def runfile(script=None,
raw_args='',
wdir='.',
code=None,
kernel=None,
**kwargs):
# this has something to do with Prefix matching rule of parse_known_args
#
# That is to say
#
# --rep 3
#
# would be parsed as
#
# args.workflow=3, unknown --rep
#
# instead of
#
# args.workflow=None, unknown --rep 3
#
# we then have to change the parse to disable args.workflow when
# there is no workflow option.
raw_args = shlex.split(raw_args) if isinstance(raw_args, str) else raw_args
if (script is None and code is None) or '-h' in raw_args:
parser = get_run_parser(interactive=True, with_workflow=True)
parser.print_help()
return
if raw_args and raw_args[0].lstrip().startswith('-'):
parser = get_run_parser(interactive=True, with_workflow=False)
parser.error = _parse_error
args, workflow_args = parser.parse_known_args(raw_args)
args.workflow = None
else:
parser = get_run_parser(interactive=True, with_workflow=True)
parser.error = _parse_error
args, workflow_args = parser.parse_known_args(raw_args)
# for reporting purpose
sys.argv = ['%run'] + raw_args
env.verbosity = args.verbosity
if kernel and not isinstance(env.logger.handlers[0],
NotebookLoggingHandler):
env.logger.handlers = []
levels = {
0: logging.ERROR,
1: logging.WARNING,
2: logging.INFO,
3: logging.DEBUG,
4: logging.TRACE,
None: logging.INFO
}
env.logger.addHandler(
NotebookLoggingHandler(levels[env.verbosity],
kernel,
title=' '.join(sys.argv)))
else:
env.logger.handers[0].setTitle(' '.join(sys.argv))
dt = datetime.datetime.now().strftime('%m%d%y_%H%M')
if args.__dag__ is None:
args.__dag__ = f'workflow_{dt}.dot'
elif args.__dag__ == '':
args.__dag__ = None
if args.__report__ is None:
args.__report__ = f'workflow_{dt}.html'
elif args.__report__ == '':
args.__report__ = None
if args.__remote__:
from sos.utils import load_config_files
cfg = load_config_files(args.__config__)
env.sos_dict.set('CONFIG', cfg)
# if executing on a remote host...
from sos.hosts import Host
host = Host(args.__remote__)
#
if script is None:
if not code.strip():
return
script = os.path.join('.sos', '__interactive__.sos')
with open(script, 'w') as s:
s.write(code)
# copy script to remote host...
host.send_to_host(script)
from sos.utils import remove_arg
argv = shlex.split(raw_args) if isinstance(raw_args, str) else raw_args
argv = remove_arg(argv, '-r')
argv = remove_arg(argv, '-c')
# execute the command on remote host
try:
with kernel.redirect_sos_io():
ret = host._host_agent.run_command(['sos', 'run', script] +
argv,
wait_for_task=True,
realtime=True)
if ret:
kernel.send_response(
kernel.iopub_socket, 'stream',
dict(name='stderr',
text=
f'remote execution of workflow exited with code {ret}'
))
except Exception as e:
if kernel:
kernel.send_response(kernel.iopub_socket, 'stream', {
'name': 'stdout',
'text': str(e)
})
return
if args.__bin_dirs__:
for d in args.__bin_dirs__:
if d == '~/.sos/bin' and not os.path.isdir(os.path.expanduser(d)):
os.makedirs(os.path.expanduser(d), exist_ok=True)
os.environ['PATH'] = os.pathsep.join(
[os.path.expanduser(x)
for x in args.__bin_dirs__]) + os.pathsep + os.environ['PATH']
# clear __step_input__, __step_output__ etc because there is
# no concept of passing input/outputs across cells.
env.sos_dict.set('__step_output__', sos_targets([]))
for k in [
'__step_input__', '__default_output__', 'step_input',
'step_output', 'step_depends', '_input', '_output', '_depends'
]:
env.sos_dict.pop(k, None)
try:
if script is None:
if not code.strip():
return
if kernel is None:
script = SoS_Script(content=code)
else:
if kernel._workflow_mode:
# in workflow mode, the content is sent by magics %run and %sosrun
script = SoS_Script(content=code)
else:
# this is a scratch step...
# if there is no section header, add a header so that the block
# appears to be a SoS script with one section
if not any([
SOS_SECTION_HEADER.match(line)
or line.startswith('%from')
or line.startswith('%include')
for line in code.splitlines()
]):
code = '[scratch_0]\n' + code
script = SoS_Script(content=code)
else:
#kernel.send_frontend_msg('stream',
# {'name': 'stdout', 'text': 'Workflow cell can only be executed with magic %run or %sosrun.'},
# title='# SoS warning')
return
else:
script = SoS_Script(filename=script)
workflow = script.workflow(args.workflow,
use_default=not args.__targets__)
env.config: DefaultDict[str, Union[None, bool, str]] = defaultdict(str)
executor = Interactive_Executor(
workflow,
args=workflow_args,
config={
'config_file':
args.__config__,
'output_dag':
args.__dag__,
'output_report':
args.__report__,
'sig_mode':
'ignore' if args.dryrun else args.__sig_mode__,
'default_queue':
'' if args.__queue__ is None else args.__queue__,
'wait_for_task':
True if args.__wait__ is True or args.dryrun else
(False if args.__no_wait__ else None),
'resume_mode':
kernel is not None and kernel._resume_execution,
'run_mode':
'dryrun' if args.dryrun else 'interactive',
'verbosity':
args.verbosity,
# wait if -w or in dryrun mode, not wait if -W, otherwise use queue default
'max_procs':
args.__max_procs__,
'max_running_jobs':
args.__max_running_jobs__,
# for infomration and resume only
'workdir':
os.getcwd(),
'script':
"interactive",
'workflow':
args.workflow,
'targets':
args.__targets__,
'bin_dirs':
args.__bin_dirs__,
'workflow_args':
workflow_args
})
return executor.run(args.__targets__)['__last_res__']
except PendingTasks:
raise
except SystemExit:
# this happens because the executor is in resume mode but nothing
# needs to be resumed, we simply pass
return
except Exception:
if args.verbosity and args.verbosity > 2:
sys.stderr.write(get_traceback())
raise
finally:
env.config['sig_mode'] = 'ignore'
env.verbosity = 2
def main():
from argparse import ArgumentParser, ArgumentDefaultsHelpFormatter, SUPPRESS
class ArgumentParserError(Exception):
pass
class MyArgParser(ArgumentParser):
def error(self, message):
raise ArgumentParserError(message)
#
p = MyArgParser(description=__doc__,
formatter_class=ArgumentDefaultsHelpFormatter,
add_help=False)
p.add_argument('dsc_file',
metavar="DSC script",
help='DSC script to execute.')
ce = p.add_argument_group('Benchmark options')
ce.add_argument('--target',
metavar="str",
nargs='+',
help='''This argument can be used in two contexts:
1) When used without "-d" it overrides "DSC::run" in DSC file.
Input should be quoted string(s) defining one or multiple valid DSC pipelines
(multiple pipelines should be separated by space).
2) When used along with "-d" it specifies one or more computational modules,
separated by space, whose output are to be removed or replaced by a (smaller) placeholder file.'''
)
ce.add_argument(
'--truncate',
action='store_true',
help='''When applied, DSC will only run one value per parameter.
For example with "--truncate", "n: 1,2,3,4,5" will be truncated to "n: 1".
This can be used in exploratory analysis and diagnostics, particularly when used in combination with "--target".'''
)
ce.add_argument(
'--replicate',
metavar='N',
type=int,
help=
'''Overrides "DSC::replicate" to set number of replicates. Will be set to 1 when "--truncate" is in action.'''
)
ce.add_argument(
'-o',
metavar="str",
dest='output',
help=
'''Benchmark output. It overrides "DSC::output" defined in DSC file.'''
)
mt = p.add_argument_group('Execution modes')
mt.add_argument('-s',
metavar="option",
choices=["strict", "lenient", "existing", "all", "none"],
dest='__construct__',
default="strict",
help='''How DSC skips or overwrites existing results.
"strict": skips jobs whose input, output and code have not been changed since previous execution.
"lenient": skips jobs whose output timestamp are newer than their input.
It can be used to avoid re-run when nuisent changes are made to module scripts that should not impact results.
"existing": skips jobs whose output exists, and mark existing output as "up-to-date" for future re-runs.
