mirror of
https://gitlab.ub.uni-bielefeld.de/sfb1288inf/nopaque.git
synced 2025-11-04 04:12:45 +00:00
move blueprints in dedicated folder
This commit is contained in:
Patrick Jentsch
@@ -1,18 +1,2 @@
|
||||
from flask import Blueprint
|
||||
from flask_login import login_required
|
||||
|
||||
|
||||
bp = Blueprint('jobs', __name__)
|
||||
|
||||
|
||||
@bp.before_request
|
||||
@login_required
|
||||
def before_request():
|
||||
'''
|
||||
Ensures that the routes in this package can only be visited by users that
|
||||
are logged in.
|
||||
'''
|
||||
pass
|
||||
|
||||
|
||||
from . import routes, json_routes
|
||||
from .handle_corpora import job as handle_corpora
|
||||
from .handle_jobs import job as handle_jobs
|
||||
|
||||
@@ -1,138 +0,0 @@
|
||||
from flask import current_app
|
||||
from flask_login import current_user
|
||||
from flask_socketio import Namespace
|
||||
from app import db, hashids, socketio
|
||||
from app.extensions.flask_socketio import admin_required, login_required
|
||||
from app.models import Job, JobStatus
|
||||
|
||||
|
||||
class JobsNamespace(Namespace):
|
||||
@login_required
|
||||
def on_delete(self, job_hashid: str):
|
||||
# Decode the job hashid
|
||||
job_id = hashids.decode(job_hashid)
|
||||
|
||||
# Validate job_id
|
||||
if not isinstance(job_id, int):
|
||||
return {
|
||||
'code': 400,
|
||||
'body': 'job_id is invalid'
|
||||
}
|
||||
|
||||
# Load job from database
|
||||
job = Job.query.get(job_id)
|
||||
if job is None:
|
||||
return {
|
||||
'code': 404,
|
||||
'body': 'Job not found'
|
||||
}
|
||||
|
||||
# Check if the current user is allowed to delete the job
|
||||
if not (job.user == current_user or current_user.is_administrator):
|
||||
return {
|
||||
'code': 403,
|
||||
'body': 'Forbidden'
|
||||
}
|
||||
|
||||
# TODO: This should be a method in the Job model
|
||||
def _delete_job(app, job_id):
|
||||
with app.app_context():
|
||||
job = Job.query.get(job_id)
|
||||
job.delete()
|
||||
db.session.commit()
|
||||
|
||||
# Delete the job in a background task
|
||||
socketio.start_background_task(
|
||||
target=_delete_job,
|
||||
app=current_app._get_current_object(),
|
||||
job_id=job_id
|
||||
)
|
||||
|
||||
return {
|
||||
'code': 202,
|
||||
'body': f'Job "{job.title}" marked for deletion'
|
||||
}
|
||||
|
||||
|
||||
@admin_required
|
||||
def on_get_log(self, job_hashid: str):
|
||||
# Decode the job hashid
|
||||
job_id = hashids.decode(job_hashid)
|
||||
|
||||
# Validate job_id
|
||||
if not isinstance(job_id, int):
|
||||
return {
|
||||
'code': 400,
|
||||
'body': 'job_id is invalid'
|
||||
}
|
||||
|
||||
# Load job from database
|
||||
job = Job.query.get(job_id)
|
||||
if job is None:
|
||||
return {
|
||||
'code': 404,
|
||||
'body': 'Job not found'
|
||||
}
|
||||
|
||||
# Check if the job is already processed
|
||||
if job.status not in [JobStatus.COMPLETED, JobStatus.FAILED]:
|
||||
return {
|
||||
'code': 409,
|
||||
'body': 'Job is not done processing'
|
||||
}
|
||||
|
||||
# Read the log file
|
||||
with open(job.path / 'pipeline_data' / 'logs' / 'pyflow_log.txt') as log_file:
|
||||
job_log = log_file.read()
|
||||
|
||||
return {
|
||||
'code': 200,
|
||||
'body': job_log
|
||||
}
|
||||
|
||||
|
||||
@login_required
|
||||
