def Run(self, args):
"""See ssh_utils.BaseSSHCLICommand.Run."""
on_prem = (
args.IsKnownAndSpecified('network') and
args.IsKnownAndSpecified('region'))
if on_prem:
args.plain = True
# These two lines are needed to ensure reauth is performed as needed, even
# for on-prem, which doesn't use the resulting variables.
holder = base_classes.ComputeApiHolder(self.ReleaseTrack())
client = holder.client
ssh_helper = ssh_utils.BaseSSHCLIHelper()
ssh_helper.Run(args)
oslogin_state = ssh.OsloginState()
if on_prem:
user, ip = ssh_utils.GetUserAndInstance(args.user_host)
remote = ssh.Remote(ip, user)
iap_tunnel_args = iap_tunnel.CreateOnPremSshTunnelArgs(
args, self.ReleaseTrack(), ip)
instance_address = ip
internal_address = ip
else:
user, instance_name = ssh_utils.GetUserAndInstance(args.user_host)
instance_ref = instance_flags.SSH_INSTANCE_RESOLVER.ResolveResources(
[instance_name], compute_scope.ScopeEnum.ZONE, args.zone,
holder.resources,
scope_lister=instance_flags.GetInstanceZoneScopeLister(client))[0]
instance = ssh_helper.GetInstance(client, instance_ref)
project = ssh_helper.GetProject(client, instance_ref.project)
host_keys = ssh_helper.GetHostKeysFromGuestAttributes(
client, instance_ref, instance, project)
iap_tunnel_args = iap_tunnel.CreateSshTunnelArgs(
args, self.ReleaseTrack(), instance_ref,
ssh_utils.GetExternalInterface(instance, no_raise=True))
internal_address = ssh_utils.GetInternalIPAddress(instance)
if args.troubleshoot:
log.status.Print(TROUBLESHOOT_HEADER.format(
instance_ref, args.zone or instance_ref.zone,
datetime.datetime.now()
))
RunTroubleshooting(project, args.zone or instance_ref.zone,
instance, iap_tunnel_args)
return
if not host_keys and host_keys is not None:
log.debug('Unable to retrieve host keys from instance metadata. '
'Continuing.')
expiration, expiration_micros = ssh_utils.GetSSHKeyExpirationFromArgs(
args)
if args.plain:
oslogin_state.oslogin_enabled = False
else:
public_key = ssh_helper.keys.GetPublicKey().ToEntry(
include_comment=True)
# If there is an '@' symbol in the user_host arg, the user is requesting
# to connect as a specific user. This may get overridden by OS Login.
username_requested = '@' in args.user_host
oslogin_state = ssh.GetOsloginState(
instance, project, user, public_key, expiration_micros,
self.ReleaseTrack(), username_requested=username_requested)
user = oslogin_state.user
log.debug(oslogin_state)
if iap_tunnel_args:
# IAP Tunnel only uses instance_address for the purpose of --ssh-flag
# substitution. In this case, dest_addr doesn't do much, it just matches
# against entries in the user's ssh_config file. It's best to use
# something unique to avoid false positive matches, thus we use
# HostKeyAlias.
instance_address = internal_address
dest_addr = ssh_utils.HostKeyAlias(instance)
elif args.internal_ip:
instance_address = internal_address
dest_addr = instance_address
else:
instance_address = ssh_utils.GetExternalIPAddress(instance)
dest_addr = instance_address
remote = ssh.Remote(dest_addr, user)
# identity_file_list will be None if security keys are not enabled.
identity_file_list = ssh.WriteSecurityKeys(oslogin_state)
identity_file = None
options = None
if not args.plain:
if not identity_file_list:
identity_file = ssh_helper.keys.key_file
options = ssh_helper.GetConfig(ssh_utils.HostKeyAlias(instance),
args.strict_host_key_checking,
host_keys_to_add=host_keys)
extra_flags = ssh.ParseAndSubstituteSSHFlags(args, remote, instance_address,
internal_address)
remainder = []
if args.ssh_args:
remainder.extend(args.ssh_args)
# Transform args.command into arg list or None if no command
command_list = args.command.split(' ') if args.command else None
tty = containers.GetTty(args.container, command_list)
remote_command = containers.GetRemoteCommand(args.container, command_list)
