def test_submit_script(self): """ Test the creation of a simple submission script. """ from aiida.schedulers.datastructures import JobTemplate from aiida.common.datastructures import CodeInfo, CodeRunMode scheduler = SlurmScheduler() job_tmpl = JobTemplate() job_tmpl.shebang = '#!/bin/bash' job_tmpl.uuid = str(uuid.uuid4()) job_tmpl.job_resource = scheduler.create_job_resource(num_machines=1, num_mpiprocs_per_machine=1) job_tmpl.max_wallclock_seconds = 24 * 3600 code_info = CodeInfo() code_info.cmdline_params = ['mpirun', '-np', '23', 'pw.x', '-npool', '1'] code_info.stdin_name = 'aiida.in' job_tmpl.codes_info = [code_info] job_tmpl.codes_run_mode = CodeRunMode.SERIAL submit_script_text = scheduler.get_submit_script(job_tmpl) assert submit_script_text.startswith('#!/bin/bash') assert '#SBATCH --no-requeue' in submit_script_text assert '#SBATCH --time=1-00:00:00' in submit_script_text assert '#SBATCH --nodes=1' in submit_script_text assert "'mpirun' '-np' '23' 'pw.x' '-npool' '1' < 'aiida.in'" in submit_script_text
def test_submit_script_with_num_cores_per_machine_and_mpiproc1(self): # pylint: disable=invalid-name """ Test to verify if scripts works fine if we pass both num_cores_per_machine and num_cores_per_mpiproc correct values. It should pass in check: res.num_cores_per_mpiproc * res.num_mpiprocs_per_machine = res.num_cores_per_machine """ from aiida.schedulers.datastructures import JobTemplate from aiida.common.datastructures import CodeInfo, CodeRunMode scheduler = SlurmScheduler() job_tmpl = JobTemplate() job_tmpl.shebang = '#!/bin/bash' job_tmpl.job_resource = scheduler.create_job_resource( num_machines=1, num_mpiprocs_per_machine=1, num_cores_per_machine=24, num_cores_per_mpiproc=24 ) job_tmpl.uuid = str(uuid.uuid4()) job_tmpl.max_wallclock_seconds = 24 * 3600 code_info = CodeInfo() code_info.cmdline_params = ['mpirun', '-np', '23', 'pw.x', '-npool', '1'] code_info.stdin_name = 'aiida.in' job_tmpl.codes_info = [code_info] job_tmpl.codes_run_mode = CodeRunMode.SERIAL submit_script_text = scheduler.get_submit_script(job_tmpl) assert '#SBATCH --no-requeue' in submit_script_text assert '#SBATCH --time=1-00:00:00' in submit_script_text assert '#SBATCH --nodes=1' in submit_script_text assert '#SBATCH --ntasks-per-node=1' in submit_script_text assert '#SBATCH --cpus-per-task=24' in submit_script_text assert "'mpirun' '-np' '23' 'pw.x' '-npool' '1' < 'aiida.in'" in submit_script_text
def test_submit_script_with_num_cores_per_mpiproc(self): """ Test to verify if scripts works fine if we pass only num_cores_per_mpiproc value """ from aiida.schedulers.datastructures import JobTemplate from aiida.common.datastructures import CodeInfo, CodeRunMode scheduler = SlurmScheduler() job_tmpl = JobTemplate() job_tmpl.shebang = '#!/bin/bash' job_tmpl.job_resource = scheduler.create_job_resource( num_machines=1, num_mpiprocs_per_machine=1, num_cores_per_mpiproc=24) job_tmpl.uuid = str(uuid.uuid4()) job_tmpl.max_wallclock_seconds = 24 * 3600 code_info = CodeInfo() code_info.cmdline_params = ['mpirun', '-np', '23', 'pw.x', '-npool', '1'] code_info.stdin_name = 'aiida.in' job_tmpl.codes_info = [code_info] job_tmpl.codes_run_mode = CodeRunMode.SERIAL submit_script_text = scheduler.get_submit_script(job_tmpl) self.assertTrue('#SBATCH --no-requeue' in submit_script_text) self.assertTrue('#SBATCH --time=1-00:00:00' in submit_script_text) self.assertTrue('#SBATCH --nodes=1' in submit_script_text) self.assertTrue('#SBATCH --ntasks-per-node=1' in submit_script_text) self.assertTrue('#SBATCH --cpus-per-task=24' in submit_script_text) self.assertTrue("'mpirun' '-np' '23' 'pw.x' '-npool' '1'" + \ " < 'aiida.in'" in submit_script_text)
