def _process_message_submit_failed(self, itask, event_time):
"""Helper for process_message, handle a submit-failed message."""
LOG.error('[%s] -%s', itask, self.EVENT_SUBMIT_FAILED)
if event_time is None:
event_time = get_current_time_string()
self.suite_db_mgr.put_update_task_jobs(
itask, {
"time_submit_exit": get_current_time_string(),
"submit_status": 1,
})
itask.summary['submit_method_id'] = None
if (TASK_STATUS_SUBMIT_RETRYING not in itask.try_timers or
itask.try_timers[TASK_STATUS_SUBMIT_RETRYING].next() is None):
# No submission retry lined up: definitive failure.
self.pflag = True
# See github #476.
if itask.state.reset_state(TASK_STATUS_SUBMIT_FAILED):
self.setup_event_handlers(itask, self.EVENT_SUBMIT_FAILED,
'job %s' % self.EVENT_SUBMIT_FAILED)
else:
# There is a submission retry lined up.
timer = itask.try_timers[TASK_STATUS_SUBMIT_RETRYING]
delay_msg = "submit-retrying in %s" % timer.delay_timeout_as_str()
msg = "%s, %s" % (self.EVENT_SUBMIT_FAILED, delay_msg)
LOG.info("[%s] -job(%02d) %s", itask, itask.submit_num, msg)
itask.set_summary_message(msg)
if itask.state.reset_state(TASK_STATUS_SUBMIT_RETRYING):
self.setup_event_handlers(
itask, self.EVENT_SUBMIT_RETRY,
"job %s, %s" % (self.EVENT_SUBMIT_FAILED, delay_msg))
self._reset_job_timers(itask)
def load_config(self):
"""Load the suite config."""
if self.suiterc:
is_reload = True
collapsed = self.suiterc.closed_families
else:
is_reload = False
collapsed = []
try:
self.suiterc = SuiteConfig(
self.suite,
self.file,
self.template_vars,
is_reload=is_reload,
collapsed=collapsed,
cli_initial_point_string=self.start_point_string,
vis_start_string=self.start_point_string,
vis_stop_string=self.stop_point_string)
except Exception as exc:
msg = "Failed - parsing error?\n\n%s" % exc
LOG.error(msg)
if self.interactive:
dia = gtk.MessageDialog(type=gtk.MESSAGE_ERROR,
buttons=gtk.BUTTONS_OK,
message_format=msg)
dia.run()
dia.destroy()
return False
sys.exit(1)
self.inherit = self.suiterc.get_parent_lists()
return True
def list_suites(self, regfilter=None):
"""Return a filtered list of valid suite registrations."""
rec_regfilter = None
if regfilter:
try:
rec_regfilter = re.compile(regfilter)
except re.error as exc:
raise ValueError("%s: %s" % (regfilter, exc))
run_d = glbl_cfg().get_host_item('run directory')
results = []
for dirpath, dnames, fnames in os.walk(run_d, followlinks=True):
# Always descend for top directory, but
# don't descend further if it has a:
# * .service/
# * cylc-suite.db: (pre-cylc-7 suites don't have ".service/").
if dirpath != run_d and (self.DIR_BASE_SRV in dnames
or "cylc-suite.db" in fnames):
dnames[:] = []
# Choose only suites with .service and matching filter
reg = os.path.relpath(dirpath, run_d)
path = os.path.join(dirpath, self.DIR_BASE_SRV)
if (not self._locate_item(self.FILE_BASE_SOURCE, path)
or rec_regfilter and not rec_regfilter.search(reg)):
continue
try:
results.append([
reg,
self.get_suite_source_dir(reg),
self.get_suite_title(reg)
])
except (IOError, SuiteServiceFileError) as exc:
LOG.error('%s: %s', reg, exc)
return results
def _process_message_submit_failed(self, itask, event_time):
"""Helper for process_message, handle a submit-failed message."""
LOG.error('[%s] -%s', itask, self.EVENT_SUBMIT_FAILED)
if event_time is None:
event_time = get_current_time_string()
self.suite_db_mgr.put_update_task_jobs(itask, {
"time_submit_exit": event_time,
"submit_status": 1,
})
itask.summary['submit_method_id'] = None
self.pflag = True
if (TASK_STATUS_SUBMIT_RETRYING not in itask.try_timers or
itask.try_timers[TASK_STATUS_SUBMIT_RETRYING].next() is None):
# No submission retry lined up: definitive failure.
# See github #476.
if itask.state.reset_state(TASK_STATUS_SUBMIT_FAILED):
self.setup_event_handlers(
itask, self.EVENT_SUBMIT_FAILED,
'job %s' % self.EVENT_SUBMIT_FAILED)
else:
# There is a submission retry lined up.
timer = itask.try_timers[TASK_STATUS_SUBMIT_RETRYING]
delay_msg = "submit-retrying in %s" % timer.delay_timeout_as_str()
msg = "%s, %s" % (self.EVENT_SUBMIT_FAILED, delay_msg)
LOG.info("[%s] -job(%02d) %s", itask, itask.submit_num, msg)
itask.set_summary_message(msg)
if itask.state.reset_state(TASK_STATUS_SUBMIT_RETRYING):
self.setup_event_handlers(
itask, self.EVENT_SUBMIT_RETRY,
"job %s, %s" % (self.EVENT_SUBMIT_FAILED, delay_msg))
self._reset_job_timers(itask)
def list_suites(self, regfilter=None):
"""Return a filtered list of valid suite registrations."""
