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 get_calculations(self, state=None):
from aiida.orm import JobCalculation
if (state == None):
return JobCalculation.query(workflow_step=self)
else:
return JobCalculation.query(workflow_step=self).filter(
dbattributes__key="state", dbattributes__tval=state)
def test_with_subclasses(self, computer):
extra_name = self.__class__.__name__ + "/test_with_subclasses"
calc_params = {
'computer': computer,
'resources': {
'num_machines': 1,
'num_mpiprocs_per_machine': 1
}
}
TemplateReplacerCalc = CalculationFactory(
'simpleplugins.templatereplacer')
ParameterData = DataFactory('parameter')
a1 = JobCalculation(**calc_params).store()
# To query only these nodes later
a1.set_extra(extra_name, True)
a2 = TemplateReplacerCalc(**calc_params).store()
# To query only these nodes later
a2.set_extra(extra_name, True)
a3 = Data().store()
a3.set_extra(extra_name, True)
a4 = ParameterData(dict={'a': 'b'}).store()
a4.set_extra(extra_name, True)
a5 = Node().store()
a5.set_extra(extra_name, True)
# I don't set the extras, just to be sure that the filtering works
# The filtering is needed because other tests will put stuff int he DB
a6 = JobCalculation(**calc_params)
a6.store()
a7 = Node()
a7.store()
# Query by calculation
results = list(JobCalculation.query(dbextras__key=extra_name))
# a3, a4, a5 should not be found because they are not JobCalculations.
# a6, a7 should not be found because they have not the attribute set.
self.assertEquals(set([i.pk for i in results]), set([a1.pk, a2.pk]))
# Same query, but by the generic Node class
results = list(Node.query(dbextras__key=extra_name))
self.assertEquals(set([i.pk for i in results]),
set([a1.pk, a2.pk, a3.pk, a4.pk, a5.pk]))
# Same query, but by the Data class
results = list(Data.query(dbextras__key=extra_name))
self.assertEquals(set([i.pk for i in results]), set([a3.pk, a4.pk]))
# Same query, but by the ParameterData subclass
results = list(ParameterData.query(dbextras__key=extra_name))
self.assertEquals(set([i.pk for i in results]), set([a4.pk]))
# Same query, but by the TemplateReplacerCalc subclass
results = list(TemplateReplacerCalc.query(dbextras__key=extra_name))
self.assertEquals(set([i.pk for i in results]), set([a2.pk]))
def fill_repo(self):
from aiida.orm import JobCalculation, CalculationFactory, Data, DataFactory
extra_name = self.__class__.__name__ + "/test_with_subclasses"
calc_params = {
'computer': self.computer,
'resources': {
'num_machines': 1,
'num_mpiprocs_per_machine': 1
}
}
TemplateReplacerCalc = CalculationFactory(
'simpleplugins.templatereplacer')
ParameterData = DataFactory('parameter')
a1 = JobCalculation(**calc_params).store()
# To query only these nodes later
a1.set_extra(extra_name, True)
a2 = TemplateReplacerCalc(**calc_params).store()
# To query only these nodes later
a2.set_extra(extra_name, True)
a3 = Data().store()
a3.set_extra(extra_name, True)
a4 = ParameterData(dict={'a': 'b'}).store()
a4.set_extra(extra_name, True)
a5 = Node().store()
a5.set_extra(extra_name, True)
# I don't set the extras, just to be sure that the filtering works
# The filtering is needed because other tests will put stuff int he DB
a6 = JobCalculation(**calc_params)
a6.store()
a7 = Node()
a7.store()
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 test_deletion(self):
from aiida.orm.computer import Computer
from aiida.orm import delete_computer, JobCalculation
from aiida.common.exceptions import InvalidOperation
newcomputer = Computer(name="testdeletioncomputer", hostname='localhost',
transport_type='local',
scheduler_type='pbspro',
workdir='/tmp/aiida').store()
# # This should be possible, because nothing is using this computer
delete_computer(newcomputer)
calc_params = {
'computer': self.computer,
'resources': {'num_machines': 1,
'num_mpiprocs_per_machine': 1}
}
_ = JobCalculation(**calc_params).store()
# This should fail, because there is at least a calculation
# using this computer (the one created just above)
with self.assertRaises(InvalidOperation):
delete_computer(self.computer)
def sub_create_bands_data(cls, user=None):
from aiida.orm.data.array.kpoints import KpointsData
from aiida.orm import JobCalculation
from aiida.orm.data.structure import StructureData
from aiida.common.links import LinkType
from aiida.orm.data.array.bands import BandsData
import numpy
s = StructureData(cell=((2., 0., 0.), (0., 2., 0.), (0., 0., 2.)))
