admin deploy {file_path}', shell=True, check=True) |
| 92 | + |
| 93 | +# Fill out the local config file in ~/.sky/local/... |
| 94 | +local_config = onprem_utils.get_local_cluster_config_or_error( |
| 95 | + local_cluster_name) |
| 96 | +local_config['auth']['ssh_user'] = 'test' |
| 97 | +local_config['auth']['ssh_private_key'] = private_key |
| 98 | +local_config_path = onprem_utils.SKY_USER_LOCAL_CONFIG_PATH.format( |
| 99 | + local_cluster_name) |
| 100 | +common_utils.dump_yaml(os.path.expanduser(local_config_path), local_config) |
| 101 | + |
| 102 | +print(( |
| 103 | + f'Sky onprem cluster {local_cluster_name} is now ready for use! ' |
| 104 | + f'You can launch jobs with `sky launch -c {local_cluster_name} -- [CMD].` ' |
| 105 | + f'After you are done, shut down the cluster by running `sky down {local_cluster_name} {onprem_name}` ' |
| 106 | + f'and removing the local cluster config in {local_config_path}.')) |
0 commit comments