rec_regfilter = None
if regfilter:
try:
rec_regfilter = re.compile(regfilter)
except re.error as exc:
raise ValueError("%s: %s" % (regfilter, exc))
run_d = glbl_cfg().get_host_item('run directory')
results = []
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/ dir
if dirpath != run_d and self.DIR_BASE_SRV in dnames:
dnames[:] = []
# Choose only suites with .service and matching filter
reg = os.path.relpath(dirpath, run_d)
path = os.path.join(dirpath, self.DIR_BASE_SRV)
if (not self._locate_item(self.FILE_BASE_SOURCE, path) or
rec_regfilter and not rec_regfilter.search(reg)):
continue
try:
results.append([
reg,
self.get_suite_source_dir(reg),
self.get_suite_title(reg)])
except (IOError, SuiteServiceFileError) as exc:
LOG.error('%s: %s', reg, exc)
return results
def _prep_submit_task_job_error(self, suite, itask, dry_run, action, exc):
"""Helper for self._prep_submit_task_job. On error."""
LOG.debug("submit_num %s" % itask.submit_num)
LOG.debug(traceback.format_exc())
LOG.error(exc)
log_task_job_activity(SubProcContext(self.JOBS_SUBMIT,
action,
err=exc,
ret_code=1),
suite,
itask.point,
itask.tdef.name,
submit_num=itask.submit_num)
if not dry_run:
# Persist
self.suite_db_mgr.put_insert_task_jobs(
itask, {
'is_manual_submit': itask.is_manual_submit,
'try_num': itask.get_try_num(),
'time_submit': get_current_time_string(),
'batch_sys_name': itask.summary.get('batch_sys_name'),
})
itask.is_manual_submit = False
self.task_events_mgr.process_message(
itask, CRITICAL, self.task_events_mgr.EVENT_SUBMIT_FAILED)
def clear_broadcast(self,
point_strings=None,
namespaces=None,
cancel_settings=None):
"""Clear broadcasts globally, or for listed namespaces and/or points.
Return a tuple (modified_settings, bad_options), where:
* modified_settings is similar to the return value of the "put" method,
but for removed broadcasts.
* bad_options is a dict in the form:
{"point_strings": ["20020202", ..."], ...}
The dict is only populated if there are options not associated with
previous broadcasts. The keys can be:
* point_strings: a list of bad point strings.
* namespaces: a list of bad namespaces.
* cancel: a list of tuples. Each tuple contains the keys of a bad
setting.
"""
# If cancel_settings defined, only clear specific broadcasts
cancel_keys_list = self._settings_to_keys_list(cancel_settings)
# Clear broadcasts
modified_settings = []
with self.lock:
for point_string, point_string_settings in self.broadcasts.items():
if point_strings and point_string not in point_strings:
continue
for namespace, namespace_settings in (
point_string_settings.items()):
if namespaces and namespace not in namespaces:
continue
stuff_stack = [([], namespace_settings)]
while stuff_stack:
keys, stuff = stuff_stack.pop()
for key, value in stuff.items():
if isinstance(value, dict):
stuff_stack.append((keys + [key], value))
elif (not cancel_keys_list
or keys + [key] in cancel_keys_list):
stuff[key] = None
setting = {key: value}
for rkey in reversed(keys):
setting = {rkey: setting}
modified_settings.append(
(point_string, namespace, setting))
# Prune any empty branches
bad_options = self._get_bad_options(self._prune(), point_strings,
namespaces, cancel_keys_list)
# Log the broadcast
self.suite_db_mgr.put_broadcast(modified_settings, is_cancel=True)
LOG.info(get_broadcast_change_report(modified_settings,
is_cancel=True))
if bad_options:
LOG.error(get_broadcast_bad_options_report(bad_options))
return (modified_settings, bad_options)
def clear_broadcast(
self, point_strings=None, namespaces=None, cancel_settings=None):
"""Clear broadcasts globally, or for listed namespaces and/or points.
Return a tuple (modified_settings, bad_options), where:
* modified_settings is similar to the return value of the "put" method,
but for removed broadcasts.
* bad_options is a dict in the form:
{"point_strings": ["20020202", ..."], ...}
The dict is only populated if there are options not associated with
previous broadcasts. The keys can be:
* point_strings: a list of bad point strings.
* namespaces: a list of bad namespaces.
* cancel: a list of tuples. Each tuple contains the keys of a bad
setting.