It can be used to avoid re-run completely even after file status cache have been deleted (as a result of "-d all" option).
"all": skips all modules and only build meta-database required to run `dsc-query` command.
It can be used for salvaging a partially completed benchmark making it possible to query from it.
"none": force executes DSC from scratch.''')
mt.add_argument('--touch',
action='store_true',
dest='__recover__',
help=SUPPRESS)
mt.add_argument('-e',
metavar='option',
choices=['stop', 'ignore', 'kill'],
dest='error_option',
help=SUPPRESS)
mt.add_argument('-d',
metavar="option",
choices=["obsolete", "replace", "all"],
dest='to_remove',
help='''How DSC deletes benchmark files.
Use option "all" to remove all output from the current benchmark.
"obsolete", when used without "--target", removes from output folder anything irrelevant
to the most recent successful execution of the benchmark.
When used with "--target" it deletes specified files, or files from specified modules or module groups.
"replace", when used with "--target", deletes files as option "obsolete" does with "--target",
but additionally puts in placeholder files with "*.zapped" extension to prevent the module from being executed
until they are needed for re-running a downstream module.
It can be used to remove large yet unused intermediate module output without triggering re-runs when possible.'''
)
ro = p.add_argument_group('Computing options')
ro.add_argument(
'-c',
type=int,
metavar='N',
default=max(min(int(os.cpu_count() / 2), 8), 1),
dest='__max_jobs__',
help=
'''Maximum number of CPU threads for local runs, or job managing sockets for remote execution.'''
)
ro.add_argument(
'-v',
'--verbosity',
type=int,
choices=list(range(5)),
default=2,
help='''Output error (0), warning (1), info (2), debug (3) and trace (4)
information.''')
ro.add_argument(
'-g',
dest='__dag__',
action='store_true',
help='''Output benchmark execution graph animation in HTML format.''')
rt = p.add_argument_group('HPC settings')
rt.add_argument(
'--host',
metavar='file',
help='''Configuration file for DSC computational environments.''')
ot = p.add_argument_group('Other options')
ot.add_argument('--version', action='version', version=__version__)
ot.add_argument("-h",
"--help",
action="help",
help="show this help message and exit")
ot.add_argument('--debug', action='store_true', help=SUPPRESS)
p.set_defaults(func=execute)
if len(sys.argv) > 2 and '-h' in sys.argv:
try:
from .dsc_parser import DSC_Script
script = DSC_Script(sys.argv[1])
script.print_help(print_version='-v' in sys.argv)
sys.exit(0)
except Exception as e:
if '--debug' in sys.argv:
raise
else:
env.logger.error(
f'No help information is available for script {sys.argv[1]}: ``{e}``'
)
sys.exit(1)
try:
args, unknown_args = p.parse_known_args()
except Exception as e:
env.logger.error(e)
env.logger.info("Please type ``{} -h`` to view available options".\
format(os.path.basename(sys.argv[0])))
sys.exit(1)
#
env.verbosity = args.verbosity
# keep `args.__recover__` to maintain backwards compatibility for `--touch` option.
if args.__recover__:
env.logger.warning(
f'Option ``--touch`` is deprecated. Please use ``-s existing`` next time.'
)
args.__construct__ = 'existing'
with Timer(verbose=True if (args.verbosity > 0) else False) as t:
try:
args.func(args, unknown_args)
except KeyboardInterrupt:
t.disable()
sys.exit('KeyboardInterrupt')
except Exception as e:
if args.debug:
raise
if args.verbosity > 2:
sys.stderr.write(get_traceback())
t.disable()
env.logger.error(e)
sys.exit(1)
def execute_scratch_cell(code, raw_args, kernel):
