def get_task_auth(self, suite_name: str,
task_name: str) -> Union[str, None]:
"""Get host for a remote task from a Cylc workflow definition.
Returns: Hostname or None if:
- task does not run remotely.
- task has not been defined.
"""
# n.b. Imports inside function to avoid dependency on Cylc and
# Cylc-Rose is Rose is being used with a different workflow engine.
from cylc.flow.platforms import get_host_from_platform
from cylc.flow.hostuserutil import is_remote_platform
from cylc.rose.platform_utils import get_platform_from_task_def
# Check whether task has been defined.
try:
platform = get_platform_from_task_def(suite_name, task_name)
except KeyError:
return None
else:
# If task has been defined return host:
if is_remote_platform(platform):
return get_host_from_platform(platform)
else:
return None
def remote_tidy(self):
"""Remove suite contact files and keys 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.
"""
# Issue all SSH commands in parallel
procs = {}
for platform, init_with_contact in self.remote_init_map.items():
platform = get_platform(platform)
host = get_host_from_platform(platform)
owner = platform['owner']
self.install_target = get_install_target_from_platform(platform)
if init_with_contact != REMOTE_INIT_DONE:
continue
cmd = ['remote-tidy']
if cylc.flow.flags.debug:
cmd.append('--debug')
cmd.append(str(f'{self.install_target}'))
cmd.append(get_remote_suite_run_dir(platform, self.suite))
if is_remote_platform(platform):
cmd = construct_platform_ssh_cmd(cmd, platform, timeout='10s')
else:
cmd = ['cylc'] + cmd
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 get_task_auth(self, suite_name: str,
task_name: str) -> Union[str, None]:
"""Get host for a remote task from a Cylc workflow definition.
Returns: Hostname, or None if:
- task does not run remotely.
- task has not been defined.
- cylc-rose is not installed(*)
(*) This function is only used by the fcm_make built-in app. Returning
None is equivalent to there being no fcm_make2 task found or no
workflow file found which is fine - 2 stage fcm_make is only supported
on the localhost install target (the workflow files aren't mirrored).
"""
# n.b. Imports inside function to avoid dependency on Cylc and
# Cylc-Rose is Rose is being used with a different workflow engine.
from cylc.flow.exceptions import WorkflowFilesError
from cylc.flow.hostuserutil import is_remote_platform
from cylc.flow.platforms import get_host_from_platform
try:
from cylc.rose.platform_utils import get_platform_from_task_def
except ModuleNotFoundError:
# Allow single stage fcm_make app to work without requiring
# cylc.rose
return None
try:
platform = get_platform_from_task_def(suite_name, task_name)
except KeyError:
return None
except (WorkflowFilesError):
raise WorkflowFileNotFoundError
else:
if platform is None:
return 'localhost'
# If task has been defined return host:
if is_remote_platform(platform):
return get_host_from_platform(platform)
else:
return None
def _run_job_cmd(self, cmd_key, suite, itasks, callback):
"""Run job commands, e.g. poll, kill, etc.
Group itasks with their platform_name and host.
Put a job command for each group to the multiprocess pool.
"""
if not itasks:
return
# sort itasks into lists based upon where they were run.
auth_itasks = {}
for itask in itasks:
platform_n = itask.platform['name']
if platform_n not in auth_itasks:
auth_itasks[platform_n] = []
auth_itasks[platform_n].append(itask)
# Go through each list of itasks and carry out commands as required.
for platform_n, itasks in sorted(auth_itasks.items()):
platform = get_platform(platform_n)
if is_remote_platform(platform):
remote_mode = True
cmd = [cmd_key]
else:
cmd = ["cylc", cmd_key]
remote_mode = False
if LOG.isEnabledFor(DEBUG):
cmd.append("--debug")
cmd.append("--")
cmd.append(get_remote_suite_run_job_dir(platform, suite))
job_log_dirs = []
if remote_mode:
cmd = construct_ssh_cmd(cmd, platform)
for itask in sorted(itasks, key=lambda itask: itask.identity):
job_log_dirs.append(
get_task_job_id(itask.point, itask.tdef.name,
itask.submit_num))
cmd += job_log_dirs
self.proc_pool.put_command(SubProcContext(cmd_key, cmd), callback,
[suite, itasks])
def main(parser: COP,
options: 'Values',
reg: str,
task_id: Optional[str] = None,
color: bool = False) -> None:
"""Implement cylc cat-log CLI.