"""
# If cancel_settings defined, only clear specific broadcasts
cancel_keys_list = self._settings_to_keys_list(cancel_settings)
# Clear broadcasts
modified_settings = []
with self.lock:
for point_string, point_string_settings in self.broadcasts.items():
if point_strings and point_string not in point_strings:
continue
for namespace, namespace_settings in (
point_string_settings.items()):
if namespaces and namespace not in namespaces:
continue
stuff_stack = [([], namespace_settings)]
while stuff_stack:
keys, stuff = stuff_stack.pop()
for key, value in stuff.items():
if isinstance(value, dict):
stuff_stack.append((keys + [key], value))
elif (not cancel_keys_list or
keys + [key] in cancel_keys_list):
stuff[key] = None
setting = {key: value}
for rkey in reversed(keys):
setting = {rkey: setting}
modified_settings.append(
(point_string, namespace, setting))
# Prune any empty branches
bad_options = self._get_bad_options(
self._prune(), point_strings, namespaces, cancel_keys_list)
# Log the broadcast
self.suite_db_mgr.put_broadcast(modified_settings, is_cancel=True)
LOG.info(
get_broadcast_change_report(modified_settings, is_cancel=True))
if bad_options:
LOG.error(get_broadcast_bad_options_report(bad_options))
return (modified_settings, bad_options)
def _run_event_handlers_callback(proc_ctx, abort_on_error=False):
"""Callback on completion of a suite event handler."""
if proc_ctx.ret_code:
msg = '%s EVENT HANDLER FAILED' % proc_ctx.cmd_key[1]
LOG.error(str(proc_ctx))
LOG.error(msg)
if abort_on_error:
raise SuiteEventError(msg)
else:
LOG.info(str(proc_ctx))
def _manip_task_jobs_callback(
self, ctx, suite, itasks, summary_callback, more_callbacks=None):
"""Callback when submit/poll/kill tasks command exits."""
if ctx.ret_code:
LOG.error(ctx)
else:
LOG.debug(ctx)
# A dict for easy reference of (CYCLE, NAME, SUBMIT_NUM) -> TaskProxy
#
# Note for "reload": A TaskProxy instance may be replaced on reload, so
# the "itasks" list may not reference the TaskProxy objects that
# replace the old ones. The .reload_successor attribute provides the
# link(s) for us to get to the latest replacement.
#
# Note for "kill": It is possible for a job to trigger its trap and
# report back to the suite back this logic is called. If so, the task
# will no longer be TASK_STATUS_SUBMITTED or TASK_STATUS_RUNNING, and
# its output line will be ignored here.
tasks = {}
for itask in itasks:
while itask.reload_successor is not None:
itask = itask.reload_successor
if itask.point is not None and itask.submit_num:
submit_num = "%02d" % (itask.submit_num)
tasks[(str(itask.point), itask.tdef.name, submit_num)] = itask
handlers = [(self.batch_sys_mgr.OUT_PREFIX_SUMMARY, summary_callback)]
if more_callbacks:
for prefix, callback in more_callbacks.items():
handlers.append((prefix, callback))
out = ctx.out
if not out:
out = ""
bad_tasks = dict(tasks)
for line in out.splitlines(True):
for prefix, callback in handlers:
if line.startswith(prefix):
line = line[len(prefix):].strip()
try:
path = line.split("|", 2)[1] # timestamp, path, status
point, name, submit_num = path.split(os.sep, 2)
if prefix == self.batch_sys_mgr.OUT_PREFIX_SUMMARY:
del bad_tasks[(point, name, submit_num)]
itask = tasks[(point, name, submit_num)]
callback(suite, itask, ctx, line)
except (LookupError, ValueError, KeyError) as exc:
LOG.warning(
'Unhandled %s output: %s', ctx.cmd_key, line)
LOG.exception(exc)
# Task jobs that are in the original command but did not get a status
# in the output. Handle as failures.
for key, itask in sorted(bad_tasks.items()):
line = (
"|".join([ctx.timestamp, os.sep.join(key), "1"]) + "\n")
summary_callback(suite, itask, ctx, line)
def _manip_task_jobs_callback(
self, ctx, suite, itasks, summary_callback, more_callbacks=None):
"""Callback when submit/poll/kill tasks command exits."""
if ctx.ret_code:
LOG.error(ctx)
else:
LOG.debug(ctx)