# we then have to change the parse to disable args.workflow when
# there is no workflow option.
raw_args = shlex.split(raw_args) if isinstance(raw_args, str) else raw_args
if code is None or '-h' in raw_args:
parser = get_run_parser(interactive=True, with_workflow=True)
parser.print_help()
return
if raw_args and raw_args[0].lstrip().startswith('-'):
parser = get_run_parser(interactive=True, with_workflow=False)
parser.error = _parse_error
args, workflow_args = parser.parse_known_args(raw_args)
args.workflow = None
else:
parser = get_run_parser(interactive=True, with_workflow=True)
parser.error = _parse_error
args, workflow_args = parser.parse_known_args(raw_args)
if not code.strip():
return
# for reporting purpose
sys.argv = ['%run'] + raw_args
env.verbosity = args.verbosity
if not any(
isinstance(x, NotebookLoggingHandler)
for x in env.logger.handlers):
env.logger.handlers = [
x for x in env.logger.handlers
if type(x) is not logging.StreamHandler
]
levels = {
0: logging.ERROR,
1: logging.WARNING,
2: logging.INFO,
3: logging.DEBUG,
4: logging.DEBUG,
None: logging.INFO
}
env.logger.addHandler(
NotebookLoggingHandler(levels[env.verbosity],
kernel,
title=' '.join(sys.argv)))
else:
env.logger.handers[0].setTitle(' '.join(sys.argv))
global last_cell_id
# we retain step_input etc only when we step through a cell #256
if kernel and kernel.cell_id != last_cell_id:
# clear __step_input__, __step_output__ etc because there is
# no concept of passing input/outputs across cells.
env.sos_dict.set('__step_output__', sos_targets([]))
for k in [
'__step_input__', '__default_output__', 'step_input',
'step_output', 'step_depends', '_input', '_output', '_depends'
]:
env.sos_dict.pop(k, None)
last_cell_id = kernel.cell_id
config = {
'config_file':
args.__config__,
'default_queue':
args.__queue__,
'run_mode':
'dryrun' if args.dryrun else 'interactive',
# issue 230, ignore sig mode in interactive mode
'sig_mode':
'ignore',
'verbosity':
args.verbosity,
# for backward compatibility, we try both args.__worker_procs__ and args.__max_procs__
'worker_procs':
args.__worker_procs__
if hasattr(args, '__worker_procs__') else args.__max_procs__,
'max_running_jobs':
args.__max_running_jobs__,
# for infomration and resume only
'workdir':
os.getcwd(),
'workflow':
args.workflow,
'targets':
args.__targets__,
'workflow_args':
workflow_args,
'workflow_id':
textMD5(code),
# interactive work is also a slave of the controller
'slave_id':
kernel.cell_id,
}
env.sos_dict.set('workflow_id', config['workflow_id'])
env.config.update(config)
try:
if not any([
SOS_SECTION_HEADER.match(line) or line.startswith('%from')
or line.startswith('%include') for line in code.splitlines()
]):
code = f'[cell_{str(kernel.cell_id)[:8] if kernel and kernel.cell_id else "0"}]\n' + code
script = SoS_Script(content=code)
else:
return
workflow = script.workflow(args.workflow)
section = workflow.sections[0]
res = analyze_section(section)
env.sos_dict.quick_update({
'__signature_vars__': res['signature_vars'],
'__environ_vars__': res['environ_vars'],
'__changed_vars__': res['changed_vars']
})
executor = Interactive_Step_Executor(section, mode='interactive')
ret = executor.run()
try:
return ret['__last_res__']
except Exception as e:
raise RuntimeError(
f'Unknown result returned from executor {ret}: {e}')
except (UnknownTarget, RemovedTarget) as e:
raise RuntimeError(f'Unavailable target {e.target}')
except TerminateExecution as e:
return
except SystemExit:
# this happens because the executor is in resume mode but nothing
# needs to be resumed, we simply pass
return
except Exception:
env.log_to_file('PROCESS', get_traceback())
