def submit_jobs():
"""
Submit all jobs in the TOSUBMIT state.
"""
from aiida.orm import JobCalculation, Computer, User
from aiida.utils.logger import get_dblogger_extra
from aiida.backends.utils import get_authinfo
computers_users_to_check = list(
JobCalculation._get_all_with_state(state=calc_states.TOSUBMIT,
only_computer_user_pairs=True,
only_enabled=True))
for computer, aiidauser in computers_users_to_check:
#~ user = User.search_for_users(id=dbuser_id)
#~ computer = Computer.get(dbcomputer_id)
execlogger.debug("({},{}) pair to submit".format(
aiidauser.email, computer.name))
try:
try:
authinfo = get_authinfo(computer.dbcomputer, aiidauser._dbuser)
except AuthenticationError:
# TODO!!
# Put each calculation in the SUBMISSIONFAILED state because
# I do not have AuthInfo to submit them
calcs_to_inquire = JobCalculation._get_all_with_state(
state=calc_states.TOSUBMIT,
computer=computer,
user=aiidauser)
for calc in calcs_to_inquire:
try:
calc._set_state(calc_states.SUBMISSIONFAILED)
except ModificationNotAllowed:
# Someone already set it, just skip
pass
logger_extra = get_dblogger_extra(calc)
execlogger.error("Submission of calc {} failed, "
"computer pk= {} ({}) is not configured "
"for aiidauser {}".format(
calc.pk, computer.pk,
computer.get_name(), aiidauser.email),
extra=logger_extra)
# Go to the next (dbcomputer,aiidauser) pair
continue
submitted_calcs = submit_jobs_with_authinfo(authinfo)
except Exception as e:
import traceback
msg = ("Error while submitting jobs "
"for aiidauser={} on computer={}, "
"error type is {}, traceback: {}".format(
aiidauser.email, computer.name, e.__class__.__name__,
traceback.format_exc()))
print msg
execlogger.error(msg)
# Continue with next computer
continue
def update_jobs():
"""
calls an update for each set of pairs (machine, aiidauser)
"""
from aiida.orm import JobCalculation, Computer, User
from aiida.backends.utils import get_authinfo
# I create a unique set of pairs (computer, aiidauser)
computers_users_to_check = list(
JobCalculation._get_all_with_state(state=calc_states.WITHSCHEDULER,
only_computer_user_pairs=True,
only_enabled=True))
for computer, aiidauser in computers_users_to_check:
execlogger.debug("({},{}) pair to check".format(
aiidauser.email, computer.name))
try:
authinfo = get_authinfo(computer.dbcomputer, aiidauser._dbuser)
computed_calcs = update_running_calcs_status(authinfo)
except Exception as e:
msg = ("Error while updating calculation status "
"for aiidauser={} on computer={}, "
"error type is {}, error message: {}".format(
aiidauser.email, computer.name, e.__class__.__name__,
e.message))
execlogger.error(msg)
# Continue with next computer
continue
def retrieve_jobs():
from aiida.orm import JobCalculation, Computer
from aiida.backends.utils import get_authinfo
# I create a unique set of pairs (computer, aiidauser)
computers_users_to_check = list(
JobCalculation._get_all_with_state(state=calc_states.COMPUTED,
only_computer_user_pairs=True,
only_enabled=True))
for computer, aiidauser in computers_users_to_check:
execlogger.debug("({},{}) pair to check".format(
aiidauser.email, computer.name))
try:
authinfo = get_authinfo(computer.dbcomputer, aiidauser._dbuser)
retrieve_computed_for_authinfo(authinfo)
except Exception as e:
msg = ("Error while retrieving calculation status for "
"aiidauser={} on computer={}, "
"error type is {}, error message: {}".format(
aiidauser.email, computer.name, e.__class__.__name__,
e.message))
execlogger.error(msg)
# Continue with next computer
continue
def retrieve_computed_for_authinfo(authinfo):
from aiida.orm import JobCalculation
from aiida.common.folders import SandboxFolder
from aiida.orm.data.folder import FolderData
from aiida.utils.logger import get_dblogger_extra
from aiida.orm import DataFactory
import os
if not authinfo.enabled:
return
calcs_to_retrieve = list(
JobCalculation._get_all_with_state(state=calc_states.COMPUTED,
computer=authinfo.dbcomputer,
