Change directory structure (move ./nopaque/* to ./)

This commit is contained in:
Patrick Jentsch
2021-07-20 15:07:42 +02:00
parent ff39d8d650
commit d6ab379418
231 changed files with 26 additions and 23 deletions

54
app/tasks/__init__.py Normal file
View File

@ -0,0 +1,54 @@
from .corpus_utils import CheckCorporaMixin
from .job_utils import CheckJobsMixin
from ..import db, socketio
import docker
class TaskRunner(CheckCorporaMixin, CheckJobsMixin):
def __init__(self):
self.docker = docker.from_env()
self._socketio_message_buffer = {}
def run(self):
self.check_corpora()
self.check_jobs()
db.session.commit()
self.flush_socketio_messages()
def buffer_socketio_message(self, event, payload, room,
msg_id=None, override_policy='replace'):
if room not in self._socketio_message_buffer:
self._socketio_message_buffer[room] = {}
if event not in self._socketio_message_buffer[room]:
self._socketio_message_buffer[room][event] = {}
if msg_id is None:
msg_id = len(self._socketio_message_buffer[room][event].keys())
if override_policy == 'append':
if msg_id in self._socketio_message_buffer[room][event]:
# If the current message value isn't a list, convert it!
if not isinstance(self._socketio_message_buffer[room][event][msg_id], list): # noqa
self._socketio_message_buffer[room][event][msg_id] = [self._socketio_message_buffer[room][event][msg_id]] # noqa
else:
self._socketio_message_buffer[room][event][msg_id] = []
self._socketio_message_buffer[room][event][msg_id].append(payload)
elif override_policy == 'replace':
self._socketio_message_buffer[room][event][msg_id] = payload
else:
raise Exception('Unknown override policy: {}'.format(override_policy)) # noqa
return msg_id
def buffer_user_patch_operation(self, ressource, patch_operation):
self.buffer_socketio_message('user_{}_patch'.format(ressource.user_id),
patch_operation,
'user_{}'.format(ressource.user_id),
msg_id='_', override_policy='append')
def clear_socketio_message_buffer(self):
self._socketio_message_buffer = {}
def flush_socketio_messages(self):
for room in self._socketio_message_buffer:
for event in self._socketio_message_buffer[room]:
for message in self._socketio_message_buffer[room][event]:
socketio.emit(event, self._socketio_message_buffer[room][event][message], room=room) # noqa
self.clear_socketio_message_buffer()

231
app/tasks/corpus_utils.py Normal file
View File

@ -0,0 +1,231 @@
from ..models import Corpus
import docker
import logging
import os
import shutil
class CheckCorporaMixin:
def check_corpora(self):
corpora = Corpus.query.all()
queued_corpora = list(filter(lambda corpus: corpus.status == 'queued', corpora)) # noqa
running_corpora = list(filter(lambda corpus: corpus.status == 'running', corpora)) # noqa
start_analysis_corpora = list(filter(lambda corpus: corpus.status == 'start analysis', corpora)) # noqa
analysing_corpora = list(filter(lambda corpus: corpus.status == 'analysing', corpora)) # noqa
stop_analysis_corpora = list(filter(lambda corpus: corpus.status == 'stop analysis', corpora)) # noqa
submitted_corpora = list(filter(lambda corpus: corpus.status == 'submitted', corpora)) # noqa
for corpus in submitted_corpora:
self.create_build_corpus_service(corpus)
for corpus in queued_corpora + running_corpora:
self.checkout_build_corpus_service(corpus)
for corpus in start_analysis_corpora:
self.create_cqpserver_container(corpus)
for corpus in analysing_corpora:
self.checkout_analysing_corpus_container(corpus)
for corpus in stop_analysis_corpora:
self.remove_cqpserver_container(corpus)
def create_build_corpus_service(self, corpus):
corpus_data_dir = os.path.join(corpus.path, 'data')
shutil.rmtree(corpus_data_dir, ignore_errors=True)
os.mkdir(corpus_data_dir)
corpus_registry_dir = os.path.join(corpus.path, 'registry')
shutil.rmtree(corpus_registry_dir, ignore_errors=True)
os.mkdir(corpus_registry_dir)
corpus_file = os.path.join(corpus.path, 'merged', 'corpus.vrt')
service_kwargs = {
'command': 'docker-entrypoint.sh build-corpus',
'constraints': ['node.role==worker'],