raise
def runfile(script=None,
raw_args='',
wdir='.',
code=None,
kernel=None,
**kwargs):
# this has something to do with Prefix matching rule of parse_known_args
#
# That is to say
#
# --rep 3
#
# would be parsed as
#
# args.workflow=3, unknown --rep
#
# instead of
#
# args.workflow=None, unknown --rep 3
#
# we then have to change the parse to disable args.workflow when
# there is no workflow option.
args = shlex.split(raw_args) if isinstance(raw_args, str) else raw_args
if (script is None and code is None) or '-h' in args:
parser = get_run_parser(interactive=True, with_workflow=True)
parser.print_help()
return
if args and args[0].lstrip().startswith('-'):
parser = get_run_parser(interactive=True, with_workflow=False)
parser.error = _parse_error
args, workflow_args = parser.parse_known_args(args)
args.workflow = None
else:
parser = get_run_parser(interactive=True, with_workflow=True)
parser.error = _parse_error
args, workflow_args = parser.parse_known_args(args)
# no multi-processing in interactive mode
env.max_jobs = 1
env.verbosity = args.verbosity
if args.__queue__ == '':
from sos.hosts import list_queues
list_queues(args.__config__, args.verbosity)
return
if args.__remote__:
from sos.utils import load_config_files
cfg = load_config_files(args.__config__)
env.sos_dict.set('CONFIG', cfg)
if args.__remote__ == '':
from .hosts import list_queues
list_queues(cfg, args.verbosity)
return
# if executing on a remote host...
from sos.hosts import Host
host = Host(args.__remote__)
#
if script is None:
if not code.strip():
return
script = os.path.join('.sos', '__interactive__.sos')
with open(script, 'w') as s:
s.write(code)
# copy script to remote host...
host.send_to_host(script)
from sos.utils import remove_arg
argv = shlex.split(raw_args) if isinstance(raw_args, str) else raw_args
argv = remove_arg(argv, '-r')
argv = remove_arg(argv, '-c')
# execute the command on remote host
try:
with kernel.redirect_sos_io():
ret = host._host_agent.run_command(['sos', 'run', script] +
argv,
wait_for_task=True,
realtime=True)
if ret:
kernel.send_response(
kernel.iopub_socket, 'stream',
dict(name='stderr',
text=
f'remote execution of workflow exited with code {ret}'
))
except Exception as e:
if kernel:
kernel.send_response(kernel.iopub_socket, 'stream', {
'name': 'stdout',
'text': str(e)
})
return
if args.__bin_dirs__:
import fasteners
for d in args.__bin_dirs__:
if d == '~/.sos/bin' and not os.path.isdir(os.path.expanduser(d)):
with fasteners.InterProcessLock(
os.path.join(tempfile.gettempdir(), 'sos_lock_bin')):
os.makedirs(os.path.expanduser(d))
elif not os.path.isdir(os.path.expanduser(d)):
raise ValueError(f'directory does not exist: {d}')
os.environ['PATH'] = os.pathsep.join(
[os.path.expanduser(x)
for x in args.__bin_dirs__]) + os.pathsep + os.environ['PATH']
# clear __step_input__, __step_output__ etc because there is
# no concept of passing input/outputs across cells.
env.sos_dict.set('__step_output__', [])
for k in ['__step_input__', '__default_output__', 'input', 'output', \
'depends', '_input', '_output', '_depends']:
env.sos_dict.pop(k, None)
try:
if script is None:
if not code.strip():
return
if kernel is None:
script = SoS_Script(content=code)
else:
if kernel._workflow_mode:
# in workflow mode, the content is sent by magics %run and %sosrun
script = SoS_Script(content=code)
else:
# this is a scratch step...
# if there is no section header, add a header so that the block
# appears to be a SoS script with one section
if not any([
SOS_SECTION_HEADER.match(line)
or line.startswith('%from')
or line.startswith('%include')
for line in code.splitlines()
]):
code = '[scratch_0]\n' + code
script = SoS_Script(content=code)
else:
if kernel.cell_idx == -1:
kernel.send_frontend_msg(
'stream', {
'name':
'stdout',
'text':
'Workflow can only be executed with magic %run or %sosrun.'
