transitions.py 9.66 KB
Newer Older
Michael Salim's avatar
Michael Salim committed
1 2
'''BalsamJob pre and post execution'''
from collections import namedtuple
3 4 5
import logging

from django.core.exceptions import ObjectDoesNotExist
Michael Salim's avatar
Michael Salim committed
6
from django.conf import settings
Michael Salim's avatar
Michael Salim committed
7
from django import db
8

Michael Salim's avatar
Michael Salim committed
9
from common import transfer
Michael Salim's avatar
Michael Salim committed
10
from balsam.launcher.exceptions import *
Michael Salim's avatar
Michael Salim committed
11
logger = logging.getLogger(__name__)
Michael Salim's avatar
Michael Salim committed
12 13 14 15 16


StatusMsg = namedtuple('Status', ['pk', 'state', 'msg'])
JobMsg =   namedtuple('JobMsg', ['pk', 'transition_function'])

Michael Salim's avatar
Michael Salim committed
17

Michael Salim's avatar
Michael Salim committed
18
def main(job_queue, status_queue):
Michael Salim's avatar
Michael Salim committed
19
    db.connection.close()
Michael Salim's avatar
Michael Salim committed
20 21
    while True:
        job, process_function = job_queue.get()
Michael Salim's avatar
Michael Salim committed
22 23
        if job == 'end': return

Michael Salim's avatar
Michael Salim committed
24 25 26 27 28 29 30 31 32
        try:
            process_function(job)
        except BalsamTransitionError as e:
            s = StatusMsg(job.pk, 'FAILED', str(e))
            status_queue.put(s)
        else:
            s = StatusMsg(job.pk, job.state, 'success')
            status_queue.put(s)

Michael Salim's avatar
Michael Salim committed
33

Michael Salim's avatar
Michael Salim committed
34
class TransitionProcessPool:
Michael Salim's avatar
Michael Salim committed
35
    
Michael Salim's avatar
Michael Salim committed
36
    NUM_PROC = settings.BALSAM_MAX_CONCURRENT_TRANSITIONS
Michael Salim's avatar
Michael Salim committed
37

Michael Salim's avatar
Michael Salim committed
38 39 40 41 42 43 44 45 46 47 48
    def __init__(self):
        
        self.job_queue = multiprocessing.Queue()
        self.status_queue = multiprocessing.Queue()
        self.transitions_pk_list = []

        self.procs = [
            multiprocessing.Process( target=main, 
                                    args=(self.job_queue, self.status_queue))
            for i in range(NUM_PROC)
        ]
Michael Salim's avatar
Michael Salim committed
49
        db.connections.close_all()
Michael Salim's avatar
Michael Salim committed
50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72
        for proc in self.procs: proc.start()

    def __contains__(self, job):
        return job.pk in self.transitions_pk_list

    def add_job(self, job):
        if job in self: raise BalsamTransitionError("already in transition")
        if job.state not in TRANSITIONS: raise TransitionNotFoundError
        pk = job.pk
        transition_function = TRANSITIONS[job.state]
        m = JobMsg(pk, transition_function)
        self.job_queue.put(m)
        self.transitions_pk_list.append(pk)

    def get_statuses():
        while not self.status_queue.empty():
            try:
                stat = self.status_queue.get_nowait()
                self.transitions_pk_list.remove(stat.pk)
                yield stat
            except queue.Empty:
                break

Michael Salim's avatar
Michael Salim committed
73
    def flush_job_queue(self):
Michael Salim's avatar
Michael Salim committed
74 75 76 77 78
        while not self.job_queue.empty():
            try:
                self.job_queue.get_nowait()
            except queue.Empty:
                break
Michael Salim's avatar
Michael Salim committed
79 80

    def end_and_wait(self):
Michael Salim's avatar
Michael Salim committed
81 82 83
        m = JobMsg('end', None)
        for proc in self.procs:
            self.job_queue.put(m)
Michael Salim's avatar
Michael Salim committed
84
        for proc in self.procs: proc.wait()
Michael Salim's avatar
Michael Salim committed
85 86
        self.transitions_pk_list = []

87 88

def check_parents(job):
Michael Salim's avatar
Michael Salim committed
89 90 91 92 93 94
    parents = job.get_parents()
    ready = all(p.state == 'JOB_FINISHED' for p in parents)
    if ready:
        job.update_state('READY', 'dependencies satisfied')
    elif job.state == 'CREATED':
        job.update_state('NOT_READY', 'awaiting dependencies')
95 96 97


def stage_in(job):
Michael Salim's avatar
Michael Salim committed
98
    # Create workdirs for jobs: use job.create_working_path
99
    logger.debug('in stage_in')
Michael Salim's avatar
Michael Salim committed
100 101 102 103 104
    job.update_state('STAGING_IN')

    if not os.path.exists(job.working_directory):
        job.create_working_path()

105 106 107 108 109 110 111 112 113 114 115 116
    if job.input_url != '':
        try:
            transfer.stage_in(job.input_url + '/', job.working_directory + '/')
            job.state = STAGED_IN.name
        except Exception as e:
            message = 'Exception received during stage_in: ' + str(e)
            logger.error(message)
            job.state = STAGE_IN_FAILED.name
    else:
        # no input url specified so stage in is complete
        job.state = STAGED_IN.name