def test_submit_script_with_num_cores_per_machine(self): """ Test to verify if script works fine if we specify only num_cores_per_machine value. """ from aiida.schedulers.datastructures import JobTemplate from aiida.common.datastructures import CodeInfo, CodeRunMode scheduler = TorqueScheduler() job_tmpl = JobTemplate() job_tmpl.shebang = '#!/bin/bash' job_tmpl.job_resource = scheduler.create_job_resource( num_machines=1, num_mpiprocs_per_machine=1, num_cores_per_machine=24) job_tmpl.uuid = str(uuid.uuid4()) job_tmpl.max_wallclock_seconds = 24 * 3600 code_info = CodeInfo() code_info.cmdline_params = [ 'mpirun', '-np', '23', 'pw.x', '-npool', '1' ] code_info.stdin_name = 'aiida.in' job_tmpl.codes_info = [code_info] job_tmpl.codes_run_mode = CodeRunMode.SERIAL submit_script_text = scheduler.get_submit_script(job_tmpl) self.assertTrue('#PBS -r n' in submit_script_text) self.assertTrue(submit_script_text.startswith('#!/bin/bash')) self.assertTrue( '#PBS -l nodes=1:ppn=24,walltime=24:00:00' in submit_script_text) self.assertTrue("'mpirun' '-np' '23' 'pw.x' '-npool' '1'" + " < 'aiida.in'" in submit_script_text)
def test_submit_script_bad_shebang(self): """ Test to verify if scripts works fine with default options """ from aiida.schedulers.datastructures import JobTemplate from aiida.common.datastructures import CodeInfo, CodeRunMode scheduler = PbsproScheduler() code_info = CodeInfo() code_info.cmdline_params = ['mpirun', '-np', '23', 'pw.x', '-npool', '1'] code_info.stdin_name = 'aiida.in' for (shebang, expected_first_line) in ((None, '#!/bin/bash'), ('', ''), ('NOSET', '#!/bin/bash')): job_tmpl = JobTemplate() if shebang == 'NOSET': pass else: job_tmpl.shebang = shebang job_tmpl.job_resource = scheduler.create_job_resource(num_machines=1, num_mpiprocs_per_machine=1) job_tmpl.codes_info = [code_info] job_tmpl.codes_run_mode = CodeRunMode.SERIAL submit_script_text = scheduler.get_submit_script(job_tmpl) # This tests if the implementation correctly chooses the default: self.assertEqual(submit_script_text.split('\n')[0], expected_first_line)
def test_submit_script_with_num_cores_per_machine_and_mpiproc1(self): """ Test to verify if scripts works fine if we pass both num_cores_per_machine and num_cores_per_mpiproc correct values. It should pass in check: res.num_cores_per_mpiproc * res.num_mpiprocs_per_machine = res.num_cores_per_machine """ from aiida.schedulers.datastructures import JobTemplate from aiida.common.datastructures import CodeInfo, CodeRunMode scheduler = PbsproScheduler() job_tmpl = JobTemplate() job_tmpl.shebang = '#!/bin/bash' job_tmpl.job_resource = scheduler.create_job_resource( num_machines=1, num_mpiprocs_per_machine=1, num_cores_per_machine=24, num_cores_per_mpiproc=24 ) job_tmpl.uuid = str(uuid.uuid4()) job_tmpl.max_wallclock_seconds = 24 * 3600 code_info = CodeInfo() code_info.cmdline_params = ['mpirun', '-np', '23', 'pw.x', '-npool', '1'] code_info.stdin_name = 'aiida.in' job_tmpl.codes_info = [code_info] job_tmpl.codes_run_mode = CodeRunMode.SERIAL submit_script_text = scheduler.get_submit_script(job_tmpl) self.assertTrue('#PBS -r n' in submit_script_text) self.assertTrue(submit_script_text.startswith('#!/bin/bash')) self.assertTrue('#PBS -l select=1:mpiprocs=1:ppn=24' in submit_script_text) # Note: here 'num_cores_per_machine' should NOT override the mpiprocs self.assertTrue("'mpirun' '-np' '23' 'pw.x' '-npool' '1'" + " < 'aiida.in'" in submit_script_text)