})
return
else:
script = SoS_Script(filename=script)
workflow = script.workflow(args.workflow)
executor = Interactive_Executor(
workflow,
args=workflow_args,
config={
'config_file':
args.__config__,
'output_dag':
args.__dag__,
'sig_mode':
args.__sig_mode__,
'default_queue':
'' if args.__queue__ is None else args.__queue__,
'wait_for_task':
True if args.__wait__ is True or args.dryrun else
(False if args.__no_wait__ else None),
'resume_mode':
kernel is not None and kernel._resume_execution,
'run_mode':
'dryrun' if args.dryrun else 'interactive',
'verbosity':
args.verbosity,
# wait if -w or in dryrun mode, not wait if -W, otherwise use queue default
'max_procs':
1,
'max_running_jobs':
args.__max_running_jobs__,
# for infomration and resume only
'workdir':
os.getcwd(),
'script':
"interactive",
'workflow':
args.workflow,
'targets':
args.__targets__,
'bin_dirs':
args.__bin_dirs__,
'workflow_args':
workflow_args
})
return executor.run(args.__targets__)
except PendingTasks:
raise
except SystemExit:
# this happens because the executor is in resume mode but nothing
# needs to be resumed, we simply pass
return
except Exception:
if args.verbosity and args.verbosity > 2:
sys.stderr.write(get_traceback())
raise
finally:
env.config['sig_mode'] = 'ignore'
env.verbosity = 2
def run(self,
targets=None,
parent_pipe=None,
my_workflow_id=None,
mode='run'):
'''Execute a block of SoS script that is sent by iPython/Jupyer/Spyer
The code can be simple SoS/Python statements, one SoS step, or more
or more SoS workflows with multiple steps. This executor,
1. adds a section header to the script if there is no section head
2. execute the workflow in interactive mode, which is different from
batch mode in a number of ways, which most notably without support
for nested workflow.
3. Optionally execute the workflow in preparation mode for debugging purposes.
'''
# if there is no valid code do nothing
self.reset_dict()
# this is the result returned by the workflow, if the
# last stement is an expression.
last_res = None
# process step of the pipelinp
if isinstance(targets, str):
targets = [targets]
dag = self.initialize_dag(targets=targets)
#
# if targets are specified and there are only signatures for them, we need
# to remove the signature and really generate them
if targets:
for t in targets:
if not file_target(t).target_exists('target') and file_target(
t).target_exists('signature'):
env.logger.debug(f'Re-generating {t}')
file_target(t).remove('signature')
else:
env.logger.debug(f'Target {t} already exists')
#
while True:
# find any step that can be executed and run it, and update the DAT
# with status.
runnable = dag.find_executable()
if runnable is None:
# no runnable
#dag.show_nodes()
break
# find the section from runnable
section = self.workflow.section_by_id(runnable._step_uuid)
#
# this is to keep compatibility of dag run with sequential run because
# in sequential run, we evaluate global section of each step in
# order to determine values of options such as skip.
# The consequence is that global definitions are available in
# SoS namespace.
try:
SoS_exec(section.global_def)
except Exception as e:
if env.verbosity > 2:
sys.stderr.write(get_traceback())
raise RuntimeError(
f'Failed to execute statements\n"{section.global_def}"\n{e}'
)
# clear existing keys, otherwise the results from some random result
# might mess with the execution of another step that does not define input
for k in [
'__step_input__', '__default_output__', '__step_output__'
]:
if k in env.sos_dict:
env.sos_dict.pop(k)
# if the step has its own context
env.sos_dict.quick_update(runnable._context)
# execute section with specified input
runnable._status = 'running'
try:
executor = Interactive_Step_Executor(section)
res = executor.run()
for k, v in res.items():
env.sos_dict.set(k, v)
last_res = res['__last_res__']
# set context to the next logic step.
for edge in dag.out_edges(runnable):
node = edge[1]
# if node is the logical next step...
if node._node_index is not None and runnable._node_index is not None:
#and node._node_index == runnable._node_index + 1:
node._context.update(env.sos_dict.clone_selected_vars(
node._context['__signature_vars__'] | node._context['__environ_vars__'] \
| {'_input', '__step_output__', '__default_output__', '__args__'}))
node._context['__completed__'].append(res['__step_name__'])
runnable._status = 'completed'
except (UnknownTarget, RemovedTarget) as e:
runnable._status = None
target = e.target
if dag.regenerate_target(target):
#runnable._depends_targets.append(target)
#dag._all_dependent_files[target].append(runnable)
dag.build(self.workflow.auxiliary_sections)
#
cycle = dag.circular_dependencies()
if cycle:
raise RuntimeError(
f'Circular dependency detected {cycle} after regeneration. It is likely a later step produces input of a previous step.'
)
else:
if self.resolve_dangling_targets(dag, [target]) == 0:
raise RuntimeError(
f'Failed to regenerate or resolve {target}{dag.steps_depending_on(target, self.workflow)}.'