Determine log path, user@host, batchview_cmd, and action (print, dir-list,
cat, edit, or tail), and then if the log path is:
a) local: perform action on log path, or
b) remote: re-invoke cylc cat-log as a) on the remote account
"""
if options.remote_args:
# Invoked on job hosts for job logs only, as a wrapper to view_log().
# Tail and batchview commands from global config on workflow host).
logpath, mode, tail_tmpl = options.remote_args[0:3]
logpath = expand_path(logpath)
tail_tmpl = expand_path(tail_tmpl)
try:
batchview_cmd = options.remote_args[3]
except IndexError:
batchview_cmd = None
res = view_log(logpath,
mode,
tail_tmpl,
batchview_cmd,
remote=True,
color=color)
if res == 1:
sys.exit(res)
return
workflow_name, _ = parse_reg(reg)
# Get long-format mode.
try:
mode = MODES[options.mode]
except KeyError:
mode = options.mode
if not task_id:
# Cat workflow logs, local only.
if options.filename is not None:
raise UserInputError("The '-f' option is for job logs only.")
logpath = get_workflow_run_log_name(workflow_name)
if options.rotation_num:
logs = glob('%s.*' % logpath)
logs.sort(key=os.path.getmtime, reverse=True)
try:
logpath = logs[int(options.rotation_num)]
except IndexError:
raise UserInputError("max rotation %d" % (len(logs) - 1))
tail_tmpl = os.path.expandvars(get_platform()["tail command template"])
out = view_log(logpath, mode, tail_tmpl, color=color)
if out == 1:
sys.exit(1)
if mode == 'edit':
tmpfile_edit(out, options.geditor)
return
if task_id:
# Cat task job logs, may be on workflow or job host.
if options.rotation_num is not None:
raise UserInputError("only workflow (not job) logs get rotated")
try:
task, point = TaskID.split(task_id)
except ValueError:
parser.error("Illegal task ID: %s" % task_id)
if options.submit_num != NN:
try:
options.submit_num = "%02d" % int(options.submit_num)
except ValueError:
parser.error("Illegal submit number: %s" % options.submit_num)
if options.filename is None:
options.filename = JOB_LOG_OUT
else:
# Convert short filename args to long (e.g. 'o' to 'job.out').
with suppress(KeyError):
options.filename = JOB_LOG_OPTS[options.filename]
# KeyError: Is already long form (standard log, or custom).
platform_name, job_runner_name, live_job_id = get_task_job_attrs(
workflow_name, point, task, options.submit_num)
platform = get_platform(platform_name)
batchview_cmd = None
if live_job_id is not None:
# Job is currently running. Get special job runner log view
# command (e.g. qcat) if one exists, and the log is out or err.
conf_key = None
if options.filename == JOB_LOG_OUT:
if mode == 'cat':
conf_key = "out viewer"
elif mode == 'tail':
conf_key = "out tailer"
elif options.filename == JOB_LOG_ERR:
if mode == 'cat':
conf_key = "err viewer"
elif mode == 'tail':
conf_key = "err tailer"
if conf_key is not None:
batchview_cmd_tmpl = None
with suppress(KeyError):
batchview_cmd_tmpl = platform[conf_key]
if batchview_cmd_tmpl is not None:
batchview_cmd = batchview_cmd_tmpl % {
"job_id": str(live_job_id)
}
log_is_remote = (is_remote_platform(platform)
and (options.filename != JOB_LOG_ACTIVITY))
log_is_retrieved = (platform['retrieve job logs']
and live_job_id is None)
if log_is_remote and (not log_is_retrieved or options.force_remote):
logpath = os.path.normpath(
get_remote_workflow_run_job_dir(workflow_name, point, task,
options.submit_num,
options.filename))
tail_tmpl = platform["tail command template"]
# Reinvoke the cat-log command on the remote account.
cmd = ['cat-log', *verbosity_to_opts(cylc.flow.flags.verbosity)]
for item in [logpath, mode, tail_tmpl]:
cmd.append('--remote-arg=%s' % shlex.quote(item))
if batchview_cmd:
cmd.append('--remote-arg=%s' % shlex.quote(batchview_cmd))
cmd.append(workflow_name)
is_edit_mode = (mode == 'edit')
# TODO: Add Intelligent Host selection to this
try:
proc = remote_cylc_cmd(cmd,
platform,
capture_process=is_edit_mode,
manage=(mode == 'tail'))
except KeyboardInterrupt:
# Ctrl-C while tailing.
pass
else:
if is_edit_mode:
# Write remote stdout to a temp file for viewing in editor.
# Only BUFSIZE bytes at a time in case huge stdout volume.
out = NamedTemporaryFile()
data = proc.stdout.read(BUFSIZE)
while data:
out.write(data)
data = proc.stdout.read(BUFSIZE)
os.chmod(out.name, S_IRUSR)
out.seek(0, 0)
else:
# Local task job or local job log.
logpath = os.path.normpath(
get_workflow_run_job_dir(workflow_name, point, task,
options.submit_num, options.filename))
tail_tmpl = os.path.expandvars(platform["tail command template"])
out = view_log(logpath,
mode,
tail_tmpl,
batchview_cmd,
color=color)
if mode != 'edit':
sys.exit(out)
if mode == 'edit':
tmpfile_edit(out, options.geditor)
def submit_task_jobs(self,
suite,
itasks,
curve_auth,
client_pub_key_dir,
is_simulation=False):
"""Prepare and submit task jobs.
Submit tasks where possible. Ignore tasks that are waiting for host
select command to complete, or tasks that are waiting for remote
initialisation. Bad host select command, error writing to a job file or
bad remote initialisation will cause a bad task - leading to submission
failure.
This method uses prep_submit_task_job() as helper.
Return (list): list of tasks that attempted submission.
"""
if is_simulation:
return self._simulation_submit_task_jobs(itasks)
# Prepare tasks for job submission
prepared_tasks, bad_tasks = self.prep_submit_task_jobs(suite, itasks)
# Reset consumed host selection results
self.task_remote_mgr.subshell_eval_reset()
if not prepared_tasks:
return bad_tasks
# Group task jobs by (install target)
auth_itasks = {} # {install target: [itask, ...], ...}
for itask in prepared_tasks:
install_target = get_install_target_from_platform(itask.platform)
auth_itasks.setdefault(install_target, [])
auth_itasks[install_target].append(itask)
# Submit task jobs for each platform
done_tasks = bad_tasks
for install_target, itasks in sorted(auth_itasks.items()):
# Re-fetch a copy of platform
platform = itasks[0].platform
is_init = self.task_remote_mgr.remote_init(platform, curve_auth,
client_pub_key_dir)
if is_init is None:
# Remote is waiting to be initialised
for itask in itasks:
itask.set_summary_message(self.REMOTE_INIT_MSG)
self.job_pool.add_job_msg(
get_task_job_id(itask.point, itask.tdef.name,
itask.submit_num),
self.REMOTE_INIT_MSG)
continue
# Ensure that localhost background/at jobs are recorded as running
# on the host name of the current suite host, rather than just
# "localhost". On suite restart on a different suite host, this
# allows the restart logic to correctly poll the status of the
# background/at jobs that may still be running on the previous
# suite host.
host = get_host_from_platform(platform)
if (self.batch_sys_mgr.is_job_local_to_host(
itask.summary['batch_sys_name'])
and not is_remote_platform(platform)):
host = get_host()
now_str = get_current_time_string()
done_tasks.extend(itasks)
for itask in itasks:
# Log and persist
LOG.info('[%s] -submit-num=%02d, owner@host=%s', itask,
itask.submit_num, host)
self.suite_db_mgr.put_insert_task_jobs(
itask, {
'is_manual_submit': itask.is_manual_submit,
'try_num': itask.get_try_num(),
'time_submit': now_str,
'platform_name': platform['name'],
'batch_sys_name': itask.summary['batch_sys_name'],
})
itask.is_manual_submit = False
if is_init == REMOTE_INIT_FAILED:
# Remote has failed to initialise
# Set submit-failed for all affected tasks
for itask in itasks:
itask.local_job_file_path = None # reset for retry
log_task_job_activity(
SubProcContext(self.JOBS_SUBMIT,
'(init %s)' % host,
err=REMOTE_INIT_FAILED,
ret_code=1), suite, itask.point,
itask.tdef.name)
self.task_events_mgr.process_message(
itask, CRITICAL,
self.task_events_mgr.EVENT_SUBMIT_FAILED)
continue
# Build the "cylc jobs-submit" command
cmd = [self.JOBS_SUBMIT]
if LOG.isEnabledFor(DEBUG):
cmd.append('--debug')
if get_utc_mode():
cmd.append('--utc-mode')
if is_remote_platform(itask.platform):
remote_mode = True
cmd.append('--remote-mode')
else:
remote_mode = False
cmd.append('--')
cmd.append(get_remote_suite_run_job_dir(platform, suite))