s.append_atom(position=(0., 0., 0.),
symbols=['Ba', 'Ti'],
weights=(1., 0.),
name='mytype')
if user is not None:
s.dbnode.user = user._dbuser
s.store()
c = JobCalculation(computer=cls.computer,
resources={
'num_machines': 1,
'num_mpiprocs_per_machine': 1
})
if user is not None:
c.dbnode.user = user._dbuser
c.store()
c.add_link_from(s, "S1", LinkType.INPUT)
c._set_state(calc_states.RETRIEVING)
# define a cell
alat = 4.
cell = numpy.array([
[alat, 0., 0.],
[0., alat, 0.],
[0., 0., alat],
])
k = KpointsData()
k.set_cell(cell)
k.set_kpoints_path()
if user is not None:
k.dbnode.user = user._dbuser
k.store()
b = BandsData()
b.set_kpointsdata(k)
input_bands = numpy.array(
[numpy.ones(4) * i for i in range(k.get_kpoints().shape[0])])
b.set_bands(input_bands, units='eV')
if user is not None:
b.dbnode.user = user._dbuser
b.store()
b.add_link_from(c, link_type=LinkType.CREATE)
return b
def setUpClass(cls):
"""
Create some code to test the CalculationParamType parameter type for the command line infrastructure
We create an initial code with a random name and then on purpose create two code with a name
that matches exactly the ID and UUID, respectively, of the first one. This allows us to test
the rules implemented to solve ambiguities that arise when determing the identifier type
"""
super(TestCalculationParamType, cls).setUpClass()
cls.param = CalculationParamType()
cls.entity_01 = Calculation().store()
cls.entity_02 = Calculation().store()
cls.entity_03 = Calculation().store()
cls.entity_04 = FunctionCalculation()
cls.entity_05 = InlineCalculation()
cls.entity_06 = JobCalculation()
cls.entity_07 = WorkCalculation()
cls.entity_01.label = 'calculation_01'
cls.entity_02.label = str(cls.entity_01.pk)
cls.entity_03.label = str(cls.entity_01.uuid)
def get_calculations(self):
from aiida.orm import JobCalculation
return JobCalculation.query(workflow_step=self.steps)
def execute_steps():
"""
This method loops on the RUNNING workflows and handled the execution of the
steps until each workflow reaches an end (or gets stopped for errors).
In the loop for each RUNNING workflow the method loops also in each of its
RUNNING steps, testing if all the calculation and subworkflows attached to
the step are FINISHED. In this case the step is set as FINISHED and the
workflow is advanced to the step's next method present in the db with
``advance_workflow``, otherwise if any step's JobCalculation is found in
NEW state the method will submit. If none of the previous conditions apply
the step is flagged as ERROR and cannot proceed anymore, blocking the future
execution of the step and, connected, the workflow.
Finally, for each workflow the method tests if there are INITIALIZED steps
to be launched, and in case reloads the workflow and execute the specific
those steps. In case or error the step is flagged in ERROR state and the
stack is reported in the workflow report.
