def get_jobs(self, jobs=None, user=None, as_dict=False):
"""
Overrides original method from DirectScheduler in order to list
missing processes as DONE.
"""
job_stats = super().get_jobs(jobs=jobs, user=user, as_dict=as_dict)
found_jobs = []
# Get the list of known jobs
if as_dict:
found_jobs = job_stats.keys()
else:
found_jobs = [j.job_id for j in job_stats]
# Now check if there are any the user requested but were not found
not_found_jobs = list(set(jobs) - set(found_jobs)) if jobs else []
for job_id in not_found_jobs:
job = JobInfo()
job.job_id = job_id
job.job_state = JobState.DONE
# Owner and wallclock time is unknown
if as_dict:
job_stats[job_id] = job
else:
job_stats.append(job)
return job_stats
def test_serialization(self):
"""Test the serialization/deserialization of JobInfo classes."""
from aiida.schedulers.datastructures import JobInfo, JobState
from datetime import datetime
dict_serialized_content = {
'job_id': '12723',
'title': 'some title',
'queue_name': 'some_queue',
'account': 'my_account'
}
to_serialize = {'job_state': (JobState.QUEUED, 'job_state'), 'submission_time': (datetime.now(), 'date')}
job_info = JobInfo()
for key, val in dict_serialized_content.items():
setattr(job_info, key, val)
for key, (val, field_type) in to_serialize.items():
setattr(job_info, key, val)
# Also append to the dictionary for easier comparison later
dict_serialized_content[key] = JobInfo.serialize_field(value=val, field_type=field_type)
self.assertEqual(job_info.get_dict(), dict_serialized_content)
# Full loop via JSON, moving data from job_info to job_info2;
# we check that the content is fully preserved
job_info2 = JobInfo.load_from_serialized(job_info.serialize())
self.assertEqual(job_info2.get_dict(), dict_serialized_content)
# Check that fields are properly re-serialized with the correct type
self.assertEqual(job_info2.job_state, to_serialize['job_state'][0])
# Check that fields are properly re-serialized with the correct type
self.assertEqual(job_info2.submission_time, to_serialize['submission_time'][0])
def get_last_job_info(self):
"""Return the last information asked to the scheduler about the status of the job.
:return: a `JobInfo` object (that closely resembles a dictionary) or None.
"""
from aiida.schedulers.datastructures import JobInfo
last_job_info_serialized = self.get_attribute(
self.SCHEUDLER_LAST_JOB_INFO_KEY, None)
if last_job_info_serialized is not None:
job_info = JobInfo()
job_info.load_from_serialized(last_job_info_serialized)
else:
job_info = None
return job_info
def _parse_joblist_output(self, retval, stdout, stderr):
"""
Parse the queue output string, as returned by executing the
command returned by _get_joblist_command command,
that is here implemented as a list of lines, one for each
job, with _field_separator as separator. The order is described
in the _get_joblist_command function.
Return a list of JobInfo objects, one of each job,
each relevant parameters implemented.
"""
if stderr.strip():
self.logger.warning("Stderr when parsing joblist: {}".format(
stderr.strip()))
job_list = [job.split() for job in stdout.split('\n') if job]
job_infos = []
for job_id, status in job_list:
job = JobInfo()
job.job_id = job_id
job.job_state = _MAP_STATUS_YASCHEDULER[status]
job_infos.append(job)
return job_infos
def _parse_joblist_output(self, retval, stdout, stderr):
"""
Parse the queue output string, as returned by executing the
command returned by _get_joblist_command command,
that is here implemented as a list of lines, one for each
job, with _field_separator as separator. The order is described
in the _get_joblist_command function.
Return a list of JobInfo objects, one of each job,
each relevant parameters implemented.
Note: depending on the scheduler configuration, finished jobs may
either appear here, or not.
This function will only return one element for each job find
in the qstat output; missing jobs (for whatever reason) simply
will not appear here.
"""
num_fields = len(self.fields)
# I don't raise because if I pass a list of jobs,
# I get a non-zero status
# if one of the job is not in the list anymore
# retval should be zero
# if retval != 0:
# self.logger.warning("Error in _parse_joblist_output: retval={}; "
# "stdout={}; stderr={}".format(retval, stdout, stderr))
# issue a warning if there is any stderr output and
# there is no line containing "Invalid job id specified", that happens
# when I ask for specific calculations, and they are all finished
if stderr.strip() and 'Invalid job id specified' not in stderr:
self.logger.warning("Warning in _parse_joblist_output, non-empty stderr='{}'".format(stderr.strip()))
if retval != 0:
raise SchedulerError('Error during squeue parsing (_parse_joblist_output function)')