def on_restart(self, job_hashid: str):
|
||||
# Decode the job hashid
|
||||
job_id = hashids.decode(job_hashid)
|
||||
|
||||
# Validate job_id
|
||||
if not isinstance(job_id, int):
|
||||
return {
|
||||
'code': 400,
|
||||
'body': 'job_id is invalid'
|
||||
}
|
||||
|
||||
# Load job from database
|
||||
job = Job.query.get(job_id)
|
||||
if job is None:
|
||||
return {
|
||||
'code': 404,
|
||||
'body': 'Job not found'
|
||||
}
|
||||
|
||||
# Check if the current user is allowed to restart the job
|
||||
if not (job.user == current_user or current_user.is_administrator):
|
||||
return {
|
||||
'code': 403,
|
||||
'body': 'Forbidden'
|
||||
}
|
||||
|
||||
# TODO: This should be a method in the Job model
|
||||
def _restart_job(app, job_id):
|
||||
with app.app_context():
|
||||
job = Job.query.get(job_id)
|
||||
job.restart()
|
||||
db.session.commit()
|
||||
|
||||
# Restart the job in a background task
|
||||
socketio.start_background_task(
|
||||
target=_restart_job,
|
||||
app=current_app._get_current_object(),
|
||||
job_id=job_id
|
||||
)
|
||||
|
||||
return {
|
||||
'code': 202,
|
||||
'body': f'Job "{job.title}" restarted'
|
||||
}
|
||||
230
app/jobs/handle_corpora.py
Normal file
230
app/jobs/handle_corpora.py
Normal file
@@ -0,0 +1,230 @@
|
||||
from app import db, docker_client, scheduler
|
||||
from app.models import Corpus, CorpusStatus
|
||||
from flask import current_app
|
||||
import docker
|
||||
import os
|
||||
import shutil
|
||||
|
||||
|
||||
def job():
|
||||
with scheduler.app.app_context():
|
||||
_handle_corpora()
|
||||
|
||||
def _handle_corpora():
|
||||
corpora = Corpus.query.all()
|
||||
for corpus in [x for x in corpora if x.status == CorpusStatus.SUBMITTED]:
|
||||
_create_build_corpus_service(corpus)
|
||||
for corpus in [x for x in corpora if x.status in [CorpusStatus.QUEUED, CorpusStatus.BUILDING]]:
|
||||
_checkout_build_corpus_service(corpus)
|
||||
for corpus in [x for x in corpora if x.status == CorpusStatus.BUILT and x.num_analysis_sessions > 0]:
|
||||
corpus.status = CorpusStatus.STARTING_ANALYSIS_SESSION
|
||||
for corpus in [x for x in corpora if x.status == CorpusStatus.RUNNING_ANALYSIS_SESSION and x.num_analysis_sessions == 0]:
|
||||
corpus.status = CorpusStatus.CANCELING_ANALYSIS_SESSION
|
||||
for corpus in [x for x in corpora if x.status == CorpusStatus.RUNNING_ANALYSIS_SESSION]:
|
||||
_checkout_analysing_corpus_container(corpus)
|
||||
for corpus in [x for x in corpora if x.status == CorpusStatus.STARTING_ANALYSIS_SESSION]:
|
||||
_create_cqpserver_container(corpus)
|
||||
for corpus in [x for x in corpora if x.status == CorpusStatus.CANCELING_ANALYSIS_SESSION]:
|
||||
_remove_cqpserver_container(corpus)
|
||||
db.session.commit()
|
||||
|
||||
def _create_build_corpus_service(corpus):
|
||||
''' # Docker service settings # '''
|
||||
''' ## Command ## '''
|
||||
command = ['bash', '-c']
|
||||
command.append(
|
||||
f'mkdir /corpora/data/nopaque-{corpus.hashid.lower()}'
|
||||
' && '
|
||||
'cwb-encode'
|
||||
' -c utf8'
|
||||
f' -d /corpora/data/nopaque-{corpus.hashid.lower()}'
|
||||
' -f /root/files/corpus.vrt'
|
||||
f' -R /usr/local/share/cwb/registry/nopaque-{corpus.hashid.lower()}'
|
||||
' -P pos -P lemma -P simple_pos'
|
||||
' -S ent:0+type -S s:0'