"""
from aiida.orm import JobCalculation
from aiida.orm.implementation import get_all_running_steps
logger.info("Querying the worflow DB")
running_steps = get_all_running_steps()
for s in running_steps:
if s.parent.state == wf_states.FINISHED:
s.set_state(wf_states.FINISHED)
continue
w = s.parent.get_aiida_class()
logger.info("[{0}] Found active step: {1}".format(w.pk, s.name))
s_calcs_new = [c.pk for c in s.get_calculations() if c._is_new()]
s_calcs_finished = [
c.pk for c in s.get_calculations() if c.has_finished_ok()
]
s_calcs_failed = [c.pk for c in s.get_calculations() if c.has_failed()]
s_calcs_num = len(s.get_calculations())
s_sub_wf_finished = [
sw.pk for sw in s.get_sub_workflows() if sw.has_finished_ok()
]
s_sub_wf_failed = [
sw.pk for sw in s.get_sub_workflows() if sw.has_failed()
]
s_sub_wf_num = len(s.get_sub_workflows())
if (s_calcs_num == (len(s_calcs_finished) + len(s_calcs_failed))
and s_sub_wf_num
== (len(s_sub_wf_finished) + len(s_sub_wf_failed))):
logger.info("[{0}] Step: {1} ready to move".format(w.pk, s.name))
s.set_state(wf_states.FINISHED)
advance_workflow(w, s)
elif len(s_calcs_new) > 0:
for pk in s_calcs_new:
obj_calc = JobCalculation.get_subclass_from_pk(pk=pk)
try:
obj_calc.submit()
logger.info(
"[{0}] Step: {1} launched calculation {2}".format(
w.pk, s.name, pk))
except:
logger.error(
"[{0}] Step: {1} cannot launch calculation {2}".format(
w.pk, s.name, pk))
def setUpClass(cls, *args, **kwargs):
super(TestVerdiCalculation, cls).setUpClass(*args, **kwargs)
from aiida.backends.tests.utils.fixtures import import_archive_fixture
from aiida.common.exceptions import ModificationNotAllowed
from aiida.common.links import LinkType
from aiida.orm import Code, Computer, Group, Node, JobCalculation, CalculationFactory
from aiida.orm.data.parameter import ParameterData
from aiida.orm.querybuilder import QueryBuilder
from aiida.work.processes import ProcessState
rmq_config = rmq.get_rmq_config()
# These two need to share a common event loop otherwise the first will never send
# the message while the daemon is running listening to intercept
cls.runner = runners.Runner(rmq_config=rmq_config,
rmq_submit=True,
poll_interval=0.)
cls.daemon_runner = runners.DaemonRunner(rmq_config=rmq_config,
rmq_submit=True,
poll_interval=0.)
cls.computer = Computer(name='comp',
hostname='localhost',
transport_type='local',
scheduler_type='direct',
workdir='/tmp/aiida').store()
cls.code = Code(remote_computer_exec=(cls.computer,
'/bin/true')).store()
cls.group = Group(name='test_group').store()
cls.node = Node().store()
cls.calcs = []
from aiida.orm.backend import construct_backend
backend = construct_backend()
authinfo = backend.authinfos.create(
computer=cls.computer, user=backend.users.get_automatic_user())
authinfo.store()
# Create 13 JobCalculations (one for each CalculationState)
for calculation_state in calc_states:
calc = JobCalculation(computer=cls.computer,
resources={
'num_machines': 1,
'num_mpiprocs_per_machine': 1
}).store()
# Trying to set NEW will raise, but in this case we don't need to change the state
try:
calc._set_state(calculation_state)
except ModificationNotAllowed:
pass
try:
exit_status = JobCalculationExitStatus[calculation_state]
except KeyError:
if calculation_state == 'IMPORTED':
calc._set_process_state(ProcessState.FINISHED)
else:
calc._set_process_state(ProcessState.RUNNING)
else:
calc._set_exit_status(exit_status)
calc._set_process_state(ProcessState.FINISHED)
cls.calcs.append(calc)
if calculation_state == 'PARSING':
cls.KEY_ONE = 'key_one'
cls.KEY_TWO = 'key_two'
cls.VAL_ONE = 'val_one'
cls.VAL_TWO = 'val_two'
output_parameters = ParameterData(dict={
cls.KEY_ONE: cls.VAL_ONE,
cls.KEY_TWO: cls.VAL_TWO,
}).store()
output_parameters.add_link_from(calc,
'output_parameters',
link_type=LinkType.RETURN)
# Create shortcut for easy dereferencing
cls.result_job = calc
# Add a single calc to a group
cls.group.add_nodes([calc])
# Load the fixture containing a single ArithmeticAddCalculation node
import_archive_fixture(
'calculation/simpleplugins.arithmetic.add.aiida')
# Get the imported ArithmeticAddCalculation node
ArithmeticAddCalculation = CalculationFactory(
'simpleplugins.arithmetic.add')
calculations = QueryBuilder().append(ArithmeticAddCalculation).all()[0]
cls.arithmetic_job = calculations[0]
def mounet_daemon():
return JobCalculation.query(dbattributes__key='state',
dbattributes__tval='WITHSCHEDULER').count()
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
def execute_steps():
"""
This method loops on the RUNNING workflows and handled the execution of the steps until
each workflow reaches an end (or gets stopped for errors).