)
runnable._depends_targets.append(target)
dag._all_dependent_files[target].append(runnable)
dag.build(self.workflow.auxiliary_sections)
#
cycle = dag.circular_dependencies()
if cycle:
raise RuntimeError(
f'Circular dependency detected {cycle}. It is likely a later step produces input of a previous step.'
)
self.save_dag(dag)
except UnavailableLock as e:
runnable._status = 'pending'
runnable._signature = (e.output, e.sig_file)
env.logger.debug(
f'Waiting on another process for step {section.step_name()}'
)
except PendingTasks as e:
self.record_quit_status(e.tasks)
raise
# if the job is failed
except Exception as e:
runnable._status = 'failed'
raise
if self.md5:
self.save_workflow_signature(dag)
env.logger.debug(
f'Workflow {self.workflow.name} (ID={self.md5}) is executed successfully.'
)
# remove task pending status if the workflow is completed normally
try:
wf_status = os.path.join(os.path.expanduser('~'), '.sos',
self.md5 + '.status')
if os.path.isfile(wf_status):
os.remove(wf_status)
except Exception as e:
env.logger.warning(f'Failed to clear workflow status file: {e}')
return last_res
def runfile(script=None, args='', wdir='.', code=None, **kwargs):
# this has something to do with Prefix matching rule of parse_known_args
#
# That is to say
#
# --rep 3
#
# would be parsed as
#
# args.workflow=3, unknown --rep
#
# instead of
#
# args.workflow=None, unknown --rep 3
#
# we then have to change the parse to disable args.workflow when
# there is no workflow option.
if isinstance(args, str):
args = shlex.split(args)
if (script is None and code is None) or '-h' in args:
parser = get_run_parser(interactive=True, with_workflow=True)
parser.print_help()
return
if args and args[0].lstrip().startswith('-'):
parser = get_run_parser(interactive=True, with_workflow=False)
parser.error = _parse_error
args, workflow_args = parser.parse_known_args(args)
args.workflow = None
else:
parser = get_run_parser(interactive=True, with_workflow=True)
parser.error = _parse_error
args, workflow_args = parser.parse_known_args(args)
# no multi-processing in interactive mode
env.max_jobs = 1
env.verbosity = args.verbosity
env.__task_engine__ = 'interactive'
#
env.sig_mode = args.__sigmode__
if args.__bin_dirs__:
import fasteners
for d in args.__bin_dirs__:
if d == '~/.sos/bin' and not os.path.isdir(os.path.expanduser(d)):
with fasteners.InterProcessLock('/tmp/sos_lock_bin'):
os.makedirs(os.path.expanduser(d))
elif not os.path.isdir(os.path.expanduser(d)):
raise ValueError('directory does not exist: {}'.format(d))
os.environ['PATH'] = os.pathsep.join([os.path.expanduser(x) for x in args.__bin_dirs__]) + os.pathsep + os.environ['PATH']
# clear __step_input__, __step_output__ etc because there is
# no concept of passing input/outputs across cells.
env.sos_dict.set('__step_output__', [])
for k in ['__step_input__', '__default_output__', 'input', 'output', \
'depends', '_input', '_output', '_depends']:
env.sos_dict.pop(k, None)
try:
if script is None:
if not code.strip():
return
# if there is no section header, add a header so that the block
# appears to be a SoS script with one section
if not any([SOS_SECTION_HEADER.match(line) for line in code.splitlines()]):
code = '[interactive_0]\n' + code
script = SoS_Script(content=code, global_sigil=get_default_global_sigil())
else:
script = SoS_Script(filename=script, global_sigil=get_default_global_sigil())
workflow = script.workflow(args.workflow)
executor = Interactive_Executor(workflow, args=workflow_args, config={
'config_file': args.__config__,
'output_dag': args.__dag__,
'report_output': args.__report__})
if args.__dryrun__:
return executor.dryrun(args.__targets__)
else:
return executor.run(args.__targets__)
except Exception:
if args.verbosity and args.verbosity > 2:
sys.stderr.write(get_traceback())
raise
finally:
env.sig_mode = 'default'
env.verbosity = 1
def run(self, targets=None, mode='interactive'):
'''Execute a block of SoS script that is sent by iPython/Jupyer/Spyer
The code can be simple SoS/Python statements, one SoS step, or more
or more SoS workflows with multiple steps. This executor,
1. adds a section header to the script if there is no section head
2. execute the workflow in interactive mode, which is different from
batch mode in a number of ways, which most notably without support
for nested workflow.