user=authinfo.aiidauser))
retrieved = []
# I avoid to open an ssh connection if there are no
# calcs with state not COMPUTED
if len(calcs_to_retrieve):
# Open connection
with authinfo.get_transport() as t:
for calc in calcs_to_retrieve:
logger_extra = get_dblogger_extra(calc)
t._set_logger_extra(logger_extra)
try:
calc._set_state(calc_states.RETRIEVING)
except ModificationNotAllowed:
# Someone else has already started to retrieve it,
# just log and continue
execlogger.debug("Attempting to retrieve more than once "
"calculation {}: skipping!".format(
calc.pk),
extra=logger_extra)
continue # with the next calculation to retrieve
try:
execlogger.debug("Retrieving calc {}".format(calc.pk),
extra=logger_extra)
workdir = calc._get_remote_workdir()
retrieve_list = calc._get_retrieve_list()
retrieve_singlefile_list = calc._get_retrieve_singlefile_list(
)
execlogger.debug("[retrieval of calc {}] "
"chdir {}".format(calc.pk, workdir),
extra=logger_extra)
t.chdir(workdir)
retrieved_files = FolderData()
retrieved_files.add_link_from(
calc,
label=calc._get_linkname_retrieved(),
link_type=LinkType.CREATE)
# First, retrieve the files of folderdata
with SandboxFolder() as folder:
for item in retrieve_list:
# I have two possibilities:
# * item is a string
# * or is a list
# then I have other two possibilities:
# * there are file patterns
# * or not
# First decide the name of the files
if isinstance(item, list):
tmp_rname, tmp_lname, depth = item
# if there are more than one file I do something differently
if t.has_magic(tmp_rname):
remote_names = t.glob(tmp_rname)
local_names = []
for rem in remote_names:
to_append = rem.split(
os.path.sep
)[-depth:] if depth > 0 else []
local_names.append(
os.path.sep.join([tmp_lname] +
to_append))
else:
remote_names = [tmp_rname]
to_append = remote_names.split(
os.path.sep
)[-depth:] if depth > 0 else []
local_names = [
os.path.sep.join([tmp_lname] +
to_append)
]
if depth > 1: # create directories in the folder, if needed
for this_local_file in local_names:
new_folder = os.path.join(
folder.abspath,
os.path.split(this_local_file)[0])
if not os.path.exists(new_folder):
os.makedirs(new_folder)
else: # it is a string
if t.has_magic(item):
remote_names = t.glob(item)
local_names = [
os.path.split(rem)[1]
for rem in remote_names
]
else:
remote_names = [item]
local_names = [os.path.split(item)[1]]
for rem, loc in zip(remote_names, local_names):
execlogger.debug(
"[retrieval of calc {}] "
"Trying to retrieve remote item '{}'".
format(calc.pk, rem),
extra=logger_extra)
t.get(rem,
os.path.join(folder.abspath, loc),
ignore_nonexisting=True)
# Here I retrieved everything;
# now I store them inside the calculation
retrieved_files.replace_with_folder(folder.abspath,
overwrite=True)
# Second, retrieve the singlefiles
with SandboxFolder() as folder:
singlefile_list = []
for (linkname, subclassname,
filename) in retrieve_singlefile_list:
execlogger.debug(
"[retrieval of calc {}] Trying "
"to retrieve remote singlefile '{}'".format(
calc.pk, filename),
extra=logger_extra)
localfilename = os.path.join(
folder.abspath,
os.path.split(filename)[1])
t.get(filename,
localfilename,
ignore_nonexisting=True)
singlefile_list.append(
(linkname, subclassname, localfilename))
# ignore files that have not been retrieved
singlefile_list = [
i for i in singlefile_list if os.path.exists(i[2])
]
# after retrieving from the cluster, I create the objects
singlefiles = []
for (linkname, subclassname,
filename) in singlefile_list:
SinglefileSubclass = DataFactory(subclassname)
singlefile = SinglefileSubclass()
singlefile.set_file(filename)
singlefile.add_link_from(calc,
label=linkname,
link_type=LinkType.CREATE)
singlefiles.append(singlefile)
# Finally, store
execlogger.debug("[retrieval of calc {}] "
"Storing retrieved_files={}".format(
calc.pk, retrieved_files.dbnode.pk),
extra=logger_extra)
retrieved_files.store()
for fil in singlefiles:
execlogger.debug(
"[retrieval of calc {}] "
