async def test_load_contact_file_async(myflow):
cont = await load_contact_file_async(myflow.suite)
assert cont[CFF.HOST] == myflow.host
# compare the async interface to the sync interface
cont2 = load_contact_file(myflow.suite)
assert cont == cont2
def get_location(suite: str, owner: str, host: str):
"""Extract host and port from a suite's contact file.
NB: if it fails to load the suite contact file, it will exit.
Args:
suite (str): suite name
owner (str): owner of the suite
host (str): host name
Returns:
Tuple[str, int, int]: tuple with the host name and port numbers.
Raises:
ClientError: if the suite is not running.
"""
try:
contact = load_contact_file(suite, owner, host)
except SuiteServiceFileError:
raise ClientError(f'Contact info not found for suite '
f'"{suite}", suite not running?')
if not host:
host = contact[ContactFileFields.HOST]
host = get_fqdn_by_host(host)
port = int(contact[ContactFileFields.PORT])
pub_port = int(contact[ContactFileFields.PUBLISH_PORT])
return host, port, pub_port
def main(parser, options, reg):
"""CLI for "cylc get-suite-contact"."""
try:
data = load_contact_file(reg)
except SuiteServiceFileError:
raise CylcError(f"{reg}: cannot get contact info, suite not running?")
else:
for key, value in sorted(data.items()):
print("%s=%s" % (key, value))
def _check_contact_file(scheduler):
try:
contact_data = suite_files.load_contact_file(
scheduler.suite)
if contact_data != scheduler.contact_data:
raise CylcError('contact file modified')
except (AssertionError, IOError, ValueError, SuiteServiceFileError):
raise CylcError(
'%s: contact file corrupted/modified and may be left'
% suite_files.get_contact_file(scheduler.suite)
)
def send_request(self, command, args=None, timeout=None):
"""Send a request, using ssh.
Determines ssh_cmd, cylc_path and login_shell settings from the contact
file.
Converts message to JSON and sends this to stdin. Executes the Cylc
command, then deserialises the output.
Use ``__call__`` to call this method.
Args:
command (str): The name of the endpoint to call.
args (dict): Arguments to pass to the endpoint function.
timeout (float): Override the default timeout (seconds).
Raises:
ClientError: Coverall, on error from function call
Returns:
object: Deserialized output from function called.
"""
# Set environment variable to determine the communication for use on
# the scheduler
os.environ["CLIENT_COMMS_METH"] = CommsMeth.SSH
cmd = ["client"]
if timeout:
cmd += [f'comms_timeout={timeout}']
cmd += [self.suite, command]
contact = load_contact_file(self.suite)
ssh_cmd = contact[ContactFileFields.SCHEDULER_SSH_COMMAND]
login_shell = contact[ContactFileFields.SCHEDULER_USE_LOGIN_SHELL]
cylc_path = contact[ContactFileFields.SCHEDULER_CYLC_PATH]
cylc_path = None if cylc_path == 'None' else cylc_path
if not args:
args = {}
message = json.dumps(args)
proc = _remote_cylc_cmd(cmd,
host=self.host,
stdin_str=message,
ssh_cmd=ssh_cmd,
remote_cylc_path=cylc_path,
ssh_login_shell=login_shell,
capture_process=True)
out, err = (f.decode() for f in proc.communicate())
return_code = proc.wait()
if return_code:
raise ClientError(err, f"return-code={return_code}")
return json.loads(out)
async def test_scan_cleans_stuck_contact_files(run, scheduler, flow, one_conf,
run_dir, test_dir):
"""Ensure scan tidies up contact files from crashed flows."""