# will contain raw data parsed from output: only lines with the
# separator, and already split in fields
# I put num_fields, because in this way
# if the symbol _field_separator appears in the title (that is
# the last field), I don't split the title.
# This assumes that _field_separator never
# appears in any previous field.
jobdata_raw = [l.split(_FIELD_SEPARATOR, num_fields) for l in stdout.splitlines() if _FIELD_SEPARATOR in l]
# Create dictionary and parse specific fields
job_list = []
for job in jobdata_raw:
thisjob_dict = {k[1]: v for k, v in zip(self.fields, job)}
this_job = JobInfo()
try:
this_job.job_id = thisjob_dict['job_id']
this_job.annotation = thisjob_dict['annotation']
job_state_raw = thisjob_dict['state_raw']
except KeyError:
# I skip this calculation if I couldn't find this basic info
# (I don't append anything to job_list before continuing)
self.logger.error("Wrong line length in squeue output! '{}'".format(job))
continue
try:
job_state_string = _MAP_STATUS_SLURM[job_state_raw]
except KeyError:
self.logger.warning("Unrecognized job_state '{}' for job "
'id {}'.format(job_state_raw, this_job.job_id))
job_state_string = JobState.UNDETERMINED
# QUEUED_HELD states are not specific states in SLURM;
# they are instead set with state QUEUED, and then the
# annotation tells if the job is held.
# I check for 'Dependency', 'JobHeldUser',
# 'JobHeldAdmin', 'BeginTime'.
# Other states should not bring the job in QUEUED_HELD, I believe
# (the man page of slurm seems to be incomplete, for instance
# JobHeld* are not reported there; I also checked at the source code
# of slurm 2.6 on github (https://github.com/SchedMD/slurm),
# file slurm/src/common/slurm_protocol_defs.c,
# and these seem all the states to be taken into account for the
# QUEUED_HELD status).
# There are actually a few others, like possible
# failures, or partition-related reasons, but for the moment I
# leave them in the QUEUED state.
if (job_state_string == JobState.QUEUED and
this_job.annotation in ['Dependency', 'JobHeldUser', 'JobHeldAdmin', 'BeginTime']):
job_state_string = JobState.QUEUED_HELD
this_job.job_state = job_state_string
####
# Up to here, I just made sure that there were at least three
# fields, to set the most important fields for a job.
# I now check if the length is equal to the number of fields
if len(job) < num_fields:
# I store this job only with the information
# gathered up to now, and continue to the next job
# Also print a warning
self.logger.warning('Wrong line length in squeue output!'
"Skipping optional fields. Line: '{}'"
''.format(jobdata_raw))
# I append this job before continuing
job_list.append(this_job)
continue
# TODO: store executing_host?
this_job.job_owner = thisjob_dict['username']
try:
this_job.num_machines = int(thisjob_dict['number_nodes'])
except ValueError:
self.logger.warning('The number of allocated nodes is not '
'an integer ({}) for job id {}!'.format(thisjob_dict['number_nodes'],
this_job.job_id))
try:
this_job.num_mpiprocs = int(thisjob_dict['number_cpus'])
except ValueError:
self.logger.warning('The number of allocated cores is not '
'an integer ({}) for job id {}!'.format(thisjob_dict['number_cpus'],
this_job.job_id))
