def _subshell_eval_callback(self, proc_ctx, cmd_str):
"""Callback when subshell eval command exits"""
self.ready = True
if proc_ctx.ret_code == 0 and proc_ctx.out:
self.remote_command_map[cmd_str] = proc_ctx.out.splitlines()[0]
else:
# Bad status
LOG.error(proc_ctx)
self.remote_command_map[cmd_str] = TaskRemoteMgmtError(
TaskRemoteMgmtError.MSG_SELECT, (cmd_str, None), cmd_str,
proc_ctx.ret_code, proc_ctx.out, proc_ctx.err)
def _remote_host_select_callback(self, proc_ctx, cmd_str):
"""Callback when host select command exits"""
self.ready = True
if proc_ctx.ret_code == 0 and proc_ctx.out:
# Good status
LOG.debug(proc_ctx)
self.remote_host_str_map[cmd_str] = proc_ctx.out.splitlines()[0]
else:
# Bad status
LOG.error(proc_ctx)
self.remote_host_str_map[cmd_str] = TaskRemoteMgmtError(
TaskRemoteMgmtError.MSG_SELECT, (cmd_str, None), cmd_str,
proc_ctx.ret_code, proc_ctx.out, proc_ctx.err)
def _remote_init_callback(self, proc_ctx, platform, tmphandle, curve_auth,
client_pub_key_dir):
"""Callback when "cylc remote-init" exits.
Write public key for install target into client public key
directory.
Set remote_init__map status to REMOTE_INIT_DONE on success which
in turn will trigger file installation to start.
Set remote_init_map status to REMOTE_INIT_FAILED on error.
"""
try:
tmphandle.close()
except OSError: # E.g. ignore bad unlink, etc
pass
install_target = platform['install target']
if proc_ctx.ret_code == 0:
if "KEYSTART" in proc_ctx.out:
regex_result = re.search('KEYSTART((.|\n|\r)*)KEYEND',
proc_ctx.out)
key = regex_result.group(1)
workflow_srv_dir = get_workflow_srv_dir(self.workflow)
public_key = KeyInfo(KeyType.PUBLIC,
KeyOwner.CLIENT,
workflow_srv_dir=workflow_srv_dir,
install_target=install_target)
old_umask = os.umask(0o177)
with open(public_key.full_key_path, 'w',
encoding='utf8') as text_file:
text_file.write(key)
os.umask(old_umask)
# configure_curve must be called every time certificates are
# added or removed, in order to update the Authenticator's
# state.
curve_auth.configure_curve(domain='*',
location=(client_pub_key_dir))
self.remote_init_map[install_target] = REMOTE_INIT_DONE
self.ready = True
return
# Bad status
LOG.error(
TaskRemoteMgmtError(TaskRemoteMgmtError.MSG_INIT, install_target,
' '.join(quote(item) for item in proc_ctx.cmd),
proc_ctx.ret_code, proc_ctx.out, proc_ctx.err))
self.remote_init_map[platform['install target']] = REMOTE_INIT_FAILED
self.ready = True
def file_install(self, platform):
"""Install required files on the remote install target.