def test_submit_script(self): """ Test the creation of a simple submission script. """ from aiida.schedulers.datastructures import JobTemplate from aiida.common.datastructures import CodeInfo, CodeRunMode scheduler = LsfScheduler() job_tmpl = JobTemplate() job_tmpl.shebang = '#!/bin/bash' job_tmpl.uuid = str(uuid.uuid4()) job_tmpl.job_resource = scheduler.create_job_resource(tot_num_mpiprocs=2, parallel_env='b681e480bd.cern.ch') job_tmpl.max_wallclock_seconds = 24 * 3600 code_info = CodeInfo() code_info.cmdline_params = ['mpirun', '-np', '2', 'pw.x', '-npool', '1'] code_info.stdin_name = 'aiida.in' job_tmpl.codes_info = [code_info] job_tmpl.codes_run_mode = CodeRunMode.SERIAL submit_script_text = scheduler.get_submit_script(job_tmpl) self.assertTrue(submit_script_text.startswith('#!/bin/bash')) self.assertTrue('#BSUB -rn' in submit_script_text) self.assertTrue('#BSUB -W 24:00' in submit_script_text) self.assertTrue('#BSUB -n 2' in submit_script_text) self.assertTrue("'mpirun' '-np' '2' 'pw.x' '-npool' '1'" + \ " < 'aiida.in'" in submit_script_text)
def test_submit_script(): scheduler = PbsproCx1Scheduler() job_tmpl = JobTemplate() job_tmpl.uuid = str(uuid.uuid4()) job_tmpl.shebang = '#!/bin/bash -l' job_tmpl.job_resource = scheduler.create_job_resource( num_machines=1, num_mpiprocs_per_machine=32) job_tmpl.max_memory_kb = 2e6 job_tmpl.max_wallclock_seconds = 24.5 * 3600 job_tmpl.queue_name = 'myqueue' code_info = CodeInfo() code_info.cmdline_params = ['mpirun', '-np', '23', 'pw.x', '-npool', '1'] code_info.stdin_name = 'aiida.in' job_tmpl.codes_info = [code_info] job_tmpl.codes_run_mode = CodeRunMode.SERIAL submit_script_text = scheduler.get_submit_script(job_tmpl).splitlines() assert '#PBS -l select=1:ncpus=32:mem=2gb' in submit_script_text assert '#PBS -l walltime=24:30:00' in submit_script_text assert '#PBS -q myqueue' in submit_script_text
def presubmit(self, folder): """Prepares the calculation folder with all inputs, ready to be copied to the cluster. :param folder: a SandboxFolder that can be used to write calculation input files and the scheduling script. :type folder: :class:`aiida.common.folders.Folder` :return calcinfo: the CalcInfo object containing the information needed by the daemon to handle operations. :rtype calcinfo: :class:`aiida.common.CalcInfo` """ # pylint: disable=too-many-locals,too-many-statements,too-many-branches import os from aiida.common.exceptions import PluginInternalError, ValidationError, InvalidOperation, InputValidationError from aiida.common import json from aiida.common.utils import validate_list_of_string_tuples from aiida.common.datastructures import CodeInfo, CodeRunMode from aiida.orm import load_node, Code, Computer from aiida.plugins import DataFactory from aiida.schedulers.datastructures import JobTemplate computer = self.node.computer inputs = self.node.get_incoming(link_type=LinkType.INPUT_CALC) if not self.inputs.metadata.dry_run and self.node.has_cached_links(): raise InvalidOperation('calculation node has unstored links in cache') codes = [_ for _ in inputs.all_nodes() if isinstance(_, Code)] for code in codes: if not code.can_run_on(computer): raise InputValidationError('The selected code {} for calculation {} cannot run on computer {}'.format( code.pk, self.node.pk, computer.name)) if code.is_local() and code.get_local_executable() in folder.get_content_list(): raise PluginInternalError('The plugin created a file {} that is also the executable name!'.format( code.get_local_executable())) calc_info = self.prepare_for_submission(folder) calc_info.uuid = str(self.node.uuid) scheduler = computer.get_scheduler() # I create the job template to pass to the scheduler job_tmpl = JobTemplate() job_tmpl.shebang = computer.get_shebang() job_tmpl.submit_as_hold = False job_tmpl.rerunnable = False job_tmpl.job_environment = {} # 'email', 'email_on_started', 'email_on_terminated', job_tmpl.job_name = 'aiida-{}'.format(self.node.pk) job_tmpl.sched_output_path = self.options.scheduler_stdout if self.options.scheduler_stderr == self.options.scheduler_stdout: job_tmpl.sched_join_files = True else: job_tmpl.sched_error_path = self.options.scheduler_stderr job_tmpl.sched_join_files = False # Set retrieve path, add also scheduler STDOUT and STDERR retrieve_list = (calc_info.retrieve_list if calc_info.retrieve_list is not None else []) if (job_tmpl.sched_output_path is not None and job_tmpl.sched_output_path not in retrieve_list): retrieve_list.append(job_tmpl.sched_output_path) if not job_tmpl.sched_join_files: if (job_tmpl.sched_error_path is not None and job_tmpl.sched_error_path not in retrieve_list): retrieve_list.append(job_tmpl.sched_error_path) self.node.set_retrieve_list(retrieve_list) retrieve_singlefile_list = (calc_info.retrieve_singlefile_list if calc_info.retrieve_singlefile_list is not None else []) # a validation on the subclasses of retrieve_singlefile_list for _, subclassname, _ in retrieve_singlefile_list: file_sub_class = DataFactory(subclassname) if not issubclass(file_sub_class, orm.SinglefileData): raise PluginInternalError( '[presubmission of calc {}] retrieve_singlefile_list subclass problem: {} is ' 'not subclass of SinglefileData'.format(self.node.pk, file_sub_class.__name__)) if retrieve_singlefile_list: self.node.set_retrieve_singlefile_list(retrieve_singlefile_list) # Handle the retrieve_temporary_list retrieve_temporary_list = (calc_info.retrieve_temporary_list if calc_info.retrieve_temporary_list is not None else []) self.node.set_retrieve_temporary_list(retrieve_temporary_list) # the if is done so that if the method returns None, this is # not added. This has two advantages: # - it does not add too many \n\n if most of the prepend_text are empty # - most importantly, skips the cases in which one of the methods # would return None, in which case the join method would raise # an exception prepend_texts = [computer.get_prepend_text()] + \ [code.get_prepend_text() for code in codes] + \ [calc_info.prepend_text, self.node.get_option('prepend_text')] job_tmpl.prepend_text = '\n\n'.join(prepend_text for prepend_text in prepend_texts if prepend_text) append_texts = [self.node.get_option('append_text'), calc_info.append_text] + \ [code.get_append_text() for code in codes] + \ [computer.get_append_text()] job_tmpl.append_text = '\n\n'.join(append_text for append_text in append_texts if append_text) # Set resources, also with get_default_mpiprocs_per_machine resources = self.node.get_option('resources') scheduler.preprocess_resources(resources, computer.get_default_mpiprocs_per_machine()) job_tmpl.job_resource = scheduler.create_job_resource(**resources) subst_dict = {'tot_num_mpiprocs': job_tmpl.job_resource.get_tot_num_mpiprocs()} for key, value in job_tmpl.job_resource.items(): subst_dict[key] = value mpi_args = [arg.format(**subst_dict) for arg in computer.get_mpirun_command()] extra_mpirun_params = self.node.get_option('mpirun_extra_params') # same for all codes in the same calc # set the codes_info if not isinstance(calc_info.codes_info, (list, tuple)): raise PluginInternalError('codes_info passed to CalcInfo must be a list of CalcInfo objects') codes_info = [] for code_info in calc_info.codes_info: if not isinstance(code_info, CodeInfo): raise PluginInternalError('Invalid codes_info, must be a list of CodeInfo objects') if code_info.code_uuid is None: raise PluginInternalError('CalcInfo should have ' 'the information of the code ' 'to be launched') this_code = load_node(code_info.code_uuid, sub_classes=(Code,)) this_withmpi = code_info.withmpi # to decide better how to set the default if this_withmpi is None: if len(calc_info.codes_info) > 1: raise PluginInternalError('For more than one code, it is ' 'necessary to set withmpi in ' 'codes_info') else: this_withmpi = self.node.get_option('withmpi') if