'labels': {'origin': 'nopaque',
'type': 'corpus.build',
'corpus_id': str(corpus.id)},
'mounts': [corpus_file + ':/root/files/corpus.vrt:ro',
corpus_data_dir + ':/corpora/data:rw',
corpus_registry_dir + ':/usr/local/share/cwb/registry:rw'],
'name': 'build-corpus_{}'.format(corpus.id),
'restart_policy': docker.types.RestartPolicy()
}
service_image = \
'gitlab.ub.uni-bielefeld.de:4567/sfb1288inf/cqpserver:latest'
try:
self.docker.services.create(service_image, **service_kwargs)
except docker.errors.APIError as e:
logging.error(
'Create "{}" service raised '.format(service_kwargs['name'])
+ '"docker.errors.APIError" The server returned an error. '
+ 'Details: {}'.format(e)
)
else:
corpus.status = 'queued'
patch_operation = {
'op': 'replace', 'path': '/corpora/{}/status'.format(corpus.id), 'value': corpus.status}
self.buffer_user_patch_operation(corpus, patch_operation)
def checkout_build_corpus_service(self, corpus):
service_name = 'build-corpus_{}'.format(corpus.id)
try:
service = self.docker.services.get(service_name)
except docker.errors.NotFound:
logging.error(
'Get "{}" service raised '.format(service_name)
+ '"docker.errors.NotFound" The service does not exist. '
+ '(corpus.status: {} -> failed)'.format(corpus.status)
)
corpus.status = 'failed'
patch_operation = {
'op': 'replace', 'path': '/corpora/{}/status'.format(corpus.id), 'value': corpus.status}
self.buffer_user_patch_operation(corpus, patch_operation)
except docker.errors.APIError as e:
logging.error(
'Get "{}" service raised '.format(service_name)
+ '"docker.errors.APIError" The server returned an error. '
+ 'Details: {}'.format(e)
)
except docker.errors.InvalidVersion:
logging.error(
'Get "{}" service raised '.format(service_name)
+ '"docker.errors.InvalidVersion" One of the arguments is '
+ 'not supported with the current API version.'
)
else:
service_tasks = service.tasks()
if not service_tasks:
return
task_state = service_tasks[0].get('Status').get('State')
if corpus.status == 'queued' and task_state != 'pending':
corpus.status = 'running'
patch_operation = {
'op': 'replace', 'path': '/corpora/{}/status'.format(corpus.id), 'value': corpus.status}
self.buffer_user_patch_operation(corpus, patch_operation)
elif (corpus.status == 'running'
and task_state in ['complete', 'failed']):
try:
service.remove()
except docker.errors.APIError as e:
logging.error(
'Remove "{}" service raised '.format(service_name)
+ '"docker.errors.APIError" The server returned an error. '
+ 'Details: {}'.format(e)
)
return
else:
corpus.status = 'prepared' if task_state == 'complete' \
else 'failed'
patch_operation = {
'op': 'replace', 'path': '/corpora/{}/status'.format(corpus.id), 'value': corpus.status}
self.buffer_user_patch_operation(corpus, patch_operation)
def create_cqpserver_container(self, corpus):
corpus_data_dir = os.path.join(corpus.path, 'data')
corpus_registry_dir = os.path.join(corpus.path, 'registry')
container_kwargs = {
'command': 'cqpserver',
'detach': True,
'volumes': [corpus_data_dir + ':/corpora/data:rw',
corpus_registry_dir + ':/usr/local/share/cwb/registry:rw'],
'name': 'cqpserver_{}'.format(corpus.id),
'network': 'nopaque_default'
}
container_image = \
'gitlab.ub.uni-bielefeld.de:4567/sfb1288inf/cqpserver:latest'
# Check if a cqpserver container already exists. If this is the case,
# remove it and create a new one
try:
container = self.docker.containers.get(container_kwargs['name'])
except docker.errors.NotFound:
pass
except docker.errors.APIError as e:
logging.error(
'Get "{}" container raised '.format(container_kwargs['name'])
+ '"docker.errors.APIError" The server returned an error. '
+ 'Details: {}'.format(e)
)
return
else:
try:
container.remove(force=True)
except docker.errors.APIError as e:
logging.error(
'Remove "{}" container raised '.format(
container_kwargs['name'])
+ '"docker.errors.APIError" The server returned an error. '
+ 'Details: {}'.format(e)
)
return
try:
self.docker.containers.run(container_image, **container_kwargs)
except docker.errors.ContainerError:
# This case should not occur, because detach is True.
logging.error(
'Run "{}" container raised '.format(container_kwargs['name'])
+ '"docker.errors.ContainerError" The container exits with a '
+ 'non-zero exit code and detach is False.'
)
corpus.status = 'failed'
patch_operation = {
'op': 'replace', 'path': '/corpora/{}/status'.format(corpus.id), 'value': corpus.status}
self.buffer_user_patch_operation(corpus, patch_operation)
except docker.errors.ImageNotFound:
logging.error(
'Run "{}" container raised '.format(container_kwargs['name'])
+ '"docker.errors.ImageNotFound" The specified image does not '
+ 'exist.'
)
corpus.status = 'failed'
patch_operation = {
'op': 'replace', 'path': '/corpora/{}/status'.format(corpus.id), 'value': corpus.status}
self.buffer_user_patch_operation(corpus, patch_operation)
except docker.errors.APIError as e:
logging.error(
'Run "{}" container raised '.format(container_kwargs['name'])
+ '"docker.errors.APIError" The server returned an error. '
+ 'Details: {}'.format(e)
)
else:
corpus.status = 'analysing'
patch_operation = {
'op': 'replace', 'path': '/corpora/{}/status'.format(corpus.id), 'value': corpus.status}
self.buffer_user_patch_operation(corpus, patch_operation)
def checkout_analysing_corpus_container(self, corpus):
container_name = 'cqpserver_{}'.format(corpus.id)
try:
self.docker.containers.get(container_name)
except docker.errors.NotFound:
logging.error('Could not find "{}" but the corpus state is "analysing".') # noqa
corpus.status = 'prepared'
except docker.errors.APIError as e:
logging.error(
'Get "{}" container raised '.format(container_name)
+ '"docker.errors.APIError" The server returned an error. '
+ 'Details: {}'.format(e)
)
return
def remove_cqpserver_container(self, corpus):
container_name = 'cqpserver_{}'.format(corpus.id)
try:
container = self.docker.containers.get(container_name)
except docker.errors.NotFound:
pass
except docker.errors.APIError as e:
logging.error(
'Get "{}" container raised '.format(container_name)
+ '"docker.errors.APIError" The server returned an error. '
+ 'Details: {}'.format(e)
)
return
else:
try:
container.remove(force=True)
except docker.errors.APIError as e:
logging.error(
'Remove "{}" container raised '.format(container_name)
+ '"docker.errors.APIError" The server returned an error. '
+ 'Details: {}'.format(e)
)
return
corpus.status = 'prepared'
patch_operation = {
'op': 'replace', 'path': '/corpora/{}/status'.format(corpus.id), 'value': corpus.status}
self.buffer_user_patch_operation(corpus, patch_operation)

206
app/tasks/job_utils.py Normal file
View File

@ -0,0 +1,206 @@
from datetime import datetime
from werkzeug.utils import secure_filename
from .. import db, mail
from ..email import create_message
from ..models import Job, JobResult
import docker
import logging
import json
import os
# TODO: Integrate the service_settings into app/services/__init__.py
service_settings = {
'file-setup': {
'default_args': ' --mem-mb 2048 --n-cores 2',
'ressources': docker.types.Resources(cpu_reservation=2 * (10 ** 9),
mem_reservation=2048 * (10 ** 6))
},
'nlp': {
'default_args': ' --mem-mb 2048 --n-cores 2',
'ressources': docker.types.Resources(cpu_reservation=2 * (10 ** 9),
mem_reservation=2048 * (10 ** 6))
},
'ocr': {
'default_args': ' --mem-mb 4096 --n-cores 4',