# Do not include default port since that will prevent users from
# specifying a custom port (b/121998342).
ssh_cmd_args = {'remote': remote,
'identity_file': identity_file,
'options': options,
'extra_flags': extra_flags,
'remote_command': remote_command,
'tty': tty,
'iap_tunnel_args': iap_tunnel_args,
'remainder': remainder,
'identity_list': identity_file_list}
cmd = ssh.SSHCommand(**ssh_cmd_args)
if args.dry_run:
# Add quotes around any arguments that contain spaces.
log.out.Print(' '.join('"{0}"'.format(arg) if ' ' in arg else arg
for arg in cmd.Build(ssh_helper.env)))
return
# Raise errors if instance requires a security key but the local
# envionment doesn't support them. This is after the 'dry-run' because
# we want to allow printing the command regardless.
if self.enable_security_keys:
ssh_utils.ConfirmSecurityKeyStatus(oslogin_state)
if args.plain or oslogin_state.oslogin_enabled:
keys_newly_added = False
else:
keys_newly_added = ssh_helper.EnsureSSHKeyExists(
client, remote.user, instance, project, expiration=expiration)
if keys_newly_added:
poller = ssh_utils.CreateSSHPoller(remote, identity_file, options,
iap_tunnel_args,
extra_flags=extra_flags)
log.status.Print('Waiting for SSH key to propagate.')
# TODO(b/35355795): Don't force_connect
try:
poller.Poll(
ssh_helper.env,
force_connect=properties.VALUES.ssh.putty_force_connect.GetBool())
except retry.WaitException:
raise ssh_utils.NetworkError()
if args.internal_ip and not on_prem:
ssh_helper.PreliminarilyVerifyInstance(instance.id, remote, identity_file,
options)
# Errors from SSH itself result in an ssh.CommandError being raised
try:
return_code = cmd.Run(
ssh_helper.env,
force_connect=properties.VALUES.ssh.putty_force_connect.GetBool())
except ssh.CommandError as e:
if not on_prem:
log.status.Print(self.createRecommendMessage(args, instance_name,
instance_ref, project))
raise e
if return_code:
# This is the return code of the remote command. Problems with SSH itself
# will result in ssh.CommandError being raised above.
sys.exit(return_code)
def Run(self, args):
"""See ssh_utils.BaseSSHCLICommand.Run."""
holder = base_classes.ComputeApiHolder(self.ReleaseTrack())
client = holder.client
ssh_helper = ssh_utils.BaseSSHCLIHelper()
ssh_helper.Run(args)
user, instance_name = ssh_utils.GetUserAndInstance(args.user_host)
instance_ref = instance_flags.SSH_INSTANCE_RESOLVER.ResolveResources(
[instance_name], compute_scope.ScopeEnum.ZONE, args.zone,
holder.resources,
scope_lister=instance_flags.GetInstanceZoneScopeLister(client))[0]
instance = ssh_helper.GetInstance(client, instance_ref)
project = ssh_helper.GetProject(client, instance_ref.project)
if self.get_host_keys:
host_keys = ssh_helper.GetHostKeysFromGuestAttributes(
client, instance_ref)
if not host_keys:
log.warning('Unable to retrieve host keys from instance metadata. '
'Continuing.')
else:
host_keys = {}
expiration, expiration_micros = ssh_utils.GetSSHKeyExpirationFromArgs(args)
if args.plain:
use_oslogin = False
else:
public_key = ssh_helper.keys.GetPublicKey().ToEntry(include_comment=True)
user, use_oslogin = ssh.CheckForOsloginAndGetUser(
instance, project, user, public_key, expiration_micros,
self.ReleaseTrack())
iap_tunnel_args = iap_tunnel.SshTunnelArgs.FromArgs(
args, self.ReleaseTrack(), instance_ref,
ssh_utils.GetExternalInterface(instance, no_raise=True))
internal_address = ssh_utils.GetInternalIPAddress(instance)
if iap_tunnel_args:
# IAP Tunnel only uses instance_address for the purpose of --ssh-flag
# substitution. In this case, dest_addr doesn't do much, it just matches
# against entries in the user's ssh_config file. It's best to use
# something unique to avoid false positive matches, thus we use
# HostKeyAlias.
instance_address = internal_address
dest_addr = ssh_utils.HostKeyAlias(instance)
elif args.internal_ip:
instance_address = internal_address
dest_addr = instance_address
else:
instance_address = ssh_utils.GetExternalIPAddress(instance)
dest_addr = instance_address
remote = ssh.Remote(dest_addr, user)
identity_file = None
options = None
if not args.plain:
identity_file = ssh_helper.keys.key_file
options = ssh_helper.GetConfig(ssh_utils.HostKeyAlias(instance),
args.strict_host_key_checking,
host_keys_to_add=host_keys)
extra_flags = ssh.ParseAndSubstituteSSHFlags(args, remote, instance_address,
internal_address)
remainder = []
if args.ssh_args:
remainder.extend(args.ssh_args)
# Transform args.command into arg list or None if no command
command_list = args.command.split(' ') if args.command else None
tty = containers.GetTty(args.container, command_list)
remote_command = containers.GetRemoteCommand(args.container, command_list)
# Do not include default port since that will prevent users from
# specifying a custom port (b/121998342).
ssh_cmd_args = {'remote': remote,
'identity_file': identity_file,
'options': options,
'extra_flags': extra_flags,
'remote_command': remote_command,
'tty': tty,
'iap_tunnel_args': iap_tunnel_args,
'remainder': remainder}
cmd = ssh.SSHCommand(**ssh_cmd_args)
if args.dry_run:
log.out.Print(' '.join(cmd.Build(ssh_helper.env)))
return
if args.plain or use_oslogin:
keys_newly_added = False
else:
keys_newly_added = ssh_helper.EnsureSSHKeyExists(
client, remote.user, instance, project, expiration=expiration)
if keys_newly_added:
poller = ssh_utils.CreateSSHPoller(remote, identity_file, options,
iap_tunnel_args,
extra_flags=extra_flags)
log.status.Print('Waiting for SSH key to propagate.')
# TODO(b/35355795): Don't force_connect
try:
poller.Poll(ssh_helper.env, force_connect=True)
except retry.WaitException:
raise ssh_utils.NetworkError()
if args.internal_ip:
ssh_helper.PreliminarilyVerifyInstance(instance.id, remote, identity_file,
options)
# Errors from SSH itself result in an ssh.CommandError being raised
return_code = cmd.Run(ssh_helper.env, force_connect=True)
if return_code:
# This is the return code of the remote command. Problems with SSH itself
# will result in ssh.CommandError being raised above.
sys.exit(return_code)
def Run(self, args):
"""See ssh_utils.BaseSSHCommand.Run."""
holder = base_classes.ComputeApiHolder(self.ReleaseTrack())
client = holder.client
ssh_helper = ssh_utils.BaseSSHHelper()
ssh_helper.Run(args)
ssh_helper.keys.EnsureKeysExist(args.force_key_file_overwrite,
allow_passphrase=True)
remote = ssh.Remote.FromArg(args.user_host)
if not remote:
raise ssh_utils.ArgumentError(
'Expected argument of the form [USER@]INSTANCE. Received [{0}].'
.format(args.user_host))
if not remote.user:
remote.user = ssh.GetDefaultSshUsername()
public_key = ssh_helper.keys.GetPublicKey().ToEntry(include_comment=True)
hostname = '[{0}]:{1}'.format(args.serial_port_gateway, CONNECTION_PORT)
# Update google_compute_known_hosts file with published host key
if args.serial_port_gateway == SERIAL_PORT_GATEWAY:
http_client = http.Http()
http_response = http_client.request(HOST_KEY_URL)
known_hosts = ssh.KnownHosts.FromDefaultFile()
if http_response[0]['status'] == '200':
host_key = http_response[1].strip()
known_hosts.Add(hostname, host_key, overwrite=True)
known_hosts.Write()
elif known_hosts.ContainsAlias(hostname):
log.warning(
'Unable to download and update Host Key for [{0}] from [{1}]. '
'Attempting to connect using existing Host Key in [{2}]. If '
'the connection fails, please try again to update the Host '
'Key.'.format(SERIAL_PORT_GATEWAY, HOST_KEY_URL,
known_hosts.file_path))
else:
known_hosts.Add(hostname, DEFAULT_HOST_KEY)
known_hosts.Write()
log.warning(
'Unable to download Host Key for [{0}] from [{1}]. To ensure '
'the security of the SSH connection, gcloud will attempt to '
'connect using a hard-coded Host Key value. If the connection '
'fails, please try again. If the problem persists, try '
'updating gcloud and connecting again.'