# ALLOCATED NODES HERE
# string may be in the format
# nid00[684-685,722-723,748-749,958-959]
# therefore it requires some parsing, that is unnecessary now.
# I just store is as a raw string for the moment, and I leave
# this_job.allocated_machines undefined
if this_job.job_state == JobState.RUNNING:
this_job.allocated_machines_raw = thisjob_dict['allocated_machines']
this_job.queue_name = thisjob_dict['partition']
try:
this_job.requested_wallclock_time_seconds = (self._convert_time(thisjob_dict['time_limit']))
except ValueError:
self.logger.warning('Error parsing the time limit for job id {}'.format(this_job.job_id))
# Only if it is RUNNING; otherwise it is not meaningful,
# and may be not set (in my test, it is set to zero)
if this_job.job_state == JobState.RUNNING:
try:
this_job.wallclock_time_seconds = (self._convert_time(thisjob_dict['time_used']))
except ValueError:
self.logger.warning('Error parsing time_used for job id {}'.format(this_job.job_id))
try:
this_job.dispatch_time = self._parse_time_string(thisjob_dict['dispatch_time'])
except ValueError:
self.logger.warning('Error parsing dispatch_time for job id {}'.format(this_job.job_id))
try:
this_job.submission_time = self._parse_time_string(thisjob_dict['submission_time'])
except ValueError:
self.logger.warning('Error parsing submission_time for job id {}'.format(this_job.job_id))
this_job.title = thisjob_dict['job_name']
# Everything goes here anyway for debugging purposes
this_job.raw_data = job
# Double check of redundant info
# Not really useful now, allocated_machines in this
# version of the plugin is never set
if (this_job.allocated_machines is not None and this_job.num_machines is not None):
if len(this_job.allocated_machines) != this_job.num_machines:
self.logger.error('The length of the list of allocated '
'nodes ({}) is different from the '
'expected number of nodes ({})!'.format(
len(this_job.allocated_machines), this_job.num_machines))
# I append to the list of jobs to return
job_list.append(this_job)
return job_list
def _parse_joblist_output(self, retval, stdout, stderr):
# pylint: disable=too-many-statements,too-many-branches
if retval != 0:
self.logger.error(
f'Error in _parse_joblist_output: retval={retval}; stdout={stdout}; stderr={stderr}'
)
raise SchedulerError(
f'Error during joblist retrieval, retval={retval}')
if stderr.strip():
self.logger.warning(
f'in _parse_joblist_output for {str(self.transport)}: there was some text in stderr: {stderr}'
)
if stdout:
try:
xmldata = xml.dom.minidom.parseString(stdout)
except xml.parsers.expat.ExpatError:
self.logger.error(
f'in sge._parse_joblist_output: xml parsing of stdout failed: {stdout}'
)
raise SchedulerParsingError(
'Error during joblist retrieval, xml parsing of stdout failed'
)
else:
self.logger.error(
f'Error in sge._parse_joblist_output: retval={retval}; stdout={stdout}; stderr={stderr}'
)
raise SchedulerError(
'Error during joblist retrieval, no stdout produced')
try:
first_child = xmldata.firstChild
second_childs = first_child.childNodes
tag_names_sec = [elem.tagName for elem in second_childs \
if elem.nodeType == 1]
if 'queue_info' not in tag_names_sec:
self.logger.error(
f'Error in sge._parse_joblist_output: no queue_info: {stdout}'
)
raise SchedulerError
if 'job_info' not in tag_names_sec:
self.logger.error(
f'Error in sge._parse_joblist_output: no job_info: {stdout}'
)
raise SchedulerError
except SchedulerError:
self.logger.error(
f'Error in sge._parse_joblist_output: stdout={stdout}')
raise SchedulerError('Error during xml processing, of stdout:'
"There is no 'job_info' or no 'queue_info'"
'element or there are no jobs!')
# If something weird happens while firstChild, pop, etc:
except Exception:
self.logger.error(
f'Error in sge._parse_joblist_output: stdout={stdout}')
raise SchedulerError('Error during xml processing, of stdout')
jobs = list(first_child.getElementsByTagName('job_list'))
# jobs = [i for i in jobinfo.getElementsByTagName('job_list')]
# print [i[0].childNodes[0].data for i in job_numbers if i]
joblist = []
for job in jobs:
this_job = JobInfo()
# In case the user needs more information the xml-data for
# each job is stored:
this_job.raw_data = job.toxml()
try:
job_element = job.getElementsByTagName('JB_job_number').pop(0)
element_child = job_element.childNodes.pop(0)
this_job.job_id = str(element_child.data).strip()
if not this_job.job_id:
raise SchedulerError
except SchedulerError:
self.logger.error(
f'Error in sge._parse_joblist_output:no job id is given, stdout={stdout}'
)
raise SchedulerError(
'Error in sge._parse_joblist_output: no job id is given')
except IndexError:
self.logger.error("No 'job_number' given for job index {} in "
'job list, stdout={}'.format(jobs.index(job) \
, stdout))
raise IndexError(
'Error in sge._parse_joblist_output: no job id is given')
try:
job_element = job.getElementsByTagName('state').pop(0)
element_child = job_element.childNodes.pop(0)
job_state_string = str(element_child.data).strip()
try:
this_job.job_state = _MAP_STATUS_SGE[job_state_string]
except KeyError:
self.logger.warning("Unrecognized job_state '{}' for job "
'id {}'.format(job_state_string,
this_job.job_id))
this_job.job_state = JobState.UNDETERMINED
except IndexError:
self.logger.warning("No 'job_state' field for job id {} in"
'stdout={}'.format(this_job.job_id,
