def attach(args):
def _find_dead_pids_host(host, pids):
dead_pids = []
_establish_default_terminal()
_ensure_scheduler_exists('jobs')
for pid in pids:
if not default_terminal.is_pid_alive(pid):
host_path = '.'.join(hops.list_hops())
if host:
host_path += '.' + host
dead_pids.append('{host}:{pid}'.format(host=host_path
or 'localhost',
pid=pid))
return dead_pids
def _find_dead_pids(pid_dict):
# Check the status of all provided PIDs
dead_pids = []
for host, pids in pid_dict.iteriterms():
# Establish a connection per each process
dead_pids.extend(_find_dead_pids_host(host, pids))
return dead_pids
def _parse_group_pids(expr):
pid_dict = {}
for app_instance in re.finditer('((?:(\w+):)?(\d+))', expr):
host = app_instance.group(2)
pid = int(app_instance.group(3))
if pid_dict.has_key(host):
pid_dict[host] += [pid]
else:
pid_dict[host] = [pid]
args_string = ' '.join(args)
# Verify command syntax
if len(args) < 1 or not re.match('(?:(?:\w+:)?\d+|\s)+', args_string):
raise errors.BadArgsError(
'attach', 'attach [<host>:]<pid>[ [<host>:]<pid> [...]]')
# Group by host
pid_dict = _parse_group_pids(args_string)
# Check the status of all provided PIDs
dead_pids = _find_dead_pids(pid_dict)
# Stop if all processes are alive
if len(dead_pids) != 0:
raise errors.CommandFailedError(
'attach',
'Invalid PIDs provided: {0}'.format(' ,'.join(dead_pids)))
# Launch GDB and attach to PIDs
session_manager, msgs = _attach_pids(pid_dict)
# Initialize the debugging session
return debug_session.init_debugging_mode(session_manager, msgs)
def end_sessions(args):
# Verify command syntax
if len(args) != 0:
raise errors.BadArgsError(
'ls', 'This command does not accept arguments'
)
_kill_all()
return modes.offline, None
def _ensure_valid_sessions_selected(cmd):
'''Verifies if all selected sessions are still alive'''
non_existing_sessions = [
id for id in selected_sessions if not session_manager.exists(id)
]
if non_existing_sessions:
raise errors.BadArgsError(
cmd, 'Cannot proceed. Dead session(s): {0}.'.
format(', '.format(non_existing_sessions))
)
def quit(args):
'''Exit scimitar'''
# Verify command syntax
if len(args) != 0:
raise errors.BadArgsError(
'ls', 'This command does not accept arguments'
)
_kill_all()
return modes.quit, None
def list_jobs(args):
# Verify command syntax
if args:
raise errors.BadArgsError('jobs',
'This command does not accept arguments.')
_establish_default_terminal()
_ensure_scheduler_exists('jobs')
items = csssi.ls_user_jobs(default_terminal)
jobs_str = '\t'.join(items)
return modes.offline, jobs_str
def list_sessions(args):
'''Lists all active sessions.'''
# Verify command syntax
if len(args) != 0:
raise errors.BadArgsError(
'ls', 'This command does not accept arguments'
)
pretty_session_list = _ls_out()
if not pretty_session_list:
return modes.debugging, 'ls: Empty. No sessions are alive.'