# A dict for easy reference of (CYCLE, NAME, SUBMIT_NUM) -> TaskProxy
#
# Note for "reload": A TaskProxy instance may be replaced on reload, so
# the "itasks" list may not reference the TaskProxy objects that
# replace the old ones. The .reload_successor attribute provides the
# link(s) for us to get to the latest replacement.
#
# Note for "kill": It is possible for a job to trigger its trap and
# report back to the suite back this logic is called. If so, the task
# will no longer be TASK_STATUS_SUBMITTED or TASK_STATUS_RUNNING, and
# its output line will be ignored here.
tasks = {}
for itask in itasks:
while itask.reload_successor is not None:
itask = itask.reload_successor
if itask.point is not None and itask.submit_num:
submit_num = "%02d" % (itask.submit_num)
tasks[(str(itask.point), itask.tdef.name, submit_num)] = itask
handlers = [(self.batch_sys_mgr.OUT_PREFIX_SUMMARY, summary_callback)]
if more_callbacks:
for prefix, callback in more_callbacks.items():
handlers.append((prefix, callback))
out = ctx.out
if not out:
out = ""
bad_tasks = dict(tasks)
for line in out.splitlines(True):
for prefix, callback in handlers:
if line.startswith(prefix):
line = line[len(prefix):].strip()
try:
path = line.split("|", 2)[1] # timestamp, path, status
point, name, submit_num = path.split(os.sep, 2)
if prefix == self.batch_sys_mgr.OUT_PREFIX_SUMMARY:
del bad_tasks[(point, name, submit_num)]
itask = tasks[(point, name, submit_num)]
callback(suite, itask, ctx, line)
except (LookupError, ValueError, KeyError) as exc:
LOG.warning(
'Unhandled %s output: %s', ctx.cmd_key, line)
LOG.exception(exc)
# Task jobs that are in the original command but did not get a status
# in the output. Handle as failures.
for key, itask in sorted(bad_tasks.items()):
line = (
"|".join([ctx.timestamp, os.sep.join(key), "1"]) + "\n")
summary_callback(suite, itask, ctx, line)
def _run_event_custom_handlers(self, config, ctx):
"""Helper for "run_event_handlers", custom event handlers."""
# Look for event handlers
# 1. Handlers for specific event
# 2. General handlers
handlers = self.get_events_conf(config, '%s handler' % ctx.event)
if not handlers and (ctx.event in self.get_events_conf(
config, 'handler events', [])):
handlers = self.get_events_conf(config, 'handlers')
if not handlers:
return
for i, handler in enumerate(handlers):
cmd_key = ('%s-%02d' % (self.SUITE_EVENT_HANDLER, i), ctx.event)
# Handler command may be a string for substitution
abort_on_error = self.get_events_conf(
config, 'abort if %s handler fails' % ctx.event)
try:
handler_data = {
'event': quote(ctx.event),
'message': quote(ctx.reason),
'suite': quote(ctx.suite),
'suite_uuid': quote(str(ctx.uuid_str)),
}
if config.cfg['meta']:
for key, value in config.cfg['meta'].items():
if key == "URL":
handler_data["suite_url"] = quote(value)
handler_data[key] = quote(value)
cmd = handler % (handler_data)
except KeyError as exc:
message = "%s bad template: %s" % (cmd_key, exc)
LOG.error(message)
if abort_on_error:
raise SuiteEventError(message)
continue
if cmd == handler:
# Nothing substituted, assume classic interface
cmd = "%s '%s' '%s' '%s'" % (handler, ctx.event, ctx.suite,
ctx.reason)
proc_ctx = SubProcContext(cmd_key,
cmd,
env=dict(os.environ),
shell=True)
if abort_on_error or self.proc_pool.closed:
# Run command in foreground if abort on failure is set or if
# process pool is closed
self.proc_pool.run_command(proc_ctx)
self._run_event_handlers_callback(
proc_ctx, abort_on_error=abort_on_error)
else:
# Run command using process pool otherwise
self.proc_pool.put_command(proc_ctx,
self._run_event_handlers_callback)
def restart_upgrade(self):
"""Vacuum/upgrade runtime DB on restart."""
# Backward compat, upgrade database with state file if necessary
suite_run_d = os.path.dirname(os.path.dirname(self.pub_path))
old_pri_db_path = os.path.join(suite_run_d, 'state',
CylcSuiteDAO.OLD_DB_FILE_BASE_NAME)
old_pri_db_path_611 = os.path.join(
suite_run_d, CylcSuiteDAO.OLD_DB_FILE_BASE_NAME_611[0])
old_state_file_path = os.path.join(suite_run_d, "state", "state")
if (os.path.exists(old_pri_db_path)
and os.path.exists(old_state_file_path)
and not os.path.exists(self.pri_path)):
# Upgrade pre-6.11.X runtime database + state file
copy(old_pri_db_path, self.pri_path)
pri_dao = self.get_pri_dao()
pri_dao.upgrade_with_state_file(old_state_file_path)
target = os.path.join(suite_run_d, "state.tar.gz")
cmd = ["tar", "-C", suite_run_d, "-czf", target, "state"]
if call(cmd, stdin=open(os.devnull)) == 0:
rmtree(os.path.join(suite_run_d, "state"), ignore_errors=True)
else:
try:
os.unlink(os.path.join(suite_run_d, "state.tar.gz"))
except OSError:
pass
LOG.error("cannot tar-gzip + remove old state/ directory")
# Remove old files as well
try:
os.unlink(os.path.join(suite_run_d, "cylc-suite-env"))
except OSError:
pass
elif (os.path.exists(old_pri_db_path_611)
and not os.path.exists(self.pri_path)):
# Upgrade 6.11.X runtime database
os.rename(old_pri_db_path_611, self.pri_path)
pri_dao = self.get_pri_dao()
pri_dao.upgrade_from_611()
# Remove old files as well
for name in [
CylcSuiteDAO.OLD_DB_FILE_BASE_NAME_611[1], "cylc-suite-env"
]:
try:
os.unlink(os.path.join(suite_run_d, name))
except OSError:
pass
else:
pri_dao = self.get_pri_dao()
pri_dao.upgrade_pickle_to_json()
# Vacuum the primary/private database file
pri_dao.vacuum()
pri_dao.close()
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_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 __init__(self, suite):