.format(SERIAL_PORT_GATEWAY, HOST_KEY_URL))
instance_ref = instance_flags.SSH_INSTANCE_RESOLVER.ResolveResources(
[remote.host],
compute_scope.ScopeEnum.ZONE,
args.zone,
holder.resources,
scope_lister=instance_flags.GetInstanceZoneScopeLister(client))[0]
instance = ssh_helper.GetInstance(client, instance_ref)
project = ssh_helper.GetProject(client, instance_ref.project)
expiration, expiration_micros = ssh_utils.GetSSHKeyExpirationFromArgs(args)
remote.user, use_os_login = ssh.CheckForOsloginAndGetUser(
instance, project, remote.user, public_key, expiration_micros,
self.ReleaseTrack())
# Determine the serial user, host tuple (remote)
port = 'port={0}'.format(args.port)
constructed_username_list = [instance_ref.project, instance_ref.zone,
instance_ref.Name(), remote.user, port]
if args.extra_args:
for k, v in args.extra_args.items():
constructed_username_list.append('{0}={1}'.format(k, v))
serial_user = '******'.join(constructed_username_list)
serial_remote = ssh.Remote(args.serial_port_gateway, user=serial_user)
identity_file = ssh_helper.keys.key_file
options = ssh_helper.GetConfig(hostname, strict_host_key_checking='yes')
del options['HostKeyAlias']
options['ControlPath'] = 'none'
cmd = ssh.SSHCommand(serial_remote, identity_file=identity_file,
port=CONNECTION_PORT,
options=options)
if args.dry_run:
log.out.Print(' '.join(cmd.Build(ssh_helper.env)))
return
if not use_os_login:
ssh_helper.EnsureSSHKeyExists(
client, remote.user, instance, project, expiration)
# Don't wait for the instance to become SSHable. We are not connecting to
# the instance itself through SSH, so the instance doesn't need to have
# fully booted to connect to the serial port. Also, ignore exit code 255,
# since the normal way to terminate the serial port connection is ~. and
# that causes ssh to exit with 255.
try:
return_code = cmd.Run(ssh_helper.env, force_connect=True)
except ssh.CommandError:
return_code = 255
if return_code:
sys.exit(return_code)
def Run(self, args):
dst = ssh.FileReference.FromPath(args.destination)
srcs = [ssh.FileReference.FromPath(src) for src in args.sources]
ssh.SCPCommand.Verify(srcs, dst, single_remote=True)
if dst.remote:
tpu_name = dst.remote.host
else:
tpu_name = srcs[0].remote.host
# If zone is not set, retrieve the one from the config.
if args.zone is None:
args.zone = properties.VALUES.compute.zone.Get(required=True)
# Retrieve the node.
tpu = tpu_utils.TPUNode(self.ReleaseTrack())
node = tpu.Get(tpu_name, args.zone)
if not tpu_utils.IsTPUVMNode(node):
raise exceptions.BadArgumentException(
'TPU',
'this command is only available for Cloud TPU VM nodes. To access '
'this node, please see '
'https://cloud.google.com/tpu/docs/creating-deleting-tpus.')
worker_ips = tpu_ssh_utils.ParseWorkerFlag(args.worker,
node.networkEndpoints,
args.internal_ip)
if len(worker_ips) > 1 and srcs[0].remote:
raise exceptions.InvalidArgumentException(
'--worker',
'cannot target multiple workers while copying files to '
'client.')