return modes.debugging, 'Sessions:\n' + pretty_session_list
def message_history(args):
'''Prints the selected sessions' history records at index args[0]'''
_ensure_sessions_selected('history')
index = -1
if len(args) == 1:
try:
index = -int(args[0])
except ValueError:
raise errors.BadArgsError(
'history', 'history[ <index>]'
)
if index > history_length:
raise errors.BadArgsError(
'history', 'Selected record does not exist in the history'
)
results = []
for tag in selected_sessions:
# Output header
results.append('~~~ Scimitar - Session: {} ~~~'.format(tag))
try:
ind_rec, cout, tout, lout = sessions_history[tag][index]
if ind_rec:
# Check the type of indicator
if ind_rec[0] == mi.indicator_error:
results.append(format_error(ind_rec[1]))
elif ind_rec[0] == mi.indicator_exit:
session_manager.remove(tag)
sessions_history.remove(tag)
results.append(format_error('Session {} died.', tag))
else:
results.append(cout)
else:
results.append(''.join([cout, tout, lout]))
except IndexError:
results.append(format_error('Scimitar: Record does not exist'))
return modes.debugging, '\n'.join(results)
def select_session(args):
'''Mark the provided sessions as selected'''
# Verify command syntax
if not args:
raise errors.BadArgsError(
'select', 'select all | none | <session id>[ <session id>[ ...]]'
)
if args:
if args[0] == 'all':
global selected_sessions
selected_sessions = sorted(session_manager.list_session_tags())
return modes.debugging, 'Selected session(s) #{0}\n{1}'.format(
', '.join(selected_sessions), _ls_out()
)
elif args[0] == 'none':
global selected_sessions
selected_sessions = []
return modes.debugging, 'No sessions selected'.format(
', '.join(selected_sessions), _ls_out()
)
non_existing_sessions = [
id for id in args if not session_manager.exists(id)
]
if non_existing_sessions:
raise errors.BadArgsError(
'select', 'Session(s) {0} do not exist.'.
format(', '.join(non_existing_sessions))
)
global selected_sessions
selected_sessions = args
return modes.debugging, 'Selected session(s) #{0}\n{1}'.format(
', '.join(selected_sessions), _ls_out()
)
def unhop(args):
# Verify command syntax
if len(args) > 1:
raise errors.BadArgsError('unhop',
'unhop[ <number of hops to remove>].')
n_hops_to_remove = 1
if len(args) == 1:
try:
n_hops_to_remove = int(args[0])
except ValueError:
raise errors.BadArgsError('unhop',
'unhop[ <number of hops to remove>].')
try:
for _ in range(n_hops_to_remove):
hops.remove_last()
except console.NoHopsError:
raise errors.CommandFailedError(
'unhop', 'No more hops currently exist. Nothing can be removed')
_establish_default_terminal(reestablish=True)
return modes.offline, None
def job(args):
# Verify command syntax
if len(args) > 2:
raise errors.BadArgsError('job', 'job[ <job_id> | <auto>[ <app>]]')
# Basic checks
_establish_default_terminal()
_ensure_scheduler_exists('job')
# If job name is not provided then probably only one job is active
if len(args) == 0 or args[0] == 'auto':
try:
job_id = csssi.detect_active_job(default_terminal)
# No active jobs
except csssi.NoActiveJobError:
raise errors.CommandFailedError(
'job', 'No active user jobs found. Cannot proceed.')
# More than one job is active
except csssi.MoreThanOneActiveJobError:
raise errors.CommandFailedError(
'job', 'Found more than one job. Cannot proceed.')
# Job Id provided
else:
job_id = args[0]
# In case app name is provided
app = args[1] if len(args) == 2 else None
# Hosts and PIDs will be stored here
pid_dict = None
# List job nodes and processes
try:
pid_dict = csssi.ls_job_pids(default_terminal, job_id, app)
# Job does not exist
except csssi.InvalidJobError:
raise errors.CommandFailedError(
'job', '{0} does not seem to be a valid job id'.format(job_id))
# Cannot determine app
except csssi.NoRunningAppFoundError:
raise errors.CommandFailedError(
'job',
'No application name lookup pattern provided and automatic MPI application detection did not succeed. Ensure job {0} is actually running a distributed application and provide the application name'
.format(job_id))
# Launch GDB and attach to PIDs
session_manager, msgs = _attach_pids(pid_dict)
# Initialize the debugging session
return debug_session.init_debugging_mode(session_manager, msgs)