stdout))
this_job.job_state = JobState.UNDETERMINED
try:
job_element = job.getElementsByTagName('JB_owner').pop(0)
element_child = job_element.childNodes.pop(0)
this_job.job_owner = str(element_child.data).strip()
except IndexError:
self.logger.warning(
f"No 'job_owner' field for job id {this_job.job_id}")
try:
job_element = job.getElementsByTagName('JB_name').pop(0)
element_child = job_element.childNodes.pop(0)
this_job.title = str(element_child.data).strip()
except IndexError:
self.logger.warning(
f"No 'title' field for job id {this_job.job_id}")
try:
job_element = job.getElementsByTagName('queue_name').pop(0)
element_child = job_element.childNodes.pop(0)
this_job.queue_name = str(element_child.data).strip()
except IndexError:
if this_job.job_state == JobState.RUNNING:
self.logger.warning(
f"No 'queue_name' field for job id {this_job.job_id}")
try:
job_element = job.getElementsByTagName(
'JB_submission_time').pop(0)
element_child = job_element.childNodes.pop(0)
time_string = str(element_child.data).strip()
try:
this_job.submission_time = self._parse_time_string(
time_string)
except ValueError:
self.logger.warning(
f"Error parsing 'JB_submission_time' for job id {this_job.job_id} ('{time_string}')"
)
except IndexError:
try:
job_element = job.getElementsByTagName(
'JAT_start_time').pop(0)
element_child = job_element.childNodes.pop(0)
time_string = str(element_child.data).strip()
try:
this_job.dispatch_time = self._parse_time_string(
time_string)
except ValueError:
self.logger.warning(
f"Error parsing 'JAT_start_time'for job id {this_job.job_id} ('{time_string}')"
)
except IndexError:
self.logger.warning("No 'JB_submission_time' and no "
"'JAT_start_time' field for job "
'id {}'.format(this_job.job_id))
# There is also cpu_usage, mem_usage, io_usage information available:
if this_job.job_state == JobState.RUNNING:
try:
job_element = job.getElementsByTagName('slots').pop(0)
element_child = job_element.childNodes.pop(0)
this_job.num_mpiprocs = str(element_child.data).strip()
except IndexError:
self.logger.warning(
f"No 'slots' field for job id {this_job.job_id}")
joblist.append(this_job)
# self.logger.debug("joblist final: {}".format(joblist))
return joblist
def _parse_joblist_output(self, retval, stdout, stderr):
"""
Parse the queue output string, as returned by executing the
command returned by _get_joblist_command command,
that is here implemented as a list of lines, one for each
job, with _field_separator as separator. The order is described
in the _get_joblist_command function.
Return a list of JobInfo objects, one of each job,
each relevant parameters implemented.
Note: depending on the scheduler configuration, finished jobs may
either appear here, or not.
This function will only return one element for each job find
in the qstat output; missing jobs (for whatever reason) simply
will not appear here.
"""
# pylint: disable=too-many-locals,too-many-statements,too-many-branches
num_fields = len(self._joblist_fields)
if retval != 0:
self.logger.warning('Error in _parse_joblist_output: retval={}; '
'stdout={}; stderr={}'.format(
retval, stdout, stderr))
raise SchedulerError('Error during parsing joblist output, '
'retval={}\n'
'stdout={}\nstderr={}'.format(
retval, stdout, stderr))
# will contain raw data parsed from output: only lines with the
# separator, and already split in fields
# I put num_fields, because in this way
# if the symbol _field_separator appears in the title (that is
# the last field), I don't split the title.
# This assumes that _field_separator never
# appears in any previous field.
jobdata_raw = [
l.split(_FIELD_SEPARATOR, num_fields) for l in stdout.splitlines()
if _FIELD_SEPARATOR in l
]
# Create dictionary and parse specific fields
job_list = []
for job in jobdata_raw:
# Each job should have all fields.
if len(job) != num_fields:
# I skip this calculation
# (I don't append anything to job_list before continuing)
self.logger.error(
"Wrong line length in squeue output! '{}'".format(job))
continue
this_job = JobInfo()
this_job.job_id = job[0]
this_job.annotation = job[2]
job_state_raw = job[1]
try:
job_state_string = _MAP_STATUS_LSF[job_state_raw]
except KeyError:
self.logger.warning("Unrecognized job_state '{}' for job "
'id {}'.format(job_state_raw,
this_job.job_id))
job_state_string = JobState.UNDETERMINED
this_job.job_state = job_state_string
# I get the remaining fields
# The first three were already obtained
# I know that the length is exactly num_fields because
# I used split(_field_separator, num_fields) before
# when creting 'job'
# (_, _, _, executing_host, username, number_nodes,
# number_cpus, allocated_machines, partition,
# time_limit, time_used, dispatch_time, job_name) = job
(_, _, _, _, username, number_nodes, number_cpus,
allocated_machines, partition, finish_time, start_time,
percent_complete, submission_time, job_name) = job
this_job.job_owner = username
try:
this_job.num_machines = int(number_nodes)
except ValueError:
self.logger.warning('The number of allocated nodes is not '
'an integer ({}) for job id {}!'.format(
number_nodes, this_job.job_id))
try:
this_job.num_mpiprocs = int(number_cpus)
except ValueError:
self.logger.warning('The number of allocated cores is not '
'an integer ({}) for job id {}!'.format(
number_cpus, this_job.job_id))