# Chop itasks into a series of shorter lists if it's very big
# to prevent overloading of stdout and stderr pipes.
itasks = sorted(itasks, key=lambda itask: itask.identity)
chunk_size = len(itasks) // ((len(itasks) // 100) + 1) + 1
itasks_batches = [
itasks[i:i + chunk_size]
for i in range(0, len(itasks), chunk_size)
]
LOG.debug('%s ... # will invoke in batches, sizes=%s', cmd,
[len(b) for b in itasks_batches])
if remote_mode:
cmd = construct_platform_ssh_cmd(cmd, platform)
else:
cmd = ['cylc'] + cmd
for i, itasks_batch in enumerate(itasks_batches):
stdin_files = []
job_log_dirs = []
for itask in itasks_batch:
if remote_mode:
stdin_files.append(
os.path.expandvars(
get_task_job_job_log(suite, itask.point,
itask.tdef.name,
itask.submit_num)))
job_log_dirs.append(
get_task_job_id(itask.point, itask.tdef.name,
itask.submit_num))
# The job file is now (about to be) used: reset the file
# write flag so that subsequent manual retrigger will
# generate a new job file.
itask.local_job_file_path = None
itask.state.reset(TASK_STATUS_READY)
if itask.state.outputs.has_custom_triggers():
self.suite_db_mgr.put_update_task_outputs(itask)
self.proc_pool.put_command(
SubProcContext(
self.JOBS_SUBMIT,
cmd + job_log_dirs,
stdin_files=stdin_files,
job_log_dirs=job_log_dirs,
), self._submit_task_jobs_callback, [suite, itasks_batch])
return done_tasks
def submit_task_jobs(self,
suite,
itasks,
curve_auth,
client_pub_key_dir,
is_simulation=False):
"""Prepare for job submission and submit task jobs.
Preparation (host selection, remote host init, and remote install)
is done asynchronously. Newly released tasks may be sent here several
times until these init subprocesses have returned. Failure during
preparation is considered to be job submission failure.
Once preparation has completed or failed, reset .waiting_on_job_prep in
task instances so the scheduler knows to stop sending them back here.
This method uses prep_submit_task_job() as helper.
Return (list): list of tasks that attempted submission.
"""
if is_simulation:
return self._simulation_submit_task_jobs(itasks)
# Prepare tasks for job submission
prepared_tasks, bad_tasks = self.prep_submit_task_jobs(suite, itasks)
# Reset consumed host selection results
self.task_remote_mgr.subshell_eval_reset()
if not prepared_tasks:
return bad_tasks
auth_itasks = {} # {platform: [itask, ...], ...}
for itask in prepared_tasks:
platform_name = itask.platform['name']
auth_itasks.setdefault(platform_name, [])
auth_itasks[platform_name].append(itask)
# Submit task jobs for each platform
done_tasks = bad_tasks
for platform_name, itasks in sorted(auth_itasks.items()):
platform = itasks[0].platform
install_target = get_install_target_from_platform(platform)
ri_map = self.task_remote_mgr.remote_init_map
if (ri_map.get(install_target) != REMOTE_FILE_INSTALL_DONE):
if install_target == get_localhost_install_target():
# Skip init and file install for localhost.
LOG.debug(f"REMOTE INIT NOT REQUIRED for {install_target}")
ri_map[install_target] = (REMOTE_FILE_INSTALL_DONE)
elif install_target not in ri_map:
# Remote init not in progress for target, so start it.
self.task_remote_mgr.remote_init(platform, curve_auth,
client_pub_key_dir)
for itask in itasks:
itask.set_summary_message(self.REMOTE_INIT_MSG)
self.data_store_mgr.delta_job_msg(
get_task_job_id(itask.point, itask.tdef.name,
itask.submit_num),
self.REMOTE_INIT_MSG)
continue
elif (ri_map[install_target] == REMOTE_INIT_DONE):
# Already done remote init so move on to file install
self.task_remote_mgr.file_install(platform)
continue
elif (ri_map[install_target] in self.IN_PROGRESS.keys()):