"Storing retrieved_singlefile={}".format(
calc.pk, fil.dbnode.pk),
extra=logger_extra)
fil.store()
# If I was the one retrieving, I should also be the only
# one parsing! I do not check
calc._set_state(calc_states.PARSING)
Parser = calc.get_parserclass()
# If no parser is set, the calculation is successful
successful = True
if Parser is not None:
# TODO: parse here
parser = Parser(calc)
successful, new_nodes_tuple = parser.parse_from_calc()
for label, n in new_nodes_tuple:
n.add_link_from(calc,
label=label,
link_type=LinkType.CREATE)
n.store()
if successful:
try:
calc._set_state(calc_states.FINISHED)
except ModificationNotAllowed:
# I should have been the only one to set it, but
# in order to avoid unuseful error messages, I
# just ignore
pass
else:
try:
calc._set_state(calc_states.FAILED)
except ModificationNotAllowed:
# I should have been the only one to set it, but
# in order to avoid unuseful error messages, I
# just ignore
pass
execlogger.error(
"[parsing of calc {}] "
"The parser returned an error, but it should have "
"created an output node with some partial results "
"and warnings. Check there for more information on "
"the problem".format(calc.pk),
extra=logger_extra)
retrieved.append(calc)
except Exception:
import traceback
tb = traceback.format_exc()
newextradict = logger_extra.copy()
newextradict['full_traceback'] = tb
if calc.get_state() == calc_states.PARSING:
execlogger.error("Error parsing calc {}. "
"Traceback: {}".format(calc.pk, tb),
extra=newextradict)
# TODO: add a 'comment' to the calculation
try:
calc._set_state(calc_states.PARSINGFAILED)
except ModificationNotAllowed:
pass
else:
execlogger.error("Error retrieving calc {}. "
"Traceback: {}".format(calc.pk, tb),
extra=newextradict)
try:
calc._set_state(calc_states.RETRIEVALFAILED)
except ModificationNotAllowed:
pass
raise
return retrieved
def submit_jobs_with_authinfo(authinfo):
"""
Submit jobs in TOSUBMIT status belonging
to user and machine as defined in the 'dbauthinfo' table.
"""
from aiida.orm import JobCalculation
from aiida.utils.logger import get_dblogger_extra
if not authinfo.enabled:
return
execlogger.debug("Submitting jobs for user {} "
"and machine {}".format(authinfo.aiidauser.email,
authinfo.dbcomputer.name))
# This returns an iterator over aiida JobCalculation objects
calcs_to_inquire = list(
JobCalculation._get_all_with_state(state=calc_states.TOSUBMIT,
computer=authinfo.dbcomputer,
user=authinfo.aiidauser))
# I avoid to open an ssh connection if there are
# no calcs with state WITHSCHEDULER
if len(calcs_to_inquire):
# Open connection
try:
# I do it here so that the transport is opened only once per computer
with authinfo.get_transport() as t:
for c in calcs_to_inquire:
logger_extra = get_dblogger_extra(c)
t._set_logger_extra(logger_extra)
try:
submit_calc(calc=c, authinfo=authinfo, transport=t)
except Exception as e:
# TODO: implement a counter, after N retrials
# set it to a status that
# requires the user intervention
execlogger.warning("There was an exception for "
"calculation {} ({}): {}".format(
c.pk, e.__class__.__name__,
e.message))
# I just proceed to the next calculation
continue
# Catch exceptions also at this level (this happens only if there is
# a problem opening the transport in the 'with t' statement,
# because any other exception is caught and skipped above
except Exception as e:
import traceback
from aiida.utils.logger import get_dblogger_extra
for calc in calcs_to_inquire:
logger_extra = get_dblogger_extra(calc)
try:
calc._set_state(calc_states.SUBMISSIONFAILED)
except ModificationNotAllowed:
# Someone already set it, just skip
pass
execlogger.error(
"Submission of calc {} failed, check also the "
"log file! Traceback: {}".format(calc.pk,
traceback.format_exc()),
extra=logger_extra)
raise
def update_running_calcs_status(authinfo):
"""
Update the states of calculations in WITHSCHEDULER status belonging
to user and machine as defined in the 'dbauthinfo' table.