3. Optionally execute the workflow in preparation mode for debugging purposes.
'''
# if there is no valid code do nothing
self.set_dict()
# this is the result returned by the workflow, if the
# last stement is an expression.
last_res = None
env.run_mode = mode
# process step of the pipelinp
if isinstance(targets, str):
targets = [targets]
dag = self.initialize_dag(targets=targets)
#
# if targets are specified and there are only signatures for them, we need
# to remove the signature and really generate them
if targets:
for t in targets:
if not FileTarget(t).exists('target'):
FileTarget(t).remove('signature')
#
self.set_dict()
while True:
# find any step that can be executed and run it, and update the DAT
# with status.
runnable = dag.find_executable()
if runnable is None:
# no runnable
#dag.show_nodes()
break
# find the section from runnable
section = self.workflow.section_by_id(runnable._step_uuid)
#
# this is to keep compatibility of dag run with sequential run because
# in sequential run, we evaluate global section of each step in
# order to determine values of options such as skip.
# The consequence is that global definitions are available in
# SoS namespace.
try:
SoS_exec(section.global_def, section.global_sigil)
except Exception as e:
if env.verbosity > 2:
sys.stderr.write(get_traceback())
raise RuntimeError('Failed to execute statements\n"{}"\n{}'.format(
section.global_def, e))
# clear existing keys, otherwise the results from some random result
# might mess with the execution of another step that does not define input
for k in ['__step_input__', '__default_output__', '__step_output__']:
if k in env.sos_dict:
env.sos_dict.pop(k)
# if the step has its own context
env.sos_dict.quick_update(runnable._context)
# execute section with specified input
runnable._status = 'running'
try:
executor = Interactive_Step_Executor(section)
res = executor.run()
for k, v in res.items():
env.sos_dict.set(k, v)
last_res = res['__last_res__']
# set context to the next logic step.
for edge in dag.out_edges(runnable):
node = edge[1]
# if node is the logical next step...
if node._node_index is not None and runnable._node_index is not None:
#and node._node_index == runnable._node_index + 1:
node._context.update(env.sos_dict.clone_selected_vars(
node._context['__signature_vars__'] | node._context['__environ_vars__'] \
| {'_input', '__step_output__', '__default_output__', '__args__'}))
runnable._status = 'completed'
except UnknownTarget as e:
runnable._status = None
target = e.target
if self.resolve_dangling_targets(dag, [target]) == 0:
raise RuntimeError('Failed to resolve {}{}.'
.format(target, dag.steps_depending_on(target, self.workflow)))
# now, there should be no dangling targets, let us connect nodes
# this can be done more efficiently
runnable._depends_targets.append(target)
dag._all_dependent_files[target].append(runnable)
#
dag.build(self.workflow.auxiliary_sections)
#dag.show_nodes()
cycle = dag.circular_dependencies()
if cycle:
raise RuntimeError('Circular dependency detected {}. It is likely a later step produces input of a previous step.'.format(cycle))
except RemovedTarget as e:
runnable._status = None
target = e.target
if not dag.regenerate_target(target):
if self.resolve_dangling_targets(dag, [target]) == 0:
raise RuntimeError('Failed to regenerate or resolve {}{}.'
.format(target, dag.steps_depending_on(target, self.workflow)))
runnable._depends_targets.append(target)
dag._all_dependent_files[target].append(runnable)
dag.build(self.workflow.auxiliary_sections)
#
cycle = dag.circular_dependencies()
if cycle:
raise RuntimeError('Circular dependency detected {}. It is likely a later step produces input of a previous step.'.format(cycle))
self.save_dag(dag)
except UnavailableLock as e:
runnable._status = 'pending'
runnable._signature = (e.output, e.sig_file)
env.logger.info('Waiting on another process for step {}'.format(section.step_name()))
# if the job is failed
except Exception as e:
runnable._status = 'failed'
raise
if self.md5:
self.save_workflow_signature(dag)
env.logger.info('Workflow {} (ID={}) is executed successfully.'.format(self.workflow.name, self.md5))
return last_res