# Remote init or file install in progress.
for itask in itasks:
msg = self.IN_PROGRESS[ri_map[install_target]]
itask.set_summary_message(msg)
self.data_store_mgr.delta_job_msg(
get_task_job_id(itask.point, itask.tdef.name,
itask.submit_num), msg)
continue
# Ensure that localhost background/at jobs are recorded as running
# on the host name of the current suite host, rather than just
# "localhost". On suite restart on a different suite host, this
# allows the restart logic to correctly poll the status of the
# background/at jobs that may still be running on the previous
# suite host.
host = get_host_from_platform(platform)
if (self.job_runner_mgr.is_job_local_to_host(
itask.summary['job_runner_name'])
and not is_remote_platform(platform)):
host = get_host()
now_str = get_current_time_string()
done_tasks.extend(itasks)
for itask in itasks:
# Log and persist
LOG.info('[%s] -submit-num=%02d, host=%s', itask,
itask.submit_num, host)
self.suite_db_mgr.put_insert_task_jobs(
itask, {
'is_manual_submit': itask.is_manual_submit,
'try_num': itask.get_try_num(),
'time_submit': now_str,
'platform_name': itask.platform['name'],
'job_runner_name': itask.summary['job_runner_name'],
})
itask.is_manual_submit = False
if (ri_map[install_target]
in [REMOTE_INIT_FAILED, REMOTE_FILE_INSTALL_FAILED]):
# Remote init or install failed. Set submit-failed for all
# affected tasks and remove target from remote init map
# - this enables new tasks to re-initialise that target
init_error = (ri_map[install_target])
del ri_map[install_target]
for itask in itasks:
itask.waiting_on_job_prep = False
itask.local_job_file_path = None # reset for retry
log_task_job_activity(
SubProcContext(self.JOBS_SUBMIT,
'(init %s)' % host,
err=init_error,
ret_code=1), suite, itask.point,
itask.tdef.name)
self._prep_submit_task_job_error(suite, itask,
'(remote init)', '')
continue
# Build the "cylc jobs-submit" command
cmd = [self.JOBS_SUBMIT]
if LOG.isEnabledFor(DEBUG):
cmd.append('--debug')
if get_utc_mode():
cmd.append('--utc-mode')
if is_remote_platform(itask.platform):
remote_mode = True
cmd.append('--remote-mode')
else:
remote_mode = False
if itask.platform['clean job submission environment']:
cmd.append('--clean-env')
for var in itask.platform[
'job submission environment pass-through']:
cmd.append(f"--env={var}")
for path in itask.platform[
'job submission executable paths'] + SYSPATH:
cmd.append(f"--path={path}")
cmd.append('--')
cmd.append(get_remote_suite_run_job_dir(platform, suite))
# Chop itasks into a series of shorter lists if it's very big
# to prevent overloading of stdout and stderr pipes.
itasks = sorted(itasks, key=lambda itask: itask.identity)
chunk_size = (len(itasks) // (
(len(itasks) // platform['max batch submit size']) + 1) + 1)
itasks_batches = [
itasks[i:i + chunk_size]
for i in range(0, len(itasks), chunk_size)
]
LOG.debug('%s ... # will invoke in batches, sizes=%s', cmd,
[len(b) for b in itasks_batches])
if remote_mode:
cmd = construct_ssh_cmd(cmd, platform)
else:
cmd = ['cylc'] + cmd
for i, itasks_batch in enumerate(itasks_batches):
stdin_files = []
job_log_dirs = []
for itask in itasks_batch:
if remote_mode:
stdin_files.append(
os.path.expandvars(
get_task_job_job_log(suite, itask.point,
itask.tdef.name,
itask.submit_num)))
job_log_dirs.append(
get_task_job_id(itask.point, itask.tdef.name,
itask.submit_num))
# The job file is now (about to be) used: reset the file
# write flag so that subsequent manual retrigger will
# generate a new job file.
itask.local_job_file_path = None
if itask.state.outputs.has_custom_triggers():
self.suite_db_mgr.put_update_task_outputs(itask)
itask.waiting_on_job_prep = False
self.proc_pool.put_command(
SubProcContext(
self.JOBS_SUBMIT,
cmd + job_log_dirs,
stdin_files=stdin_files,
job_log_dirs=job_log_dirs,
), self._submit_task_jobs_callback, [suite, itasks_batch])
return done_tasks