# Suite only needed for back-compat with old clients (see below):
self.suite = suite
self.engine = None
self.port = None
# Figure out the ports we are allowed to use.
self.ok_ports = glbl_cfg().get(['suite servers', 'run ports'])
random.shuffle(self.ok_ports)
comms_options = glbl_cfg().get(['communication', 'options'])
# HTTP Digest Auth uses MD5 - pretty secure in this use case.
# Extending it with extra algorithms is allowed, but won't be
# supported by most browsers. requests and urllib2 are OK though.
self.hash_algorithm = "MD5"
if "SHA1" in comms_options:
# Note 'SHA' rather than 'SHA1'.
self.hash_algorithm = "SHA"
self.srv_files_mgr = SuiteSrvFilesManager()
self.comms_method = glbl_cfg().get(['communication', 'method'])
self.get_ha1 = cherrypy.lib.auth_digest.get_ha1_dict_plain(
{
'cylc':
self.srv_files_mgr.get_auth_item(
self.srv_files_mgr.FILE_BASE_PASSPHRASE,
suite,
content=True),
'anon':
NO_PASSPHRASE
},
algorithm=self.hash_algorithm)
if self.comms_method == 'http':
self.cert = None
self.pkey = None
else: # if self.comms_method in [None, 'https']:
try:
self.cert = self.srv_files_mgr.get_auth_item(
self.srv_files_mgr.FILE_BASE_SSL_CERT, suite)
self.pkey = self.srv_files_mgr.get_auth_item(
self.srv_files_mgr.FILE_BASE_SSL_PEM, suite)
except SuiteServiceFileError:
LOG.error("no HTTPS/OpenSSL support. Aborting...")
raise CylcError("No HTTPS support. "
"Configure user's global.rc to use HTTP.")
self.start()
def _manip_task_jobs_callback(
self, ctx, suite, itasks, summary_callback, more_callbacks=None):
"""Callback when submit/poll/kill tasks command exits."""
if ctx.ret_code:
LOG.error(ctx)
else:
LOG.debug(ctx)
tasks = {}
# Note for "kill": It is possible for a job to trigger its trap and
# report back to the suite back this logic is called. If so, the task
# will no longer be TASK_STATUS_SUBMITTED or TASK_STATUS_RUNNING, and
# its output line will be ignored here.
for itask in itasks:
if itask.point is not None and itask.submit_num:
submit_num = "%02d" % (itask.submit_num)
tasks[(str(itask.point), itask.tdef.name, submit_num)] = itask
handlers = [(self.batch_sys_mgr.OUT_PREFIX_SUMMARY, summary_callback)]
if more_callbacks:
for prefix, callback in more_callbacks.items():
handlers.append((prefix, callback))
out = ctx.out
if not out:
out = ""
# Something is very wrong here
# Fallback to use "job_log_dirs" list to report the problem
job_log_dirs = ctx.cmd_kwargs.get("job_log_dirs", [])
for job_log_dir in job_log_dirs:
point, name, submit_num = job_log_dir.split(os.sep, 2)
itask = tasks[(point, name, submit_num)]
out += (self.batch_sys_mgr.OUT_PREFIX_SUMMARY +
"|".join([ctx.timestamp, job_log_dir, "1"]) + "\n")
for line in out.splitlines(True):
for prefix, callback in handlers:
if line.startswith(prefix):
line = line[len(prefix):].strip()
try:
path = line.split("|", 2)[1] # timestamp, path, status
point, name, submit_num = path.split(os.sep, 2)
itask = tasks[(point, name, submit_num)]
callback(suite, itask, ctx, line)
except (LookupError, ValueError) as exc:
LOG.warning(
'Unhandled %s output: %s', ctx.cmd_key, line)
LOG.exception(exc)
def _prep_submit_task_job_error(self, suite, itask, dry_run, action, exc):
"""Helper for self._prep_submit_task_job. On error."""