"""
from aiida.orm import JobCalculation, Computer
from aiida.scheduler.datastructures import JobInfo
from aiida.utils.logger import get_dblogger_extra
if not authinfo.enabled:
return
execlogger.debug("Updating running calc status for user {} "
"and machine {}".format(authinfo.aiidauser.email,
authinfo.dbcomputer.name))
# This returns an iterator over aiida JobCalculation objects
calcs_to_inquire = list(
JobCalculation._get_all_with_state(state=calc_states.WITHSCHEDULER,
computer=authinfo.dbcomputer,
user=authinfo.aiidauser))
# NOTE: no further check is done that machine and
# aiidauser are correct for each calc in calcs
s = Computer(dbcomputer=authinfo.dbcomputer).get_scheduler()
t = authinfo.get_transport()
computed = []
# I avoid to open an ssh connection if there are
# no calcs with state WITHSCHEDULER
if len(calcs_to_inquire):
jobids_to_inquire = [str(c.get_job_id()) for c in calcs_to_inquire]
# Open connection
with t:
s.set_transport(t)
# TODO: Check if we are ok with filtering by job (to make this work,
# I had to remove the check on the retval for getJobs,
# because if the job has computed and is not in the output of
# qstat, it gives a nonzero retval)
# TODO: catch SchedulerError exception and do something
# sensible (at least, skip this computer but continue with
# following ones, and set a counter; set calculations to
# UNKNOWN after a while?
if s.get_feature('can_query_by_user'):
found_jobs = s.getJobs(user="******", as_dict=True)
else:
found_jobs = s.getJobs(jobs=jobids_to_inquire, as_dict=True)
# I update the status of jobs
for c in calcs_to_inquire:
try:
logger_extra = get_dblogger_extra(c)
t._set_logger_extra(logger_extra)
jobid = c.get_job_id()
if jobid is None:
execlogger.error("JobCalculation {} is WITHSCHEDULER "
"but no job id was found!".format(
c.pk),
extra=logger_extra)
continue
# I check if the calculation to be checked (c)
# is in the output of qstat
if jobid in found_jobs:
# jobinfo: the information returned by
# qstat for this job
jobinfo = found_jobs[jobid]
execlogger.debug("Inquirying calculation {} (jobid "
"{}): it has job_state={}".format(
c.pk, jobid, jobinfo.job_state),
extra=logger_extra)
# For the moment, FAILED is not defined
if jobinfo.job_state in [job_states.DONE
]: # , job_states.FAILED]:
computed.append(c)
try:
c._set_state(calc_states.COMPUTED)
except ModificationNotAllowed:
# Someone already set it, just skip
pass
## Do not set the WITHSCHEDULER state multiple times,
## this would raise a ModificationNotAllowed
# else:
# c._set_state(calc_states.WITHSCHEDULER)
c._set_scheduler_state(jobinfo.job_state)
c._set_last_jobinfo(jobinfo)
else:
execlogger.debug("Inquirying calculation {} (jobid "
"{}): not found, assuming "
"job_state={}".format(
c.pk, jobid, job_states.DONE),
extra=logger_extra)
# calculation c is not found in the output of qstat
computed.append(c)
c._set_scheduler_state(job_states.DONE)
except Exception as e:
# TODO: implement a counter, after N retrials
# set it to a status that
# requires the user intervention
execlogger.warning("There was an exception for "
"calculation {} ({}): {}".format(
c.pk, e.__class__.__name__,
e.message),
extra=logger_extra)
continue
for c in computed:
try:
logger_extra = get_dblogger_extra(c)
try:
detailed_jobinfo = s.get_detailed_jobinfo(
jobid=c.get_job_id())
except NotImplementedError:
detailed_jobinfo = (
u"AiiDA MESSAGE: This scheduler does not implement "
u"the routine get_detailed_jobinfo to retrieve "
u"the information on "
u"a job after it has finished.")
last_jobinfo = c._get_last_jobinfo()
if last_jobinfo is None:
last_jobinfo = JobInfo()
last_jobinfo.job_id = c.get_job_id()
last_jobinfo.job_state = job_states.DONE
last_jobinfo.detailedJobinfo = detailed_jobinfo
c._set_last_jobinfo(last_jobinfo)
except Exception as e:
execlogger.warning("There was an exception while "
"retrieving the detailed jobinfo "
"for calculation {} ({}): {}".format(
c.pk, e.__class__.__name__,
e.message),
extra=logger_extra)
continue
finally:
# Set the state to COMPUTED as the very last thing
# of this routine; no further change should be done after
# this, so that in general the retriever can just
# poll for this state, if we want to.
try:
c._set_state(calc_states.COMPUTED)
except ModificationNotAllowed:
# Someone already set it, just skip
pass
return computed