Included by default in the file installation:
Files:
.service/server.key (required for ZMQ authentication)
Directories:
app/
bin/
etc/
lib/
"""
install_target = platform['install target']
self.remote_init_map[install_target] = REMOTE_FILE_INSTALL_IN_PROGRESS
src_path = get_workflow_run_dir(self.workflow)
dst_path = get_remote_workflow_run_dir(self.workflow)
install_target = platform['install target']
try:
cmd, host = construct_rsync_over_ssh_cmd(src_path,
dst_path,
platform,
self.rsync_includes,
bad_hosts=self.bad_hosts)
ctx = SubProcContext('file-install', cmd, host)
except NoHostsError:
LOG.error(
TaskRemoteMgmtError(TaskRemoteMgmtError.MSG_INIT,
install_target, '', '', '', ''))
self.remote_init_map[
platform['install target']] = REMOTE_FILE_INSTALL_FAILED
self.bad_hosts -= set(platform['hosts'])
self.ready = True
else:
LOG.debug(f"Begin file installation on {install_target}")
self.proc_pool.put_command(
ctx,
bad_hosts=self.bad_hosts,
callback=self._file_install_callback,
callback_args=[install_target, platform],
callback_255=self._file_install_callback_255,
)
def _remote_init_callback(self, proc_ctx, host, owner, tmphandle):
"""Callback when "cylc remote-init" exits"""
self.ready = True
try:
tmphandle.close()
except OSError: # E.g. ignore bad unlink, etc
pass
if proc_ctx.ret_code == 0:
for status in (REMOTE_INIT_DONE, REMOTE_INIT_NOT_REQUIRED):
if status in proc_ctx.out:
# Good status
LOG.debug(proc_ctx)
self.remote_init_map[(host, owner)] = status
return
# Bad status
LOG.error(
TaskRemoteMgmtError(TaskRemoteMgmtError.MSG_INIT, (host, owner),
' '.join(quote(item) for item in proc_ctx.cmd),
proc_ctx.ret_code, proc_ctx.out, proc_ctx.err))
LOG.error(proc_ctx)
self.remote_init_map[(host, owner)] = REMOTE_INIT_FAILED
def remote_tidy(self):
"""Remove suite contact files from initialised remotes.
Call "cylc remote-tidy".
This method is called on suite shutdown, so we want nothing to hang.
Timeout any incomplete commands after 10 seconds.
Also remove UUID file on suite host ".service/uuid".
"""
# Remove UUID file
uuid_fname = os.path.join(get_suite_srv_dir(self.suite),
FILE_BASE_UUID)
try:
os.unlink(uuid_fname)
except OSError:
pass
# Issue all SSH commands in parallel
procs = {}
for (host, owner), init_with_contact in self.remote_init_map.items():
if init_with_contact != REMOTE_INIT_DONE:
continue
cmd = ['timeout', '10', 'cylc', 'remote-tidy']
if is_remote_host(host):
cmd.append('--host=%s' % host)
if is_remote_user(owner):
cmd.append('--user=%s' % owner)
if cylc.flow.flags.debug:
cmd.append('--debug')
cmd.append(get_remote_suite_run_dir(host, owner, self.suite))
procs[(host, owner)] = (cmd,
Popen(cmd,
stdout=PIPE,
stderr=PIPE,
stdin=DEVNULL))
# Wait for commands to complete for a max of 10 seconds
timeout = time() + 10.0
while procs and time() < timeout:
for (host, owner), (cmd, proc) in procs.copy().items():
if proc.poll() is None:
continue
del procs[(host, owner)]
out, err = (f.decode() for f in proc.communicate())
if proc.wait():
LOG.warning(
TaskRemoteMgmtError(
TaskRemoteMgmtError.MSG_TIDY, (host, owner),
' '.join(quote(item) for item in cmd),
proc.returncode, out, err))
# Terminate any remaining commands
for (host, owner), (cmd, proc) in procs.items():
try:
proc.terminate()
except OSError:
pass
out, err = proc.communicate()
if proc.wait():
LOG.warning(
TaskRemoteMgmtError(TaskRemoteMgmtError.MSG_TIDY,
(host, owner),
' '.join(quote(item) for item in cmd),
proc.returncode, out, err))
def _remote_init_callback(
self, proc_ctx, platform, tmphandle,
curve_auth, client_pub_key_dir):
"""Callback when "cylc remote-init" exits"""
self.ready = True
try:
tmphandle.close()
except OSError: # E.g. ignore bad unlink, etc
pass
self.install_target = platform['install target']
if proc_ctx.ret_code == 0:
if REMOTE_INIT_DONE in proc_ctx.out:
src_path = get_suite_run_dir(self.suite)
dst_path = get_remote_suite_run_dir(platform, self.suite)
try:
process = procopen(construct_rsync_over_ssh_cmd(
src_path,
dst_path,
platform,
self.rsync_includes),
stdoutpipe=True,
stderrpipe=True,
universal_newlines=True)
out, err = process.communicate(timeout=600)
install_target = platform['install target']
if out:
RSYNC_LOG.info(
'File installation information for '
f'{install_target}:\n {out}')
if err:
LOG.error(
'File installation error on '
f'{install_target}:\n {err}')
except Exception as ex:
LOG.error(f"Problem during rsync: {ex}")
self.remote_init_map[self.install_target] = (
REMOTE_INIT_FAILED)
return
if "KEYSTART" in proc_ctx.out:
regex_result = re.search(
'KEYSTART((.|\n|\r)*)KEYEND', proc_ctx.out)
key = regex_result.group(1)
suite_srv_dir = get_suite_srv_dir(self.suite)
public_key = KeyInfo(
KeyType.PUBLIC,
KeyOwner.CLIENT,
suite_srv_dir=suite_srv_dir,
install_target=self.install_target
)
old_umask = os.umask(0o177)
with open(
public_key.full_key_path,
'w', encoding='utf8') as text_file:
text_file.write(key)
os.umask(old_umask)
# configure_curve must be called every time certificates are
# added or removed, in order to update the Authenticator's
# state.
curve_auth.configure_curve(
domain='*', location=(client_pub_key_dir))
for status in (REMOTE_INIT_DONE, REMOTE_INIT_NOT_REQUIRED):
if status in proc_ctx.out:
# Good status
LOG.debug(proc_ctx)
self.remote_init_map[self.install_target] = status
return
# Bad status
LOG.error(TaskRemoteMgmtError(
TaskRemoteMgmtError.MSG_INIT,
platform['install target'], ' '.join(
quote(item) for item in proc_ctx.cmd),
proc_ctx.ret_code, proc_ctx.out, proc_ctx.err))
LOG.error(proc_ctx)
self.remote_init_map[platform['install target']] = REMOTE_INIT_FAILED
def remote_tidy(self):
"""Remove workflow contact files and keys from initialised remotes.
Call "cylc remote-tidy".
This method is called on workflow shutdown, so we want nothing to hang.
Timeout any incomplete commands after 10 seconds.
"""
from cylc.flow.platforms import PlatformLookupError
# Issue all SSH commands in parallel
def construct_remote_tidy_ssh_cmd(install_target, platform):
cmd = ['remote-tidy']
if cylc.flow.flags.verbosity > 1:
cmd.append('--debug')
cmd.append(install_target)
cmd.append(get_remote_workflow_run_dir(self.workflow))
host = get_host_from_platform(platform, bad_hosts=self.bad_hosts)
cmd = construct_ssh_cmd(cmd, platform, host, timeout='10s')
return cmd, host
procs = {}
for install_target, message in self.remote_init_map.items():
if message != REMOTE_FILE_INSTALL_DONE:
continue
if install_target == get_localhost_install_target():
continue
platform = get_random_platform_for_install_target(install_target)
platform_n = platform['name']
try:
cmd, host = construct_remote_tidy_ssh_cmd(
install_target, platform)
except (NoHostsError, PlatformLookupError):
LOG.warning(
TaskRemoteMgmtError(TaskRemoteMgmtError.MSG_TIDY,
platform_n, 1, '', '', 'remote tidy'))
else:
LOG.debug("Removing authentication keys and contact file "
f"from remote: \"{install_target}\"")
procs[platform_n] = (cmd, host,
Popen(cmd,
stdout=PIPE,
stderr=PIPE,
stdin=DEVNULL))
# Wait for commands to complete for a max of 10 seconds
timeout = time() + 10.0
while procs and time() < timeout:
for platform_n, (cmd, host, proc) in procs.copy().items():
if proc.poll() is None:
continue
del procs[platform_n]
out, err = (f.decode() for f in proc.communicate())