# create a flow
reg = flow(one_conf, name='-crashed-')
schd = scheduler(reg)
srv_dir = Path(run_dir, reg, SuiteFiles.Service.DIRNAME)
tmp_dir = test_dir / 'srv'
cont = srv_dir / SuiteFiles.Service.CONTACT
# run the flow, copy the contact, stop the flow, copy back the contact
async with run(schd):
copytree(srv_dir, tmp_dir)
rmtree(srv_dir)
copytree(tmp_dir, srv_dir)
rmtree(tmp_dir)
# the old contact file check uses the CLI command that the flow was run
# with to check that whether the flow is running. Because this is an
# integration test the process is the pytest process and it is still
# running so we need to change the command so that Cylc sees the flow as
# having crashed
contact_info = load_contact_file(reg)
contact_info[ContactFileFields.PROCESS] += 'xyz'
dump_contact_file(reg, contact_info)
# make sure this flow shows for a regular filesystem-only scan
opts = ScanOptions(states='running,paused', format='name')
flows = []
await main(opts, write=flows.append, scan_dir=test_dir)
assert len(flows) == 1
assert '-crashed-' in flows[0]
# the contact file should still be there
assert cont.exists()
# make sure this flow shows for a regular filesystem-only scan
opts = ScanOptions(states='running,paused', format='name', ping=True)
flows = []
await main(opts, write=flows.append, scan_dir=test_dir)
assert len(flows) == 0
# the contact file should have been removed by the scan
assert not cont.exists()
def get_location(suite: str):
"""Extract host and port from a suite's contact file.
NB: if it fails to load the suite contact file, it will exit.
Args:
suite (str): suite name
Returns:
Tuple[str, int, int]: tuple with the host name and port numbers.
Raises:
ClientError: if the suite is not running.
"""
try:
contact = load_contact_file(suite)
except SuiteServiceFileError:
raise SuiteStopped(suite)
host = contact[ContactFileFields.HOST]
host = get_fqdn_by_host(host)
port = int(contact[ContactFileFields.PORT])
pub_port = int(contact[ContactFileFields.PUBLISH_PORT])
return host, port, pub_port
def get_scan_items_from_fs(
owner_pattern=None, reg_pattern=None, active_only=True):
"""Scrape list of suites from the filesystem.
Walk users' "~/cylc-run/" to get (host, port) from ".service/contact" for
active, or all (active plus registered but dormant), suites.
Yields:
tuple - (reg, host, port, pub_port, api)
"""
if owner_pattern is None:
# Run directory of current user only
run_dirs = [(glbl_cfg().get_host_item('run directory'), None)]
else:
# Run directory of all users matching "owner_pattern".
# But skip those with /nologin or /false shells
run_dirs = []
skips = ('/false', '/nologin')
for pwent in getpwall():
if any(pwent.pw_shell.endswith(s) for s in skips):
continue
if owner_pattern.match(pwent.pw_name):
run_dirs.append((
glbl_cfg().get_host_item(
'run directory',
owner=pwent.pw_name,
owner_home=pwent.pw_dir),
pwent.pw_name))
if cylc.flow.flags.debug:
sys.stderr.write('Listing suites:%s%s\n' % (
DEBUG_DELIM, DEBUG_DELIM.join(item[1] for item in run_dirs if
item[1] is not None)))
for run_d, owner in run_dirs:
for dirpath, dnames, _ in os.walk(run_d, followlinks=True):
# Always descend for top directory, but
# don't descend further if it has a .service/ or log/ dir
if dirpath != run_d and (
SuiteFiles.Service.DIRNAME
in dnames or 'log' in dnames):
dnames[:] = []
# Filter suites by name
reg = os.path.relpath(dirpath, run_d)
if reg_pattern and not reg_pattern.match(reg):
continue
# Choose only suites with .service and matching filter
if active_only:
try:
contact_data = load_contact_file(
reg, owner)
except (SuiteServiceFileError, IOError, TypeError, ValueError):
continue
yield (
reg,
contact_data[ContactFileFields.HOST],
contact_data[ContactFileFields.PORT],
contact_data[ContactFileFields.PUBLISH_PORT],
contact_data[ContactFileFields.API]
)
else:
try:
source_dir = get_suite_source_dir(reg)
title = get_suite_title(reg)
except (SuiteServiceFileError, IOError):
continue
yield (
reg,
source_dir,
title
)
def test_load_contact_file(myflow):
cont = load_contact_file(myflow.suite)
assert cont[CFF.HOST] == myflow.host