# ALLOCATED NODES HERE
# string may be in the format
# nid00[684-685,722-723,748-749,958-959]
# therefore it requires some parsing, that is unnecessary now.
# I just store is as a raw string for the moment, and I leave
# this_job.allocated_machines undefined
if this_job.job_state == JobState.RUNNING:
this_job.allocated_machines_raw = allocated_machines
this_job.queue_name = partition
psd_finish_time = self._parse_time_string(finish_time,
fmt='%b %d %H:%M')
psd_start_time = self._parse_time_string(start_time,
fmt='%b %d %H:%M')
psd_submission_time = self._parse_time_string(submission_time,
fmt='%b %d %H:%M')
# Now get the time in seconds which has been used
# Only if it is RUNNING; otherwise it is not meaningful,
# and may be not set (in my test, it is set to zero)
if this_job.job_state == JobState.RUNNING:
try:
requested_walltime = psd_finish_time - psd_start_time
# fix of a weird bug. Since the year is not parsed, it is assumed
# to always be 1900. Therefore, job submitted
# in december and finishing in january would produce negative time differences
if requested_walltime.total_seconds() < 0:
import datetime
old_month = psd_finish_time.month
old_day = psd_finish_time.day
old_hour = psd_finish_time.hour
old_minute = psd_finish_time.minute
new_year = psd_start_time.year + 1
# note: we assume that no job will last more than 1 year...
psd_finish_time = datetime.datetime(year=new_year,
month=old_month,
day=old_day,
hour=old_hour,
minute=old_minute)
requested_walltime = psd_finish_time - psd_start_time
this_job.requested_wallclock_time_seconds = requested_walltime.total_seconds(
) # pylint: disable=invalid-name
except (TypeError, ValueError):
self.logger.warning(
'Error parsing the time limit for job id {}'.format(
this_job.job_id))
try:
psd_percent_complete = float(
percent_complete.strip(' L').strip('%'))
this_job.wallclock_time_seconds = requested_walltime.total_seconds(
) * psd_percent_complete / 100.
except ValueError:
self.logger.warning(
'Error parsing the time used for job id {}'.format(
this_job.job_id))
try:
this_job.submission_time = psd_submission_time
except ValueError:
self.logger.warning(
'Error parsing submission time for job id {}'.format(
this_job.job_id))
this_job.title = job_name
# Everything goes here anyway for debugging purposes
this_job.raw_data = job
# Double check of redundant info
# Not really useful now, allocated_machines in this
# version of the plugin is never set
if (this_job.allocated_machines is not None
and this_job.num_machines is not None):
if len(this_job.allocated_machines) != this_job.num_machines:
self.logger.error('The length of the list of allocated '
'nodes ({}) is different from the '
'expected number of nodes ({})!'.format(
len(this_job.allocated_machines),
this_job.num_machines))
# I append to the list of jobs to return
job_list.append(this_job)
return job_list
def _parse_joblist_output(self, retval, stdout, stderr):
"""
Parse the queue output string, as returned by executing the
command returned by _get_joblist_command command (qstat -f).
Return a list of JobInfo objects, one of each job,
each relevant parameters implemented.
Note: depending on the scheduler configuration, finished jobs may
either appear here, or not.
This function will only return one element for each job find
in the qstat output; missing jobs (for whatever reason) simply
will not appear here.
"""