|
||||
' -S text:0+address+author+booktitle+chapter+editor+institution+journal+pages+publisher+publishing_year+school+title'
|
||||
' -xsB -9'
|
||||
' && '
|
||||
f'cwb-make -V NOPAQUE-{corpus.hashid.upper()}'
|
||||
)
|
||||
''' ## Constraints ## '''
|
||||
constraints = ['node.role==worker']
|
||||
''' ## Image ## '''
|
||||
image = f'{current_app.config["NOPAQUE_DOCKER_IMAGE_PREFIX"]}cwb:r1879'
|
||||
''' ## Labels ## '''
|
||||
labels = {
|
||||
'origin': current_app.config['SERVER_NAME'],
|
||||
'type': 'corpus.build',
|
||||
'corpus_id': str(corpus.id)
|
||||
}
|
||||
''' ## Mounts ## '''
|
||||
mounts = []
|
||||
''' ### Data mount ### '''
|
||||
data_mount_source = os.path.join(corpus.path, 'cwb', 'data')
|
||||
data_mount_target = '/corpora/data'
|
||||
data_mount = f'{data_mount_source}:{data_mount_target}:rw'
|
||||
# Make sure that their is no data in the data directory
|
||||
shutil.rmtree(data_mount_source, ignore_errors=True)
|
||||
os.makedirs(data_mount_source)
|
||||
mounts.append(data_mount)
|
||||
''' ### File mount ### '''
|
||||
file_mount_source = os.path.join(corpus.path, 'cwb', 'corpus.vrt')
|
||||
file_mount_target = '/root/files/corpus.vrt'
|
||||
file_mount = f'{file_mount_source}:{file_mount_target}:ro'
|
||||
mounts.append(file_mount)
|
||||
''' ### Registry mount ### '''
|
||||
registry_mount_source = os.path.join(corpus.path, 'cwb', 'registry')
|
||||
registry_mount_target = '/usr/local/share/cwb/registry'
|
||||
registry_mount = f'{registry_mount_source}:{registry_mount_target}:rw'
|
||||
# Make sure that their is no data in the registry directory
|
||||
shutil.rmtree(registry_mount_source, ignore_errors=True)
|
||||
os.makedirs(registry_mount_source)
|
||||
mounts.append(registry_mount)
|
||||
''' ## Name ## '''
|
||||
name = f'build-corpus_{corpus.id}'
|
||||
''' ## Restart policy ## '''
|
||||
restart_policy = docker.types.RestartPolicy()
|
||||
try:
|
||||
docker_client.services.create(
|
||||
image,
|
||||
command=command,
|
||||
constraints=constraints,
|
||||
labels=labels,
|
||||
mounts=mounts,
|
||||
name=name,
|
||||
restart_policy=restart_policy,
|
||||
user='0:0'
|
||||
)
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Create service "{name}" failed: {e}')
|
||||
return
|
||||
corpus.status = CorpusStatus.QUEUED
|
||||
|
||||
def _checkout_build_corpus_service(corpus):
|
||||
service_name = f'build-corpus_{corpus.id}'
|
||||
try:
|
||||
service = docker_client.services.get(service_name)
|
||||
except docker.errors.NotFound as e:
|
||||
current_app.logger.error(f'Get service "{service_name}" failed: {e}')
|
||||
corpus.status = CorpusStatus.FAILED
|
||||
return
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Get service "{service_name}" failed: {e}')
|
||||
service_tasks = service.tasks()
|
||||
if not service_tasks:
|
||||
return
|
||||
task_state = service_tasks[0].get('Status').get('State')
|
||||
if corpus.status == CorpusStatus.QUEUED and task_state != 'pending':
|
||||
corpus.status = CorpusStatus.BUILDING
|
||||
return
|
||||
elif corpus.status == CorpusStatus.BUILDING and task_state == 'complete':
|
||||
corpus.status = CorpusStatus.BUILT
|
||||
elif corpus.status == CorpusStatus.BUILDING and task_state == 'failed':
|
||||
corpus.status = CorpusStatus.FAILED