tpu_ssh_utils.ValidateTPUState(node.state,
tpu.messages.Node.StateValueValuesEnum)
# Retrieve GuestAttributes.
single_pod_worker = len(
node.networkEndpoints) > 1 and len(worker_ips) == 1
if single_pod_worker:
# Retrieve only that worker's GuestAttributes.
worker_id = list(worker_ips)[0]
guest_attributes_response = tpu.GetGuestAttributes(
tpu_name, args.zone, six.text_type((worker_id)))
host_key_suffixes = tpu_ssh_utils.GetHostKeySuffixes(
guest_attributes_response.guestAttributes,
len(node.networkEndpoints), worker_id)
else:
# Retrieve the GuestAttributes for all workers in that TPU.
guest_attributes_response = tpu.GetGuestAttributes(
tpu_name, args.zone)
host_key_suffixes = tpu_ssh_utils.GetHostKeySuffixes(
guest_attributes_response.guestAttributes)
# Generate the public key.
ssh_helper = ssh_utils.BaseSSHCLIHelper()
ssh_helper.Run(args)
public_key = ssh_helper.keys.GetPublicKey().ToEntry()
remote = dst.remote or srcs[0].remote
if not dst.remote: # Make sure all remotes point to the same ref.
for src in srcs:
src.remote = remote
if remote.user:
username_requested = True
else:
username_requested = False
remote.user = ssh.GetDefaultSshUsername(warn_on_account_user=True)
project = tpu_utils.GetProject(self.ReleaseTrack(), ssh_helper)
if not args.plain:
# If there is an '@' symbol in the user_host arg, the user is requesting
# to connect as a specific user. This may get overridden by OS Login.
_, expiration_micros = ssh_utils.GetSSHKeyExpirationFromArgs(args)
oslogin_state = ssh.GetOsloginState(
None,
project,
remote.user,
public_key,
expiration_micros,
self.ReleaseTrack(),
username_requested=username_requested,
instance_enable_oslogin=tpu_ssh_utils.TpuHasOsLoginEnabled(
node))
remote.user = oslogin_state.user
# Format the key correctly.
public_key = '{1}:{0} {1}'.format(public_key, remote.user)
if not args.plain and not args.dry_run:
tpu_ssh_utils.AddSSHKeyIfNeeded(project, tpu, node, tpu_name,
args.zone, public_key)
identity_file = None
if not args.plain:
identity_file = ssh_helper.keys.key_file
# If the user's key is not in the SSH agent, the command will stall. We
# want to verify it is added before proceeding, and raise an error if it
# is not.
if not args.dry_run and len(worker_ips) > 1:
tpu_ssh_utils.VerifyKeyInAgent(identity_file)
extra_flags = []
if args.scp_flag:
extra_flags.extend(args.scp_flag)
instance_names = {}
if (args.IsKnownAndSpecified('tunnel_through_iap')
and args.tunnel_through_iap):
# Retrieve the instance names from the GuestAttributes.
for worker in worker_ips:
# The GuestAttributes will only have one entry if we're targeting a
# single worker.
index = 0 if single_pod_worker else worker
instance_name = tpu_ssh_utils.GetFromGuestAttributes(
guest_attributes_response.guestAttributes, index,
'hostname')
if instance_name is None:
log.status.Print('Failed to connect to TPU.')
log.status.Print(tpu_ssh_utils.IAP_TROUBLESHOOTING_HELP)
raise tpu_exceptions.IapTunnelingUnavailable()
instance_names[worker] = instance_name
ssh_threads = []
exit_statuses = [None] * len(worker_ips)
for worker, ips in worker_ips.items():
options = None
if not args.plain:
options = ssh_helper.GetConfig(
tpu_ssh_utils.GetInstanceID(node.id, worker,
host_key_suffixes),
args.strict_host_key_checking, None)
iap_tunnel_args = None
if (args.IsKnownAndSpecified('tunnel_through_iap')
and args.tunnel_through_iap):
# Retrieve the instance name from the GuestAttributes.
instance_name = instance_names[worker]
iap_tunnel_args = tpu_ssh_utils.CreateSshTunnelArgs(
args, self.ReleaseTrack(), project, args.zone,
instance_name)
remote.host = ips.ip_address
cmd = ssh.SCPCommand(srcs,
dst,
identity_file=identity_file,
options=options,
recursive=args.recurse,
compress=args.compress,
extra_flags=extra_flags,
iap_tunnel_args=iap_tunnel_args)
if args.dry_run:
log.out.Print(' '.join(cmd.Build(ssh_helper.env)))
continue
if len(worker_ips) > 1:
# Run the command on multiple workers concurrently.
ssh_threads.append(
threading.Thread(
target=tpu_ssh_utils.AttemptRunWithRetries,
args=('SCP', worker, exit_statuses, cmd,
ssh_helper.env, None, True, SCPRunCmd)))
ssh_threads[-1].start()
else:
# Run on a single worker.
tpu_ssh_utils.AttemptRunWithRetries('SCP', worker,
exit_statuses, cmd,
ssh_helper.env, None,
False, SCPRunCmd)
if len(worker_ips) > 1:
# Wait for all the threads to complete.
for i in range(len(ssh_threads)):
ssh_threads[i].join()
# Exit with a nonzero status, if any.
# This ensures that if any command failed on a worker, we don't end up
# returning 0 for a value.
for status in exit_statuses:
if status:
sys.exit(status)
def RunScp(self,
compute_holder,
args,
port=None,
recursive=False,
compress=False,
extra_flags=None,
release_track=None,
ip_type=ip.IpTypeEnum.EXTERNAL):
"""SCP files between local and remote GCE instance.
Run this method from subclasses' Run methods.
Args:
compute_holder: The ComputeApiHolder.
args: argparse.Namespace, the args the command was invoked with.
port: str or None, Port number to use for SSH connection.
recursive: bool, Whether to use recursive copying using -R flag.
compress: bool, Whether to use compression.
extra_flags: [str] or None, extra flags to add to command invocation.
release_track: obj, The current release track.
ip_type: IpTypeEnum, Specify using internal ip or external ip address.
Raises:
ssh_utils.NetworkError: Network issue which likely is due to failure
of SSH key propagation.
ssh.CommandError: The SSH command exited with SSH exit code, which
usually implies that a connection problem occurred.
"""
if release_track is None:
release_track = base.ReleaseTrack.GA
super(BaseScpHelper, self).Run(args)
dst = ssh.FileReference.FromPath(args.destination)
srcs = [ssh.FileReference.FromPath(src) for src in args.sources]
# Make sure we have a unique remote
ssh.SCPCommand.Verify(srcs, dst, single_remote=True)
remote = dst.remote or srcs[0].remote
if not dst.remote: # Make sure all remotes point to the same ref
for src in srcs:
src.remote = remote
instance_ref = instance_flags.SSH_INSTANCE_RESOLVER.ResolveResources(
[remote.host],
compute_scope.ScopeEnum.ZONE,
args.zone,
compute_holder.resources,
scope_lister=instance_flags.GetInstanceZoneScopeLister(
compute_holder.client))[0]
instance = self.GetInstance(compute_holder.client, instance_ref)
project = self.GetProject(compute_holder.client, instance_ref.project)
expiration, expiration_micros = ssh_utils.GetSSHKeyExpirationFromArgs(
args)
if not remote.user:
remote.user = ssh.GetDefaultSshUsername(warn_on_account_user=True)
if args.plain:
use_oslogin = False
else:
public_key = self.keys.GetPublicKey().ToEntry(include_comment=True)
remote.user, use_oslogin = ssh.CheckForOsloginAndGetUser(
instance, project, remote.user, public_key, expiration_micros,
release_track)
identity_file = None
options = None
if not args.plain:
identity_file = self.keys.key_file
options = self.GetConfig(ssh_utils.HostKeyAlias(instance),
args.strict_host_key_checking)
iap_tunnel_args = iap_tunnel.SshTunnelArgs.FromArgs(
args, release_track, instance_ref,
ssh_utils.GetExternalInterface(instance, no_raise=True))
if iap_tunnel_args:
remote.host = ssh_utils.HostKeyAlias(instance)
elif ip_type is ip.IpTypeEnum.INTERNAL:
remote.host = ssh_utils.GetInternalIPAddress(instance)
else:
remote.host = ssh_utils.GetExternalIPAddress(instance)
cmd = ssh.SCPCommand(srcs,
dst,
identity_file=identity_file,
options=options,
recursive=recursive,
compress=compress,
port=port,
extra_flags=extra_flags,
iap_tunnel_args=iap_tunnel_args)
if args.dry_run:
log.out.Print(' '.join(cmd.Build(self.env)))
return
if args.plain or use_oslogin:
keys_newly_added = False
else:
keys_newly_added = self.EnsureSSHKeyExists(compute_holder.client,
remote.user,
instance,
project,
expiration=expiration)
if keys_newly_added:
poller = ssh_utils.CreateSSHPoller(remote,
identity_file,
options,
iap_tunnel_args,
port=port)
log.status.Print('Waiting for SSH key to propagate.')