LOG.debug("submit_num %s" % itask.submit_num)
LOG.debug(traceback.format_exc())
LOG.error(exc)
log_task_job_activity(
SubProcContext(self.JOBS_SUBMIT, action, err=exc, ret_code=1),
suite, itask.point, itask.tdef.name, submit_num=itask.submit_num)
if not dry_run:
# Persist
self.suite_db_mgr.put_insert_task_jobs(itask, {
'is_manual_submit': itask.is_manual_submit,
'try_num': itask.get_try_num(),
'time_submit': get_current_time_string(),
'batch_sys_name': itask.summary.get('batch_sys_name'),
})
itask.is_manual_submit = False
self.task_events_mgr.process_message(
itask, CRITICAL, self.task_events_mgr.EVENT_SUBMIT_FAILED)
def load(self):
"""Load or reload configuration from files."""
self.sparse.clear()
self.dense.clear()
LOG.debug("Loading site/user global config files")
conf_path_str = os.getenv("CYLC_CONF_PATH")
if conf_path_str is None:
# CYLC_CONF_PATH not defined, use default locations.
for conf_dir_1, conf_dir_2, conf_type in [
(self.SITE_CONF_DIR, self.SITE_CONF_DIR_OLD,
upgrader.SITE_CONFIG),
(self.USER_CONF_DIR_1, self.USER_CONF_DIR_2,
upgrader.USER_CONFIG)
]:
fname1 = os.path.join(conf_dir_1, self.CONF_BASE)
fname2 = os.path.join(conf_dir_2, self.CONF_BASE)
if os.access(fname1, os.F_OK | os.R_OK):
fname = fname1
elif os.access(fname2, os.F_OK | os.R_OK):
fname = fname2
else:
continue
try:
self.loadcfg(fname, conf_type)
except ParsecError as exc:
if conf_type == upgrader.SITE_CONFIG:
# Warn on bad site file (users can't fix it).
LOG.warning('ignoring bad %s %s:\n%s', conf_type,
fname, exc)
else:
# Abort on bad user file (users can fix it).
LOG.error('bad %s %s', conf_type, fname)
raise
break
elif conf_path_str:
# CYLC_CONF_PATH defined with a value
for path in conf_path_str.split(os.pathsep):
fname = os.path.join(path, self.CONF_BASE)
if os.access(fname, os.F_OK | os.R_OK):
self.loadcfg(fname, upgrader.USER_CONFIG)
# (OK if no global.rc is found, just use system defaults).
self.transform()
def load(self):
"""Load or reload configuration from files."""
self.sparse.clear()
self.dense.clear()
LOG.debug("Loading site/user global config files")
conf_path_str = os.getenv("CYLC_CONF_PATH")
if conf_path_str is None:
# CYLC_CONF_PATH not defined, use default locations.
for conf_dir_1, conf_dir_2, conf_type in [
(self.SITE_CONF_DIR, self.SITE_CONF_DIR_OLD,
upgrader.SITE_CONFIG),
(self.USER_CONF_DIR_1, self.USER_CONF_DIR_2,
upgrader.USER_CONFIG)]:
fname1 = os.path.join(conf_dir_1, self.CONF_BASE)
fname2 = os.path.join(conf_dir_2, self.CONF_BASE)
if os.access(fname1, os.F_OK | os.R_OK):
fname = fname1
elif os.access(fname2, os.F_OK | os.R_OK):
fname = fname2
else:
continue
try:
self.loadcfg(fname, conf_type)
except ParsecError as exc:
if conf_type == upgrader.SITE_CONFIG:
# Warn on bad site file (users can't fix it).
LOG.warning(
'ignoring bad %s %s:\n%s', conf_type, fname, exc)
else:
# Abort on bad user file (users can fix it).
LOG.error('bad %s %s', conf_type, fname)
raise
break
elif conf_path_str:
# CYLC_CONF_PATH defined with a value
for path in conf_path_str.split(os.pathsep):
fname = os.path.join(path, self.CONF_BASE)
if os.access(fname, os.F_OK | os.R_OK):
self.loadcfg(fname, upgrader.USER_CONFIG)
# (OK if no global.rc is found, just use system defaults).
self.transform()
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_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 log_task_job_activity(ctx, suite, point, name, submit_num=None):
"""Log an activity for a task job."""
ctx_str = str(ctx)
if not ctx_str:
return
if isinstance(ctx.cmd_key, tuple): # An event handler
submit_num = ctx.cmd_key[-1]
job_activity_log = get_task_job_activity_log(suite, point, name,
submit_num)
try:
with open(job_activity_log, "ab") as handle:
handle.write(ctx_str + '\n')
except IOError as exc:
# This happens when there is no job directory, e.g. if job host
# selection command causes an submission failure, there will be no job
# directory. In this case, just send the information to the suite log.
LOG.exception(exc)
LOG.info(ctx_str)
if ctx.cmd and ctx.ret_code:
LOG.error(ctx_str)
elif ctx.cmd:
LOG.debug(ctx_str)
def log_task_job_activity(ctx, suite, point, name, submit_num=None):
"""Log an activity for a task job."""