# I don't raise because if I pass a list of jobs, I get a non-zero status
# if one of the job is not in the list anymore
# retval should be zero
# if retval != 0:
# _LOGGER.warning("Error in _parse_joblist_output: retval={}; "
# "stdout={}; stderr={}".format(retval, stdout, stderr))
# issue a warning if there is any stderr output
# but I strip lines containing "Unknown Job Id", that happens
# also when I ask for a calculation that has finished
#
# I also strip for "Job has finished" because this happens for
# those schedulers configured to leave the job in the output
# of qstat for some time after job completion.
filtered_stderr = '\n'.join(
l for l in stderr.split('\n')
if 'Unknown Job Id' not in l and 'Job has finished' not in l)
if filtered_stderr.strip():
_LOGGER.warning('Warning in _parse_joblist_output, non-empty '
"(filtered) stderr='{}'".format(filtered_stderr))
if retval != 0:
raise SchedulerError('Error during qstat parsing, retval={}\n'
'stdout={}\nstderr={}'.format(
retval, stdout, stderr))
jobdata_raw = [] # will contain raw data parsed from qstat output
# Get raw data and split in lines
for line_num, line in enumerate(stdout.split('\n'), start=1):
# Each new job stanza starts with the string 'Job Id:': I
# create a new item in the jobdata_raw list
if line.startswith('Job Id:'):
jobdata_raw.append({
'id': line.split(':', 1)[1].strip(),
'lines': [],
'warning_lines_idx': []
})
# warning_lines_idx: lines that do not start either with
# tab or space
else:
if line.strip():
# This is a non-empty line, therefore it is an attribute
# of the last job found
if not jobdata_raw:
# The list is still empty! (This means that I found a
# non-empty line, before finding the first 'Job Id:'
# string: it is an error. However this may happen
# only before the first job.
raise SchedulerParsingError(
'I did not find the header for the first job')
# _LOGGER.warning("I found some text before the "
# "first job: {}".format(l))
else:
if line.startswith(' '):
# If it starts with a space, it is a new field
jobdata_raw[-1]['lines'].append(line)
elif line.startswith('\t'):
# If a line starts with a TAB,
# I append to the previous string
# stripping the TAB
if not jobdata_raw[-1]['lines']:
raise SchedulerParsingError(
'Line {} is the first line of the job, but it '
'starts with a TAB! ({})'.format(
line_num, line))
jobdata_raw[-1]['lines'][-1] += line[1:]
else:
# raise SchedulerParsingError(
# "Wrong starting character at line {}! ({})"
# "".format(line_num, l))
## For some reasons, the output of 'comment' and
## 'Variable_List', for instance, can have
## newlines if they are included... # I do a
## workaround
jobdata_raw[-1]['lines'][-1] += '\n{}'.format(line)
jobdata_raw[-1]['warning_lines_idx'].append(
len(jobdata_raw[-1]['lines']) - 1)
# Create dictionary and parse specific fields
job_list = []
for job in jobdata_raw:
this_job = JobInfo()
this_job.job_id = job['id']
lines_without_equals_sign = [
i for i in job['lines'] if '=' not in i
]
# There are lines without equals sign: this is bad
if lines_without_equals_sign:
# Should I only warn?
_LOGGER.error('There are lines without equals sign! {}'.format(
lines_without_equals_sign))
raise SchedulerParsingError(
'There are lines without equals sign.')
raw_data = {
i.split('=', 1)[0].strip().lower(): i.split('=',
1)[1].lstrip()
for i in job['lines'] if '=' in i
}
## I ignore the errors for the time being - this seems to be
## a problem if there are \n in the content of some variables?
## I consider this a workaround...
# for line_with_warning in set(job['warning_lines_idx']):
# if job['lines'][line_with_warning].split(
# '=',1)[0].strip().lower() != "comment":
# raise SchedulerParsingError(
# "Wrong starting character in one of the lines "
# "of job {}, and it's not a comment! ({})"
# "".format(this_job.job_id,
# job['lines'][line_with_warning]))
problematic_fields = []
for line_with_warning in set(job['warning_lines_idx']):
problematic_fields.append(
job['lines'][line_with_warning].split(
'=', 1)[0].strip().lower())
if problematic_fields:
# These are the fields that contain unexpected newlines
raw_data['warning_fields_with_newlines'] = problematic_fields
# I believe that exit_status and terminating_signal cannot be
# retrieved from the qstat -f output.
# I wrap calls in try-except clauses to avoid errors if a field
# is missing
try:
this_job.title = raw_data['job_name']
except KeyError:
_LOGGER.debug("No 'job_name' field for job id {}".format(
this_job.job_id))
try:
this_job.annotation = raw_data['comment']
except KeyError:
# Many jobs do not have a comment; I do not complain about it.
pass
# _LOGGER.debug("No 'comment' field for job id {}".format(
# this_job.job_id))
try:
job_state_string = raw_data['job_state']
try:
this_job.job_state = self._map_status[job_state_string]
except KeyError:
_LOGGER.warning("Unrecognized job_state '{}' for job "
'id {}'.format(job_state_string,
this_job.job_id))
this_job.job_state = JobState.UNDETERMINED
except KeyError:
_LOGGER.debug("No 'job_state' field for job id {}".format(
this_job.job_id))
this_job.job_state = JobState.UNDETERMINED
try:
this_job.job_substate = raw_data['substate']
except KeyError:
_LOGGER.debug("No 'substate' field for job id {}".format(
this_job.job_id))
try:
exec_hosts = raw_data['exec_host'].split('+')
except KeyError:
# No exec_host information found (it may be ok, if the job
# is not running)
pass
else:
# parse each host; syntax, from the man page:
# hosta/J1+hostb/J2*P+...
# where J1 and J2 are an index of the job
# on the named host and P is the number of
# processors allocated from that host to this job.
# P does not appear if it is 1.
try:
exec_host_list = []
for exec_host in exec_hosts:
node = MachineInfo()
node.name, data = exec_host.split('/')
data = data.split('*')
if len(data) == 1:
node.jobIndex = int(data[0])
node.num_cpus = 1
elif len(data) == 2:
node.jobIndex = int(data[0])
node.num_cpus = int(data[1])
else:
raise ValueError(
'Wrong number of pieces: {} '
'instead of 1 or 2 in exec_hosts: '
'{}'.format(len(data), exec_hosts))
exec_host_list.append(node)
this_job.allocated_machines = exec_host_list
except Exception as exc:
_LOGGER.debug('Problem parsing the node names, I '
'got Exception {} with message {}; '
'exec_hosts was {}'.format(
str(type(exc)), exc, exec_hosts))
try:
# I strip the part after the @: is this always ok?
this_job.job_owner = raw_data['job_owner'].split('@')[0]
except KeyError:
_LOGGER.debug("No 'job_owner' field for job id {}".format(
this_job.job_id))
try:
this_job.num_cpus = int(raw_data['resource_list.ncpus'])
# TODO: understand if this is the correct field also for
# multithreaded (OpenMP) jobs.
except KeyError:
_LOGGER.debug(
"No 'resource_list.ncpus' field for job id {}".format(
this_job.job_id))
except ValueError:
_LOGGER.warning("'resource_list.ncpus' is not an integer "
'({}) for job id {}!'.format(
raw_data['resource_list.ncpus'],
this_job.job_id))
try:
this_job.num_mpiprocs = int(raw_data['resource_list.mpiprocs'])
# TODO: understand if this is the correct field also for
# multithreaded (OpenMP) jobs.
except KeyError:
_LOGGER.debug(
"No 'resource_list.mpiprocs' field for job id {}".format(
this_job.job_id))
except ValueError:
_LOGGER.warning("'resource_list.mpiprocs' is not an integer "
'({}) for job id {}!'.format(
raw_data['resource_list.mpiprocs'],
this_job.job_id))
try:
this_job.num_machines = int(raw_data['resource_list.nodect'])
except KeyError:
_LOGGER.debug(
"No 'resource_list.nodect' field for job id {}".format(
this_job.job_id))
except ValueError:
_LOGGER.warning("'resource_list.nodect' is not an integer "
'({}) for job id {}!'.format(
raw_data['resource_list.nodect'],
this_job.job_id))
# Double check of redundant info
if (this_job.allocated_machines is not None
and this_job.num_machines is not None):
if len(
set(machine.name for machine in this_job.
allocated_machines)) != this_job.num_machines:
_LOGGER.error('The length of the list of allocated '
'nodes ({}) is different from the '
'expected number of nodes ({})!'.format(
len(this_job.allocated_machines),
this_job.num_machines))
try:
this_job.queue_name = raw_data['queue']
except KeyError:
_LOGGER.debug("No 'queue' field for job id {}".format(
this_job.job_id))
try:
this_job.RequestedWallclockTime = (self._convert_time(
raw_data['resource_list.walltime']))
except KeyError:
_LOGGER.debug(
"No 'resource_list.walltime' field for job id {}".format(
this_job.job_id))
except ValueError:
_LOGGER.warning(
"Error parsing 'resource_list.walltime' for job id {}".
format(this_job.job_id))
try:
this_job.wallclock_time_seconds = (self._convert_time(
raw_data['resources_used.walltime']))
except KeyError:
# May not have started yet
pass
except ValueError:
_LOGGER.warning(
"Error parsing 'resources_used.walltime' for job id {}".
format(this_job.job_id))
try:
this_job.cpu_time = (self._convert_time(
raw_data['resources_used.cput']))
except KeyError:
# May not have started yet
pass
except ValueError:
_LOGGER.warning(
"Error parsing 'resources_used.cput' for job id {}".format(
this_job.job_id))
#
# ctime: The time that the job was created
# mtime: The time that the job was last modified, changed state,
# or changed locations.
# qtime: The time that the job entered the current queue
# stime: The time when the job started execution.
# etime: The time that the job became eligible to run, i.e. in a
# queued state while residing in an execution queue.
try:
this_job.submission_time = self._parse_time_string(
raw_data['ctime'])
except KeyError:
_LOGGER.debug("No 'ctime' field for job id {}".format(
this_job.job_id))
except ValueError:
_LOGGER.warning("Error parsing 'ctime' for job id {}".format(
this_job.job_id))
try:
this_job.dispatch_time = self._parse_time_string(
raw_data['stime'])
except KeyError:
# The job may not have been started yet
pass
except ValueError:
_LOGGER.warning("Error parsing 'stime' for job id {}".format(
this_job.job_id))
# TODO: see if we want to set also finish_time for finished jobs,
# if there are any
# Everything goes here anyway for debugging purposes
this_job.raw_data = raw_data
# I append to the list of jobs to return
job_list.append(this_job)
return job_list
def _parse_joblist_output(self, retval, stdout, stderr):
"""
Parse the queue output string, as returned by executing the
command returned by _get_joblist_command command (qstat -f).