'ressources': docker.types.Resources(cpu_reservation=4 * (10 ** 9),
mem_reservation=4096 * (10 ** 6))
}
}
class CheckJobsMixin:
def check_jobs(self):
jobs = Job.query.all()
canceling_jobs = list(filter(lambda job: job.status == 'canceling', jobs)) # noqa
queued_jobs = list(filter(lambda job: job.status == 'queued', jobs))
running_jobs = list(filter(lambda job: job.status == 'running', jobs))
submitted_jobs = list(filter(lambda job: job.status == 'submitted', jobs)) # noqa
for job in submitted_jobs:
self.create_job_service(job)
for job in queued_jobs + running_jobs:
self.checkout_job_service(job)
for job in canceling_jobs:
self.remove_job_service(job)
def create_job_service(self, job):
cmd = job.service
cmd += ' -i /input'
cmd += ' -o /output'
cmd += ' --log-dir /input'
cmd += ' --zip [{}]_{}'.format(job.service, secure_filename(job.title))
cmd += service_settings[job.service]['default_args']
cmd += ' ' + ' '.join(json.loads(job.service_args))
# Setup input mount
input_mount_src = job.path
input_mount_dest = os.path.abspath('/input')
if job.service == 'file-setup':
input_mount_dest = os.path.join(input_mount_dest, secure_filename(job.title)) # noqa
input_mount = '{}:{}:rw'.format(input_mount_src, input_mount_dest)
# Setup output mount
output_mount_src = os.path.join(job.path, 'output')
output_mount_dest = os.path.abspath('/output')
os.makedirs(output_mount_src)
output_mount = '{}:{}:rw'.format(output_mount_src, output_mount_dest)
service_kwargs = {'command': cmd,
'constraints': ['node.role==worker'],
'labels': {'origin': 'nopaque',
'type': 'job',
'job_id': str(job.id)},
'mounts': [input_mount, output_mount],
'name': 'job_{}'.format(job.id),
'resources': service_settings[job.service]['ressources'], # noqa
'restart_policy': docker.types.RestartPolicy()}
service_image = 'gitlab.ub.uni-bielefeld.de:4567/sfb1288inf/{}:{}'.format(job.service, job.service_version) # noqa
try:
self.docker.services.create(service_image, **service_kwargs)
except docker.errors.APIError as e:
logging.error(
'Create "{}" service raised '.format(service_kwargs['name'])
+ '"docker.errors.APIError" The server returned an error. '
+ 'Details: {}'.format(e)
)
return
else:
job.status = 'queued'
patch_operation = {'op': 'replace', 'path': '/jobs/{}/status'.format(job.id), 'value': job.status} # noqa
self.buffer_user_patch_operation(job, patch_operation)
finally:
self.send_job_notification(job)
def checkout_job_service(self, job):
service_name = 'job_{}'.format(job.id)
try:
service = self.docker.services.get(service_name)
except docker.errors.NotFound:
logging.error('Get "{}" service raised '.format(service_name)
+ '"docker.errors.NotFound" The service does not exist. '
+ '(job.status: {} -> failed)'.format(job.status))
job.status = 'failed'
patch_operation = {'op': 'replace', 'path': '/jobs/{}/status'.format(job.id), 'value': job.status} # noqa
self.buffer_user_patch_operation(job, patch_operation)
except docker.errors.APIError as e:
logging.error(
'Get "{}" service raised '.format(service_name)
+ '"docker.errors.APIError" The server returned an error. '
+ 'Details: {}'.format(e)
)
return
except docker.errors.InvalidVersion:
logging.error(
'Get "{}" service raised '.format(service_name)
+ '"docker.errors.InvalidVersion" One of the arguments is '
+ 'not supported with the current API version.'