In the loop for each RUNNING workflow the method loops also in each of its RUNNING steps,
testing if all the calculation and subworkflows attached to the step are FINISHED. In this
case the step is set as FINISHED and the workflow is advanced to the step's next method present
in the db with ``advance_workflow``, otherwise if any step's JobCalculation is found in NEW state
the method will submit. If none of the previous conditions apply the step is flagged as
ERROR and cannot proceed anymore, blocking the future execution of the step and, connected,
the workflow.
Finally, for each workflow the method tests if there are INITIALIZED steps to be launched,
and in case reloads the workflow and execute the specific those steps. In case or error
the step is flagged in ERROR state and the stack is reported in the workflow report.
"""
from aiida.backends.utils import get_automatic_user
from aiida.orm.workflow import Workflow
from aiida.common.datastructures import wf_states
from aiida.orm import JobCalculation
logger.info("Querying the worflow DB")
w_list = Workflow.query(user=get_automatic_user(), state=wf_states.RUNNING)
for w in w_list:
logger.info("Found active workflow: {0}".format(w.uuid))
# Launch INITIALIZED Workflows
#
running_steps = w.get_steps(state=wf_states.RUNNING)
for s in running_steps:
logger.info("[{0}] Found active step: {1}".format(w.uuid, s.name))
s_calcs_new = [c.uuid for c in s.get_calculations() if c._is_new()]
s_calcs_running = [
c.uuid for c in s.get_calculations() if c._is_running()
]
s_calcs_finished = [
c.uuid for c in s.get_calculations() if c.has_finished_ok()
]
s_calcs_failed = [
c.uuid for c in s.get_calculations() if c.has_failed()
]
s_calcs_num = len(s.get_calculations())
s_sub_wf_running = [
sw.uuid for sw in s.get_sub_workflows() if sw.is_running()
]
s_sub_wf_finished = [
sw.uuid for sw in s.get_sub_workflows()
if sw.has_finished_ok()
]
s_sub_wf_failed = [
sw.uuid for sw in s.get_sub_workflows() if sw.has_failed()
]
s_sub_wf_num = len(s.get_sub_workflows())
if s_calcs_num == (len(s_calcs_finished) + len(s_calcs_failed)
) and s_sub_wf_num == (len(s_sub_wf_finished) +
len(s_sub_wf_failed)):
logger.info("[{0}] Step: {1} ready to move".format(
w.uuid, s.name))
s.set_state(wf_states.FINISHED)
advance_workflow(w, s)
elif len(s_calcs_new) > 0:
for uuid in s_calcs_new:
obj_calc = JobCalculation.get_subclass_from_uuid(uuid=uuid)
try:
obj_calc.submit()
logger.info(
"[{0}] Step: {1} launched calculation {2}".format(
w.uuid, s.name, uuid))
except:
logger.error(
"[{0}] Step: {1} cannot launch calculation {2}".