this_withmpi: this_argv = (mpi_args + extra_mpirun_params + [this_code.get_execname()] + (code_info.cmdline_params if code_info.cmdline_params is not None else [])) else: this_argv = [this_code.get_execname()] + (code_info.cmdline_params if code_info.cmdline_params is not None else []) # overwrite the old cmdline_params and add codename and mpirun stuff code_info.cmdline_params = this_argv codes_info.append(code_info) job_tmpl.codes_info = codes_info # set the codes execution mode if len(codes) > 1: try: job_tmpl.codes_run_mode = calc_info.codes_run_mode except KeyError: raise PluginInternalError('Need to set the order of the code execution (parallel or serial?)') else: job_tmpl.codes_run_mode = CodeRunMode.SERIAL ######################################################################## custom_sched_commands = self.node.get_option('custom_scheduler_commands') if custom_sched_commands: job_tmpl.custom_scheduler_commands = custom_sched_commands job_tmpl.import_sys_environment = self.node.get_option('import_sys_environment') job_tmpl.job_environment = self.node.get_option('environment_variables') queue_name = self.node.get_option('queue_name') account = self.node.get_option('account') qos = self.node.get_option('qos') if queue_name is not None: job_tmpl.queue_name = queue_name if account is not None: job_tmpl.account = account if qos is not None: job_tmpl.qos = qos priority = self.node.get_option('priority') if priority is not None: job_tmpl.priority = priority max_memory_kb = self.node.get_option('max_memory_kb') if max_memory_kb is not None: job_tmpl.max_memory_kb = max_memory_kb max_wallclock_seconds = self.node.get_option('max_wallclock_seconds') if max_wallclock_seconds is not None: job_tmpl.max_wallclock_seconds = max_wallclock_seconds max_memory_kb = self.node.get_option('max_memory_kb') if max_memory_kb is not None: job_tmpl.max_memory_kb = max_memory_kb submit_script_filename = self.node.get_option('submit_script_filename') script_content = scheduler.get_submit_script(job_tmpl) folder.create_file_from_filelike(io.StringIO(script_content), submit_script_filename, 'w', encoding='utf8') subfolder = folder.get_subfolder('.aiida', create=True) subfolder.create_file_from_filelike(io.StringIO(json.dumps(job_tmpl)), 'job_tmpl.json', 'w', encoding='utf8') subfolder.create_file_from_filelike(io.StringIO(json.dumps(calc_info)), 'calcinfo.json', 'w', encoding='utf8') if calc_info.local_copy_list is None: calc_info.local_copy_list = [] if calc_info.remote_copy_list is None: calc_info.remote_copy_list = [] # Some validation this_pk = self.node.pk if self.node.pk is not None else '[UNSTORED]' local_copy_list = calc_info.local_copy_list try: validate_list_of_string_tuples(local_copy_list, tuple_length=3) except ValidationError as exc: raise PluginInternalError('[presubmission of calc {}] ' 'local_copy_list format problem: {}'.format(this_pk, exc)) remote_copy_list = calc_info.remote_copy_list try: validate_list_of_string_tuples(remote_copy_list, tuple_length=3) except ValidationError as exc: raise PluginInternalError('[presubmission of calc {}] ' 'remote_copy_list format problem: {}'.format(this_pk, exc)) for (remote_computer_uuid, _, dest_rel_path) in remote_copy_list: try: Computer.objects.get(uuid=remote_computer_uuid) # pylint: disable=unused-variable except exceptions.NotExistent: raise PluginInternalError('[presubmission of calc {}] ' 'The remote copy requires a computer with UUID={}' 'but no such computer was found in the ' 'database'.format(this_pk, remote_computer_uuid)) if os.path.isabs(dest_rel_path): raise PluginInternalError('[presubmission of calc {}] ' 'The destination path of the remote copy ' 'is absolute! ({})'.format(this_pk, dest_rel_path)) return calc_info