)
return
else:
service_tasks = service.tasks()
if not service_tasks:
return
task_state = service_tasks[0].get('Status').get('State')
if job.status == 'queued' and task_state != 'pending':
job.status = 'running'
patch_operation = {'op': 'replace', 'path': '/jobs/{}/status'.format(job.id), 'value': job.status} # noqa
self.buffer_user_patch_operation(job, patch_operation)
elif job.status == 'running' and task_state in ['complete', 'failed']:
try:
service.remove()
except docker.errors.APIError as e:
logging.error(
'Remove "{}" service raised '.format(service_name)
+ '"docker.errors.APIError" The server returned an error. ' # noqa
+ 'Details: {}'.format(e)
)
return
else:
if task_state == 'complete':
results_dir = os.path.join(job.path, 'output')
result_files = filter(lambda x: x.endswith('.zip'),
os.listdir(results_dir))
for result_file in result_files:
job_result = JobResult(filename=result_file, job=job) # noqa
db.session.add(job_result)
db.session.flush()
db.session.refresh(job_result)
patch_operation = {'op': 'add', 'path': '/jobs/{}/results/{}'.format(job.id, job_result.id), 'value': job_result.to_dict()} # noqa
self.buffer_user_patch_operation(job, patch_operation) # noqa
job.end_date = datetime.utcnow()
patch_operation = {'op': 'replace', 'path': '/jobs/{}/end_date'.format(job.id), 'value': job.end_date.timestamp()} # noqa
self.buffer_user_patch_operation(job, patch_operation)
job.status = task_state
patch_operation = {'op': 'replace', 'path': '/jobs/{}/status'.format(job.id), 'value': job.status} # noqa
self.buffer_user_patch_operation(job, patch_operation)
finally:
self.send_job_notification(job)
def remove_job_service(self, job):
service_name = 'job_{}'.format(job.id)
try:
service = self.docker.services.get(service_name)
except docker.errors.NotFound:
job.status = 'canceled'
patch_operation = {'op': 'replace', 'path': '/jobs/{}/status'.format(job.id), 'value': job.status} # noqa
self.buffer_user_patch_operation(job, patch_operation)
except docker.errors.APIError as e:
logging.error(
'Get "{}" service raised '.format(service_name)
+ '"docker.errors.APIError" The server returned an error. '
+ 'Details: {}'.format(e)
)
return
except docker.errors.InvalidVersion:
logging.error(
'Get "{}" service raised '.format(service_name)
+ '"docker.errors.InvalidVersion" One of the arguments is '
+ 'not supported with the current API version.'
)
return
else:
try:
service.update(mounts=None)
except docker.errors.APIError as e:
logging.error(
'Update "{}" service raised '.format(service_name)
+ '"docker.errors.APIError" The server returned an error. '
+ 'Details: {}'.format(e)
)
return
try:
service.remove()
except docker.errors.APIError as e:
logging.error(
'Remove "{}" service raised '.format(service_name)
+ '"docker.errors.APIError" The server returned an error. '
+ 'Details: {}'.format(e)
)
def send_job_notification(self, job):
if job.creator.setting_job_status_mail_notifications == 'none':
return
if (job.creator.setting_job_status_mail_notifications == 'end'
and job.status not in ['complete', 'failed']):
return
msg = create_message(job.creator.email,
'Status update for your Job "{}"'.format(job.title), # noqa
'tasks/email/notification', job=job)
mail.send(msg)