def _receiver(self, message):
"""Wrap incoming messages and dispatch them to exposed methods.
Args:
message (dict): message contents
"""
# determine the server method to call
try:
method = getattr(self, message['command'])
args = message['args']
args.update({'user': message['user']})
if 'meta' in message:
args['meta'] = message['meta']
except KeyError:
# malformed message
return {'error': {
'message': 'Request missing required field(s).'}}
except AttributeError:
# no exposed method by that name
return {'error': {
'message': 'No method by the name "%s"' % message['command']}}
# generate response
try:
response = method(**args)
except Exception as exc:
# includes incorrect arguments (TypeError)
LOG.exception(exc) # note the error server side
import traceback
return {'error': {
'message': str(exc), 'traceback': traceback.format_exc()}}
return {'data': response}
def create_directory(dir_, name):
"""Create directory. Raise GlobalConfigError on error."""
try:
os.makedirs(dir_, exist_ok=True)
except OSError as exc:
LOG.exception(exc)
raise GlobalConfigError(
'Failed to create directory "' + name + '"')
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 _event_email_callback(self, proc_ctx, schd_ctx):
"""Call back when email notification command exits."""
for id_key in proc_ctx.cmd_kwargs["id_keys"]:
key1, point, name, submit_num = id_key
try:
if proc_ctx.ret_code == 0:
del self.event_timers[id_key]
log_ctx = SubProcContext((key1, submit_num), None)
log_ctx.ret_code = 0
log_task_job_activity(
log_ctx, schd_ctx.suite, point, name, submit_num)
else:
self.event_timers[id_key].unset_waiting()
except KeyError as exc:
LOG.exception(exc)
def _event_email_callback(self, proc_ctx, schd_ctx):
"""Call back when email notification command exits."""
for id_key in proc_ctx.cmd_kwargs["id_keys"]:
key1, point, name, submit_num = id_key
try:
if proc_ctx.ret_code == 0:
del self.event_timers[id_key]
log_ctx = SubProcContext((key1, submit_num), None)
log_ctx.ret_code = 0
log_task_job_activity(log_ctx, schd_ctx.suite, point, name,
submit_num)
else:
self.event_timers[id_key].unset_waiting()
except KeyError as exc:
LOG.exception(exc)
def _run_command_init(cls, ctx, callback=None, callback_args=None):
"""Prepare and launch shell command in ctx."""
try:
if ctx.cmd_kwargs.get('stdin_files'):
if len(ctx.cmd_kwargs['stdin_files']) > 1:
stdin_file = TemporaryFile()
for file_ in ctx.cmd_kwargs['stdin_files']:
if hasattr(file_, 'read'):
stdin_file.write(file_.read())
else:
stdin_file.write(open(file_, 'rb').read())
stdin_file.seek(0)
elif hasattr(ctx.cmd_kwargs['stdin_files'][0], 'read'):
stdin_file = ctx.cmd_kwargs['stdin_files'][0]
else:
stdin_file = open(ctx.cmd_kwargs['stdin_files'][0], 'rb')
elif ctx.cmd_kwargs.get('stdin_str'):
stdin_file = TemporaryFile('bw+')
stdin_file.write(ctx.cmd_kwargs.get('stdin_str').encode())
stdin_file.seek(0)
else:
stdin_file = open(os.devnull)
proc = procopen(
ctx.cmd,
stdin=stdin_file,
stdoutpipe=True,
stderrpipe=True,
# Execute command as a process group leader,
# so we can use "os.killpg" to kill the whole group.
preexec_fn=os.setpgrp,
env=ctx.cmd_kwargs.get('env'),
usesh=ctx.cmd_kwargs.get('shell'))
# calls to open a shell are aggregated in cylc_subproc.procopen()
# with logging for what is calling it and the commands given
except (IOError, OSError) as exc:
if exc.filename is None:
exc.filename = ctx.cmd[0]
LOG.exception(exc)
ctx.ret_code = 1
ctx.err = str(exc)
cls._run_command_exit(ctx, callback, callback_args)
return None
else:
LOG.debug(ctx.cmd)
return proc
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 _run_command_init(cls, ctx, callback=None, callback_args=None):
"""Prepare and launch shell command in ctx."""
try:
if ctx.cmd_kwargs.get('stdin_file_paths'):
if len(ctx.cmd_kwargs['stdin_file_paths']) > 1:
stdin_file = TemporaryFile()
for file_path in ctx.cmd_kwargs['stdin_file_paths']:
stdin_file.write(open(file_path, 'rb').read())
stdin_file.seek(0)
else:
stdin_file = open(ctx.cmd_kwargs['stdin_file_paths'][0],
'rb')
elif ctx.cmd_kwargs.get('stdin_str'):
stdin_file = TemporaryFile()
stdin_file.write(ctx.cmd_kwargs.get('stdin_str'))
stdin_file.seek(0)
else:
stdin_file = open(os.devnull)
proc = Popen(
ctx.cmd,
stdin=stdin_file,
stdout=PIPE,
stderr=PIPE,
# Execute command as a process group leader,
# so we can use "os.killpg" to kill the whole group.
preexec_fn=os.setpgrp,
env=ctx.cmd_kwargs.get('env'),
shell=ctx.cmd_kwargs.get('shell'))
except (IOError, OSError) as exc:
if exc.filename is None:
exc.filename = ctx.cmd[0]
LOG.exception(exc)
ctx.ret_code = 1
ctx.err = str(exc)
cls._run_command_exit(ctx, callback, callback_args)
return None
else:
LOG.debug(ctx.cmd)
return proc
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 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 _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 _receiver(self, message):
"""Wrap incoming messages and dispatch them to exposed methods."""
# determine the server method to call
try:
method = getattr(self, message['command'])
args = message['args']
args.update({'user': message['user']})
if 'meta' in message:
args['meta'] = message['meta']
except KeyError:
# malformed message
return {'error': {'message': 'Request missing required field(s).'}}
except AttributeError:
# no exposed method by that name
return {
'error': {
'message':
'No method by the name "%s"' % message['command']
}
}
# generate response
try:
response = method(**args)
except Exception as exc:
# includes incorrect arguments (TypeError)
LOG.exception(exc) # note the error server side
import traceback
return {
'error': {
'message': str(exc),
'traceback': traceback.format_exc()
}
}
return {'data': response}