Michael Salim's avatar
Michael Salim committed
117
    job.update_state('STAGE_IN_DONE')
118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254

def stage_out(job):
    ''' if the job has files defined via the output_files and an output_url is defined,
        they are copied from the local working_directory to the output_url '''
    logger.debug('in stage_out')
    message = None
    if job.output_url != '':
        try:
            transfer.stage_out(
                job.working_directory + '/',
                job.output_url + '/')
            job.state = STAGED_OUT.name
        except Exception as e:
            message = 'Exception received during stage_out: ' + str(e)
            logger.error(message)
            job.state = STAGE_OUT_FAILED.name
    else:
        # no output url specififed so stage out is complete
        job.state = STAGED_OUT.name

    job.save(
        update_fields=['state'],
        using=db_tools.get_db_connection_id(
            job.pk))
    status_sender = BalsamStatusSender.BalsamStatusSender(
        settings.SENDER_CONFIG)
    status_sender.send_status(job, message)

# preprocess a job


def preprocess(job):
    ''' Each job defines a task to perform, so tasks need preprocessing to prepare
        for the job to be submitted to the batch queue. '''
    logger.debug('in preprocess ')
    message = 'Job prepocess complete.'
    # get the task that is running
    try:
        app = ApplicationDefinition.objects.get(name=job.application)
        if app.preprocess:
            if os.path.exists(app.preprocess):
                stdout = run_subprocess.run_subprocess(app.preprocess)
                # write stdout to log file
                f = open(os.path.join(job.working_directory, app.name +
                                      '.preprocess.log.pid' + str(os.getpid())), 'wb')
                f.write(stdout)
                f.close()
                job.state = PREPROCESSED.name
            else:
                message = ('Preprocess, "' + app.preprocess + '", of application, "' + str(job.application)
                           + '", does not exist on filesystem.')
                logger.error(message)
                job.state = PREPROCESS_FAILED.name
        else:
            logger.debug('No preprocess specified for this job; skipping')
            job.state = PREPROCESSED.name
    except run_subprocess.SubprocessNonzeroReturnCode as e:
        message = ('Preprocess, "' + app.preprocess + '", of application, "' + str(job.application)
                   + '", exited with non-zero return code: ' + str(returncode))
        logger.error(message)
        job.state = PREPROCESS_FAILED.name
    except run_subprocess.SubprocessFailed as e:
        message = ('Received exception while running preprocess, "' + app.preprocess
                   + '", of application, "' + str(job.application) + '", exception: ' + str(e))
        logger.error(message)
        job.state = PREPROCESS_FAILED.name
    except ObjectDoesNotExist as e:
        message = 'application,' + str(job.application) + ', does not exist.'
        logger.error(message)
        job.state = PREPROCESS_FAILED.name
    except Exception as e:
        message = 'Received exception while in preprocess, "' + \
            app.preprocess + '", for application ' + str(job.application)
        logger.exception(message)
        job.state = PREPROCESS_FAILED.name

    job.save(
        update_fields=['state'],
        using=db_tools.get_db_connection_id(
            job.pk))
    status_sender = BalsamStatusSender.BalsamStatusSender(
        settings.SENDER_CONFIG)
    status_sender.send_status(job, message)

# perform any post job processing needed
def postprocess(job):
    ''' some jobs need to have some postprocessing performed,
        this function does this.'''
    logger.debug('in postprocess ')
    message = 'Job postprocess complete'
    try:
        app = ApplicationDefinition.objects.get(name=job.application)
        if app.postprocess:
            if os.path.exists(app.postprocess):
                stdout = run_subprocess.run_subprocess(app.postprocess)
                # write stdout to log file
                f = open(os.path.join(job.working_directory, app.name +
                                      '.postprocess.log.pid' + str(os.getpid())), 'wb')
                f.write(stdout)
                f.close()
                job.state = POSTPROCESSED.name
            else:
                message = ('Postprocess, "' + app.postprocess + '", of application, "' + str(job.application)
                           + '", does not exist on filesystem.')
                logger.error(message)
                job.state = POSTPROCESS_FAILED.name
        else:
            logger.debug('No postprocess specified for this job; skipping')
            job.state = POSTPROCESSED.name
    except run_subprocess.SubprocessNonzeroReturnCode as e:
        message = ('Postprocess, "' + app.postprocess + '", of application, "' + str(job.application)
                   + '", exited with non-zero return code: ' + str(returncode))
        logger.error(message)
        job.state = POSTPROCESS_FAILED.name
    except run_subprocess.SubprocessFailed as e:
        message = ('Received exception while running postprocess, "' + app.preprocess
                   + '", of application, "' + str(job.application) + '", exception: ' + str(e))
        logger.error(message)
        job.state = POSTPROCESS_FAILED.name
    except ObjectDoesNotExist as e:
        message = 'application,' + str(job.application) + ', does not exist.'
        logger.error(message)
        job.state = POSTPROCESS_FAILED.name
    except Exception as e:
        message = 'Received exception while in postprocess, "' + \
            app.postprocess + '", for application ' + str(job.application)
        logger.error(message)
        job.state = POSTPROCESS_FAILED.name

    job.save(
        update_fields=['state'],
        using=db_tools.get_db_connection_id(
            job.pk))
    status_sender = BalsamStatusSender.BalsamStatusSender(
        settings.SENDER_CONFIG)
    status_sender.send_status(job, message)

Michael Salim's avatar
Michael Salim committed
255 256 257 258 259 260 261 262 263 264 265
TRANSITIONS = {
    'CREATED':          check_parents,
    'LAUNCHER_QUEUED':  check_parents,
    'AWAITING_PARENTS': check_parents,
    'READY':            stage_in,
    'STAGED_IN':        preprocess,
    'RUN_DONE':         postprocess,
    'RUN_TIMEOUT':      postprocess,
    'RUN_ERROR':        postprocess,
    'POSTPROCESSED':    stage_out,
}