# 255 error has to be handled here becuase remote tidy doesn't
# use SubProcPool.
if proc.returncode == 255:
timeout = time() + 10.0
self.bad_hosts.add(host)
LOG.warning(
f'Tried to tidy remote platform: \'{platform_n}\' '
f'using host \'{host}\' but failed; '
'trying a different host')
try:
retry_cmd, host = construct_remote_tidy_ssh_cmd(
install_target, platform)
except (NoHostsError, PlatformLookupError):
LOG.warning(
TaskRemoteMgmtError(TaskRemoteMgmtError.MSG_TIDY,
platform_n, '', '', '', ''))
else:
procs[platform_n] = (retry_cmd, host,
Popen(retry_cmd,
stdout=PIPE,
stderr=PIPE,
stdin=DEVNULL))
if proc.wait() and proc.returncode != 255:
LOG.warning(
TaskRemoteMgmtError(
TaskRemoteMgmtError.MSG_TIDY, platform_n,
' '.join(quote(item) for item in cmd),
proc.returncode, out, err))
# Terminate any remaining commands
for platform_n, (cmd, proc) in procs.items():
with suppress(OSError):
proc.terminate()
out, err = (f.decode() for f in proc.communicate())
if proc.wait():
LOG.warning(
TaskRemoteMgmtError(TaskRemoteMgmtError.MSG_TIDY,
platform_n,
' '.join(quote(item) for item in cmd),
proc.returncode, out, err))
def remote_init(self, platform: Dict[str, Any],
curve_auth: 'ThreadAuthenticator',
client_pub_key_dir: str) -> None:
"""Initialise a remote host if necessary.
Call "cylc remote-init" to install workflow items to remote:
".service/contact": For TCP task communication
"python/": if source exists
Args:
platform: A dict containing settings relating to platform used in
this remote installation.
curve_auth: The ZMQ authenticator.
client_pub_key_dir: Client public key directory, used by the
ZMQ authenticator.
"""
install_target = platform['install target']
if install_target == get_localhost_install_target():
self.remote_init_map[install_target] = REMOTE_FILE_INSTALL_DONE
return
# Set status of install target to in progress while waiting for remote
# initialisation to finish
self.remote_init_map[install_target] = REMOTE_INIT_IN_PROGRESS
# Determine what items to install
comms_meth: CommsMeth = CommsMeth(platform['communication method'])
items = self._remote_init_items(comms_meth)
# Create a TAR archive with the service files,
# so they can be sent later via SSH's STDIN to the task remote.
tmphandle = self.proc_pool.get_temporary_file()
tarhandle = tarfile.open(fileobj=tmphandle, mode='w')
for path, arcname in items:
tarhandle.add(path, arcname=arcname)
tarhandle.close()
tmphandle.seek(0)
# Build the remote-init command to be run over ssh
cmd = ['remote-init']
if cylc.flow.flags.verbosity > 1:
cmd.append('--debug')
cmd.append(str(install_target))
cmd.append(get_remote_workflow_run_dir(self.workflow))
dirs_to_symlink = get_dirs_to_symlink(install_target, self.workflow)
for key, value in dirs_to_symlink.items():
if value is not None:
cmd.append(f"{key}={quote(value)} ")
# Create the ssh command
try:
host = get_host_from_platform(platform, bad_hosts=self.bad_hosts)
except NoHostsError:
LOG.error(
TaskRemoteMgmtError(TaskRemoteMgmtError.MSG_INIT,
install_target,
' '.join(quote(item)
for item in cmd), 42, '', ''))
self.remote_init_map[
platform['install target']] = REMOTE_INIT_FAILED
self.bad_hosts -= set(platform['hosts'])
self.ready = True
else:
cmd = construct_ssh_cmd(cmd, platform, host)
self.proc_pool.put_command(
SubProcContext('remote-init',
cmd,
stdin_files=[tmphandle],
host=host),
bad_hosts=self.bad_hosts,
callback=self._remote_init_callback,
callback_args=[
platform, tmphandle, curve_auth, client_pub_key_dir
],
callback_255=self._remote_init_callback_255,
callback_255_args=[platform])