format(w.uuid, s.name, uuid))
## DO NOT STOP ANYMORE IF A CALCULATION FAILS
# elif s_calcs_failed:
#s.set_state(wf_states.ERROR)
initialized_steps = w.get_steps(state=wf_states.INITIALIZED)
for s in initialized_steps:
import sys
try:
w_class = Workflow.get_subclass_from_uuid(w.uuid)
getattr(w, s.name)()
return True
except:
exc_type, exc_value, exc_traceback = sys.exc_info()
w.append_to_report(
"ERROR ! This workflow got an error in the {0} method, we report down the stack trace"
.format(s.name))
w.append_to_report("full traceback: {0}".format(
exc_traceback.format_exc()))
s.set_state(wf_states.ERROR)
w.set_state(wf_states.ERROR)
for w in w_list:
if w.get_steps(state=wf_states.ERROR):
w.set_state(wf_states.ERROR)
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 test_cif_structure_roundtrip(self):
from aiida.tools.dbexporters.tcod import export_cif, export_values
from aiida.orm import Code
from aiida.orm import JobCalculation
from aiida.orm.data.cif import CifData
from aiida.orm.data.parameter import ParameterData
from aiida.orm.data.upf import UpfData
from aiida.orm.data.folder import FolderData
from aiida.common.folders import SandboxFolder
from aiida.common.datastructures import calc_states
import tempfile
with tempfile.NamedTemporaryFile() as f:
f.write('''
data_test
_cell_length_a 10
_cell_length_b 10
_cell_length_c 10
_cell_angle_alpha 90
_cell_angle_beta 90
_cell_angle_gamma 90
loop_
_atom_site_label
_atom_site_fract_x
_atom_site_fract_y
_atom_site_fract_z
C 0 0 0
O 0.5 0.5 0.5
''')
f.flush()
a = CifData(file=f.name)
c = a._get_aiida_structure()
c.store()
pd = ParameterData()
code = Code(local_executable='test.sh')
with tempfile.NamedTemporaryFile() as f:
f.write("#/bin/bash\n\necho test run\n")
f.flush()
code.add_path(f.name, 'test.sh')
code.store()
calc = JobCalculation(computer=self.computer)
calc.set_resources({'num_machines': 1, 'num_mpiprocs_per_machine': 1})
calc.add_link_from(code, "code")
calc.set_environment_variables({
'PATH': '/dev/null',
'USER': '******'
})
with tempfile.NamedTemporaryFile(prefix="Fe") as f:
f.write("<UPF version=\"2.0.1\">\nelement=\"Fe\"\n")
f.flush()
upf = UpfData(file=f.name)
upf.store()
calc.add_link_from(upf, "upf")
with tempfile.NamedTemporaryFile() as f:
f.write("data_test")
f.flush()
cif = CifData(file=f.name)
cif.store()
calc.add_link_from(cif, "cif")
calc.store()
calc._set_state(calc_states.SUBMITTING)
with SandboxFolder() as f:
calc._store_raw_input_folder(f.abspath)
fd = FolderData()
with open(
fd._get_folder_pathsubfolder.get_abs_path(
calc._SCHED_OUTPUT_FILE), 'w') as f:
f.write("standard output")
f.flush()
with open(
fd._get_folder_pathsubfolder.get_abs_path(
calc._SCHED_ERROR_FILE), 'w') as f:
f.write("standard error")
f.flush()
fd.store()
fd.add_link_from(calc, calc._get_linkname_retrieved(), LinkType.CREATE)
pd.add_link_from(calc, "calc", LinkType.CREATE)
pd.store()
with self.assertRaises(ValueError):
export_cif(c, parameters=pd)
c.add_link_from(calc, "calc", LinkType.CREATE)
export_cif(c, parameters=pd)
values = export_values(c, parameters=pd)
values = values['0']
self.assertEquals(values['_tcod_computation_environment'],
['PATH=/dev/null\nUSER=unknown'])
self.assertEquals(values['_tcod_computation_command'],
['cd 1; ./_aiidasubmit.sh'])
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 setUpClass(cls, *args, **kwargs):
"""
Create some calculations with various states
"""
super(TestVerdiCalculationCommands, cls).setUpClass()
from aiida.orm import JobCalculation
# Create some calculation
calc1 = JobCalculation(computer=cls.computer,
resources={
'num_machines': 1,
'num_mpiprocs_per_machine': 1
}).store()
calc1._set_state(calc_states.TOSUBMIT)
calc2 = JobCalculation(computer=cls.computer.name,
resources={
'num_machines': 1,
'num_mpiprocs_per_machine': 1
}).store()
calc2._set_state(calc_states.COMPUTED)
calc3 = JobCalculation(computer=cls.computer.id,
resources={
'num_machines': 1,
'num_mpiprocs_per_machine': 1
}).store()
calc3._set_state(calc_states.FINISHED)
def mounet_daemon():
return JobCalculation.query(
dbattributes__key='state',
dbattributes__tval='WITHSCHEDULER').with_entities(func.count(
DbNode.id)).scalar()