Return a list of JobInfo objects, one of each job,
each relevant parameters implemented.
.. note:: depending on the scheduler configuration, finished jobs
may either appear here, or not.
This function will only return one element for each job find
in the qstat output; missing jobs (for whatever reason) simply
will not appear here.
"""
import re
filtered_stderr = '\n'.join(l for l in stderr.split('\n'))
if filtered_stderr.strip():
self.logger.warning(
'Warning in _parse_joblist_output, non-empty '
"(filtered) stderr='{}'".format(filtered_stderr)
)
if retval != 0:
raise SchedulerError('Error during direct execution parsing (_parse_joblist_output function)')
# Create dictionary and parse specific fields
job_list = []
for line in stdout.split('\n'):
if re.search(r'^\s*PID', line) or line == '':
# Skip the header if present
continue
line = re.sub(r'^\s+', '', line)
job = re.split(r'\s+', line)
this_job = JobInfo()
this_job.job_id = job[0]
if len(job) < 3:
raise SchedulerError(
'Unexpected output from the scheduler, '
"not enough fields in line '{}'".format(line)
)
try:
job_state_string = job[1][0] # I just check the first character
except IndexError:
self.logger.debug("No 'job_state' field for job id {}".format(this_job.job_id))
this_job.job_state = JobState.UNDETERMINED
else:
try:
this_job.job_state = \
_MAP_STATUS_PS[job_state_string]
except KeyError:
self.logger.warning(
"Unrecognized job_state '{}' for job "
'id {}'.format(job_state_string, this_job.job_id)
)
this_job.job_state = JobState.UNDETERMINED
try:
# I strip the part after the @: is this always ok?
this_job.job_owner = job[2]
except KeyError:
self.logger.debug("No 'job_owner' field for job id {}".format(this_job.job_id))
try:
this_job.wallclock_time_seconds = self._convert_time(job[3])
except KeyError:
# May not have started yet
pass
except ValueError:
self.logger.warning("Error parsing 'resources_used.walltime' for job id {}".format(this_job.job_id))
# I append to the list of jobs to return
job_list.append(this_job)
return job_list
def get_jobs(self, jobs=None, user=None, as_dict=False):
"""
Return the list of currently active jobs
"""
computer_id = self.transport._machine # Host name is used as the identifier
lpad = self.lpad
query = {
"spec._aiida_job_info.computer_id":
computer_id, # Limit to this machine
# Ignore completed and archived jobs
"state": {
"$not": {
"$in": ["COMPLETED", "ARCHIVED"]
}
}
}
# Limit to the specific fw_ids
if jobs:
# Convert to integer keys
jobs = [int(job_id) for job_id in jobs]
query['fw_id'] = {'$in': jobs}
fw_ids = lpad.get_fw_ids(query)
joblist = []
for fid in fw_ids:
# Get the information of the fireworks in the dict format
# this is more robust than getting Fireworks objects
try:
fw_dict = lpad.get_fw_dict_by_id(fid)
except ValueError:
raise SchedulerError(f"No FW found for id: {fid}")
spec = fw_dict.get("spec", {})
this_job = JobInfo()
this_job.job_id = str(fid)
try:
this_job.job_state = _MAP_STATUS_FW[fw_dict['state']]
except IndexError:
this_job.job_state = JobState.UNDETERMINED
this_job.title = fw_dict.get('name')
# Category or categories are mapped to queue_name attribute
category = spec.get('category')
if isinstance(category, str):
this_job.queue_name = category
elif isinstance(category, (tuple, list)):
this_job.queue_name = ":".join(category)
# The created_on is mapped to the submission time
try:
this_job.submission_time = datetime.strptime(
fw_dict['created_on'], "%Y-%m-%dT%H:%M:%S.%f")
except ValueError:
pass
# NOTE: add information about the dispatch time by looking into the launches
joblist.append(this_job)
if as_dict:
jobdict = {job.job_id: job for job in joblist}
if None in jobdict:
raise SchedulerError('Found at least one job without jobid')
return jobdict
return joblist