# TODO(b/35355795): Don't force_connect
try:
poller.Poll(self.env, force_connect=True)
except retry.WaitException:
raise ssh_utils.NetworkError()
if ip_type is ip.IpTypeEnum.INTERNAL:
# This will never happen when IAP Tunnel is enabled, because ip_type is
# always EXTERNAL when IAP Tunnel is enabled, even if the instance has no
# external IP. IAP Tunnel doesn't need verification because it uses
# unambiguous identifiers for the instance.
self.PreliminarilyVerifyInstance(instance.id, remote,
identity_file, options)
# Errors from the SCP command result in an ssh.CommandError being raised
cmd.Run(self.env, force_connect=True)
def Run(self, args):
user, tpu_name = ssh_utils.GetUserAndInstance(args.user_tpu)
# If zone is not set, retrieve the one from the config.
if args.zone is None:
args.zone = properties.VALUES.compute.zone.Get(required=True)
# Validate the output path.
if args.output_directory:
if not args.command:
raise exceptions.InvalidArgumentException(
'--output_directory',
'cannot be specified without the `--command` '
'flag. Please specify the `--command` flag or remove the '
'--output-directory flag.')
output_directory_path = os.path.abspath(
os.path.expandvars(os.path.expanduser(args.output_directory)))
if not os.path.isdir(output_directory_path):
raise exceptions.InvalidArgumentException(
'--output_directory',
'Failed to find directory {}. Please create '
'it or specify another directory'.format(
output_directory_path))
# Retrieve the node.
tpu = tpu_utils.TPUNode(self.ReleaseTrack())
node = tpu.Get(tpu_name, args.zone)
if not tpu_utils.IsTPUVMNode(node):
raise exceptions.BadArgumentException(
'TPU',
'this command is only available for Cloud TPU VM nodes. To access '
'this node, please see '
'https://cloud.google.com/tpu/docs/creating-deleting-tpus.')
tpu_ssh_utils.ValidateTPUState(node.state,
tpu.messages.Node.StateValueValuesEnum)
worker_ips = tpu_ssh_utils.ParseWorkerFlag(args.worker,
node.networkEndpoints,
args.internal_ip)
if len(worker_ips) > 1 and not args.command:
raise exceptions.InvalidArgumentException(
'--worker',
'cannot target multiple workers without the `--command` '
'flag.')
# Retrieve GuestAttributes.
single_pod_worker = len(
node.networkEndpoints) > 1 and len(worker_ips) == 1
if single_pod_worker:
# Retrieve only that worker's GuestAttributes.
worker_id = list(worker_ips)[0]
guest_attributes_response = tpu.GetGuestAttributes(
tpu_name, args.zone, six.text_type((worker_id)))
host_key_suffixes = tpu_ssh_utils.GetHostKeySuffixes(
guest_attributes_response.guestAttributes,
len(node.networkEndpoints), worker_id)
else:
# Retrieve the GuestAttributes for all workers in that TPU.
guest_attributes_response = tpu.GetGuestAttributes(
tpu_name, args.zone)
host_key_suffixes = tpu_ssh_utils.GetHostKeySuffixes(
guest_attributes_response.guestAttributes)
# Generate the public key.
ssh_helper = ssh_utils.BaseSSHCLIHelper()
ssh_helper.Run(args)
public_key = ssh_helper.keys.GetPublicKey().ToEntry()
project = tpu_utils.GetProject(self.ReleaseTrack(), ssh_helper)
if not args.plain:
# If there is an '@' symbol in the user_host arg, the user is requesting
# to connect as a specific user. This may get overridden by OS Login.
username_requested = '@' in args.user_tpu
_, expiration_micros = ssh_utils.GetSSHKeyExpirationFromArgs(args)
oslogin_state = ssh.GetOsloginState(
None,
project,
user,
public_key,
expiration_micros,
self.ReleaseTrack(),
username_requested=username_requested,
instance_enable_oslogin=tpu_ssh_utils.TpuHasOsLoginEnabled(
node))
user = oslogin_state.user
# Format the key correctly.
public_key = '{1}:{0} {1}'.format(public_key, user)
if not args.plain and not args.dry_run:
tpu_ssh_utils.AddSSHKeyIfNeeded(project, tpu, node, tpu_name,
args.zone, public_key)
command_list = args.command.split(' ') if args.command else None
remainder = []
if args.ssh_args:
remainder.extend(args.ssh_args)
if args.output_directory:
log.status.Print(
'Preparing SSH command execution; output will be logged '
'to {}'.format(output_directory_path))
instance_names = {}
if (args.IsKnownAndSpecified('tunnel_through_iap')
and args.tunnel_through_iap):
# Retrieve the instance names from the GuestAttributes.
for worker in worker_ips:
# The GuestAttributes will only have one entry if we're targeting a
# single worker.
index = 0 if single_pod_worker else worker
instance_name = tpu_ssh_utils.GetFromGuestAttributes(
guest_attributes_response.guestAttributes, index,
'hostname')
if instance_name is None:
log.status.Print('Failed to connect to TPU.')
log.status.Print(tpu_ssh_utils.IAP_TROUBLESHOOTING_HELP)
raise tpu_exceptions.IapTunnelingUnavailable()
instance_names[worker] = instance_name
ssh_threads = []
exit_statuses = [None] * len(worker_ips)
for worker, ips in worker_ips.items():
identity_file = None
options = None
if not args.plain:
identity_file = ssh_helper.keys.key_file
options = ssh_helper.GetConfig(
tpu_ssh_utils.GetInstanceID(node.id, worker,
host_key_suffixes),
args.strict_host_key_checking, None)
remote = ssh.Remote(ips.ip_address, user)
extra_flags = ssh.ParseAndSubstituteSSHFlags(
args, remote, ips.ip_address, ips.internal_address)
iap_tunnel_args = None
if (args.IsKnownAndSpecified('tunnel_through_iap')
and args.tunnel_through_iap):
# Retrieve the instance name from the GuestAttributes.
instance_name = instance_names[worker]
iap_tunnel_args = tpu_ssh_utils.CreateSshTunnelArgs(
args, self.ReleaseTrack(), project, args.zone,
instance_name)
cmd = ssh.SSHCommand(remote=remote,
identity_file=identity_file,
remote_command=command_list,
extra_flags=extra_flags,
options=options,
remainder=remainder,
iap_tunnel_args=iap_tunnel_args)
if args.dry_run:
log.out.Print(' '.join(cmd.Build(ssh_helper.env)))
continue
output_file_writer = None
if args.output_directory:
output_file_writer = FileWriter('{}/{}.log'.format(
output_directory_path, six.text_type(worker)))
if len(worker_ips) > 1:
# Run the command on multiple workers concurrently.
ssh_threads.append(
threading.Thread(
target=tpu_ssh_utils.AttemptRunWithRetries,
args=('SSH', worker, exit_statuses, cmd,
ssh_helper.env, output_file_writer, True,
SSHRunCmd)))
ssh_threads[-1].start()
else:
# Run on a single worker.
tpu_ssh_utils.AttemptRunWithRetries('SSH', worker,
exit_statuses, cmd,
ssh_helper.env,
output_file_writer, False,
SSHRunCmd)
if len(worker_ips) > 1:
# Wait for all the threads to complete.
for i in range(len(ssh_threads)):
ssh_threads[i].join()
# Exit with a nonzero code, if there are any.
# This ensures that if any command failed on a worker, we don't end up
# returning 0 for a value.
for status in exit_statuses:
if status:
sys.exit(status)