ctx_str = str(ctx)
if not ctx_str:
return
if isinstance(ctx.cmd_key, tuple): # An event handler
submit_num = ctx.cmd_key[-1]
job_activity_log = get_task_job_activity_log(
suite, point, name, submit_num)
try:
with open(job_activity_log, "ab") as handle:
handle.write((ctx_str + '\n').encode())
except IOError as exc:
# This happens when there is no job directory, e.g. if job host
# selection command causes an submission failure, there will be no job
# directory. In this case, just send the information to the suite log.
LOG.exception(exc)
LOG.info(ctx_str)
if ctx.cmd and ctx.ret_code:
LOG.error(ctx_str)
elif ctx.cmd:
LOG.debug(ctx_str)
def _job_logs_retrieval_callback(self, proc_ctx, schd_ctx):
"""Call back when log job retrieval completes."""
if proc_ctx.ret_code:
LOG.error(proc_ctx)
else:
LOG.debug(proc_ctx)
for id_key in proc_ctx.cmd_kwargs["id_keys"]:
key1, point, name, submit_num = id_key
try:
# All completed jobs are expected to have a "job.out".
fnames = [JOB_LOG_OUT]
try:
if key1[1] not in 'succeeded':
fnames.append(JOB_LOG_ERR)
except TypeError:
pass
fname_oks = {}
for fname in fnames:
fname_oks[fname] = os.path.exists(
get_task_job_log(schd_ctx.suite, point, name,
submit_num, fname))
# All expected paths must exist to record a good attempt
log_ctx = SubProcContext((key1, submit_num), None)
if all(fname_oks.values()):
log_ctx.ret_code = 0
del self.event_timers[id_key]
else:
log_ctx.ret_code = 1
log_ctx.err = "File(s) not retrieved:"
for fname, exist_ok in sorted(fname_oks.items()):
if not exist_ok:
log_ctx.err += " %s" % fname
self.event_timers[id_key].unset_waiting()
log_task_job_activity(log_ctx, schd_ctx.suite, point, name,
submit_num)
except KeyError as exc:
LOG.exception(exc)
def _job_logs_retrieval_callback(self, proc_ctx, schd_ctx):
"""Call back when log job retrieval completes."""
if proc_ctx.ret_code:
LOG.error(proc_ctx)
else:
LOG.debug(proc_ctx)
for id_key in proc_ctx.cmd_kwargs["id_keys"]:
key1, point, name, submit_num = id_key
try:
# All completed jobs are expected to have a "job.out".
fnames = [JOB_LOG_OUT]
try:
if key1[1] not in 'succeeded':
fnames.append(JOB_LOG_ERR)
except TypeError:
pass
fname_oks = {}
for fname in fnames:
fname_oks[fname] = os.path.exists(get_task_job_log(
schd_ctx.suite, point, name, submit_num, fname))
# All expected paths must exist to record a good attempt
log_ctx = SubProcContext((key1, submit_num), None)
if all(fname_oks.values()):
log_ctx.ret_code = 0
del self.event_timers[id_key]
else:
log_ctx.ret_code = 1
log_ctx.err = "File(s) not retrieved:"
for fname, exist_ok in sorted(fname_oks.items()):
if not exist_ok:
log_ctx.err += " %s" % fname
self.event_timers[id_key].unset_waiting()
log_task_job_activity(
log_ctx, schd_ctx.suite, point, name, submit_num)
except KeyError as exc:
LOG.exception(exc)
def _setup_custom_event_handlers(self, itask, event, message):
"""Set up custom task event handlers."""
handlers = self._get_events_conf(itask, event + ' handler')
if (handlers is None and event in self._get_events_conf(
itask, 'handler events', [])):
handlers = self._get_events_conf(itask, 'handlers')
if handlers is None:
return
retry_delays = self._get_events_conf(
itask, 'handler retry delays',
self.get_host_conf(itask, "task event handler retry delays"))
if not retry_delays:
retry_delays = [0]
# There can be multiple custom event handlers
for i, handler in enumerate(handlers):
if event in self.NON_UNIQUE_EVENTS:
key1 = ('%s-%02d' % (self.HANDLER_CUSTOM, i), '%s-%d' %
(event, itask.non_unique_events.get(event, 1)))
else:
key1 = ('%s-%02d' % (self.HANDLER_CUSTOM, i), event)
id_key = (key1, str(itask.point), itask.tdef.name,
itask.submit_num)
if id_key in self.event_timers:
continue
# Note: user@host may not always be set for a submit number, e.g.
# on late event or if host select command fails. Use null string to
# prevent issues in this case.
user_at_host = itask.summary['job_hosts'].get(itask.submit_num, '')
if user_at_host and '@' not in user_at_host:
# (only has 'user@' on the front if user is not suite owner).
user_at_host = '%s@%s' % (get_user(), user_at_host)
# Custom event handler can be a command template string
# or a command that takes 4 arguments (classic interface)
# Note quote() fails on None, need str(None).
try:
handler_data = {
"event":
quote(event),
"suite":
quote(self.suite),
'suite_uuid':
quote(str(self.uuid_str)),
"point":
quote(str(itask.point)),
"name":
quote(itask.tdef.name),
"submit_num":
itask.submit_num,
"try_num":
itask.get_try_num(),
"id":
quote(itask.identity),
"message":
quote(message),
"batch_sys_name":
quote(str(itask.summary['batch_sys_name'])),
"batch_sys_job_id":
quote(str(itask.summary['submit_method_id'])),
"submit_time":
quote(str(itask.summary['submitted_time_string'])),
"start_time":
quote(str(itask.summary['started_time_string'])),
"finish_time":
quote(str(itask.summary['finished_time_string'])),
"user@host":
quote(user_at_host)
}
if self.suite_cfg:
for key, value in self.suite_cfg.items():
if key == "URL":
handler_data["suite_url"] = quote(value)
else:
handler_data["suite_" + key] = quote(value)
if itask.tdef.rtconfig['meta']:
for key, value in itask.tdef.rtconfig['meta'].items():
if key == "URL":
handler_data["task_url"] = quote(value)
handler_data[key] = quote(value)
cmd = handler % (handler_data)
except KeyError as exc:
message = "%s/%s/%02d %s bad template: %s" % (
itask.point, itask.tdef.name, itask.submit_num, key1, exc)
LOG.error(message)
continue
if cmd == handler:
# Nothing substituted, assume classic interface
cmd = "%s '%s' '%s' '%s' '%s'" % (handler, event, self.suite,
itask.identity, message)
LOG.debug("[%s] -Queueing %s handler: %s", itask, event, cmd)
self.event_timers[id_key] = (TaskActionTimer(
CustomTaskEventHandlerContext(
key1,
self.HANDLER_CUSTOM,
cmd,
), retry_delays))
def _setup_custom_event_handlers(self, itask, event, message):
"""Set up custom task event handlers."""
handlers = self._get_events_conf(itask, event + ' handler')
if (handlers is None and
event in self._get_events_conf(itask, 'handler events', [])):
handlers = self._get_events_conf(itask, 'handlers')
if handlers is None:
return
retry_delays = self._get_events_conf(
itask,
'handler retry delays',
self.get_host_conf(itask, "task event handler retry delays"))
if not retry_delays:
retry_delays = [0]
# There can be multiple custom event handlers
for i, handler in enumerate(handlers):
if event in self.NON_UNIQUE_EVENTS:
key1 = (
'%s-%02d' % (self.HANDLER_CUSTOM, i),
'%s-%d' % (event, itask.non_unique_events.get(event, 1)))
else:
key1 = ('%s-%02d' % (self.HANDLER_CUSTOM, i), event)
id_key = (
key1, str(itask.point), itask.tdef.name, itask.submit_num)
if id_key in self.event_timers:
continue
# Note: user@host may not always be set for a submit number, e.g.
# on late event or if host select command fails. Use null string to
# prevent issues in this case.
user_at_host = itask.summary['job_hosts'].get(itask.submit_num, '')
if user_at_host and '@' not in user_at_host:
# (only has 'user@' on the front if user is not suite owner).
user_at_host = '%s@%s' % (get_user(), user_at_host)
# Custom event handler can be a command template string
# or a command that takes 4 arguments (classic interface)
# Note quote() fails on None, need str(None).
try:
handler_data = {
"event": quote(event),
"suite": quote(self.suite),
'suite_uuid': quote(str(self.uuid_str)),
"point": quote(str(itask.point)),
"name": quote(itask.tdef.name),
"submit_num": itask.submit_num,
"try_num": itask.get_try_num(),
"id": quote(itask.identity),
"message": quote(message),
"batch_sys_name": quote(
str(itask.summary['batch_sys_name'])),
"batch_sys_job_id": quote(
str(itask.summary['submit_method_id'])),
"submit_time": quote(
str(itask.summary['submitted_time_string'])),
"start_time": quote(
str(itask.summary['started_time_string'])),
"finish_time": quote(
str(itask.summary['finished_time_string'])),
"user@host": quote(user_at_host)
}
if self.suite_cfg:
for key, value in self.suite_cfg.items():
if key == "URL":
handler_data["suite_url"] = quote(value)
else:
handler_data["suite_" + key] = quote(value)
if itask.tdef.rtconfig['meta']:
for key, value in itask.tdef.rtconfig['meta'].items():
if key == "URL":
handler_data["task_url"] = quote(value)
handler_data[key] = quote(value)
cmd = handler % (handler_data)
except KeyError as exc:
message = "%s/%s/%02d %s bad template: %s" % (
itask.point, itask.tdef.name, itask.submit_num, key1, exc)
LOG.error(message)
continue
if cmd == handler:
# Nothing substituted, assume classic interface
cmd = "%s '%s' '%s' '%s' '%s'" % (
handler, event, self.suite, itask.identity, message)
LOG.debug("[%s] -Queueing %s handler: %s", itask, event, cmd)
self.event_timers[id_key] = (
TaskActionTimer(
CustomTaskEventHandlerContext(
key1,
self.HANDLER_CUSTOM,
cmd,
),
retry_delays))