|
||||
else:
|
||||
return
|
||||
try:
|
||||
service.remove()
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Remove service "{service_name}" failed: {e}')
|
||||
|
||||
def _create_cqpserver_container(corpus):
|
||||
''' # Docker container settings # '''
|
||||
''' ## Command ## '''
|
||||
command = []
|
||||
command.append(
|
||||
'echo "host *;" > cqpserver.init'
|
||||
' && '
|
||||
'echo "user anonymous \\"\\";" >> cqpserver.init'
|
||||
' && '
|
||||
'cqpserver -I cqpserver.init'
|
||||
)
|
||||
''' ## Detach ## '''
|
||||
detach = True
|
||||
''' ## Entrypoint ## '''
|
||||
entrypoint = ['bash', '-c']
|
||||
''' ## Image ## '''
|
||||
image = f'{current_app.config["NOPAQUE_DOCKER_IMAGE_PREFIX"]}cwb:r1879'
|
||||
''' ## Name ## '''
|
||||
name = f'cqpserver_{corpus.id}'
|
||||
''' ## Network ## '''
|
||||
network = f'{current_app.config["NOPAQUE_DOCKER_NETWORK_NAME"]}'
|
||||
''' ## Volumes ## '''
|
||||
volumes = []
|
||||
''' ### Corpus data volume ### '''
|
||||
data_volume_source = os.path.join(corpus.path, 'cwb', 'data')
|
||||
data_volume_target = '/corpora/data'
|
||||
# data_volume_source = os.path.join(corpus.path, 'cwb', 'data', f'nopaque_{corpus.id}')
|
||||
# data_volume_target = f'/corpora/data/nopaque_{corpus.hashid.lower()}'
|
||||
data_volume = f'{data_volume_source}:{data_volume_target}:rw'
|
||||
volumes.append(data_volume)
|
||||
''' ### Corpus registry volume ### '''
|
||||
registry_volume_source = os.path.join(corpus.path, 'cwb', 'registry')
|
||||
registry_volume_target = '/usr/local/share/cwb/registry'
|
||||
# registry_volume_source = os.path.join(corpus.path, 'cwb', 'registry', f'nopaque_{corpus.id}')
|
||||
# registry_volume_target = f'/usr/local/share/cwb/registry/nopaque_{corpus.hashid.lower()}'
|
||||
registry_volume = f'{registry_volume_source}:{registry_volume_target}:rw'
|
||||
volumes.append(registry_volume)
|
||||
# Check if a cqpserver container already exists. If this is the case,
|
||||
# remove it and create a new one
|
||||
try:
|
||||
container = docker_client.containers.get(name)
|
||||
except docker.errors.NotFound:
|
||||
pass
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Get container "{name}" failed: {e}')
|
||||
return
|
||||
else:
|
||||
try:
|
||||
container.remove(force=True)
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Remove container "{name}" failed: {e}')
|
||||
return
|
||||
try:
|
||||
docker_client.containers.run(
|
||||
image,
|
||||
command=command,
|
||||
detach=detach,
|
||||
entrypoint=entrypoint,
|
||||
name=name,
|
||||
network=network,
|
||||
user='0:0',
|
||||
volumes=volumes
|
||||
)
|
||||
except docker.errors.ImageNotFound as e:
|
||||
current_app.logger.error(
|
||||
f'Run container "{name}" failed '
|
||||
f'due to "docker.errors.ImageNotFound" error: {e}'
|
||||
)
|
||||
corpus.status = CorpusStatus.FAILED
|
||||
return
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Run container "{name}" failed: {e}')
|
||||
return
|
||||
corpus.status = CorpusStatus.RUNNING_ANALYSIS_SESSION
|
||||
|
||||
def _checkout_analysing_corpus_container(corpus):
|
||||
container_name = f'cqpserver_{corpus.id}'
|
||||
try:
|
||||
docker_client.containers.get(container_name)
|
||||
except docker.errors.NotFound as e:
|
||||
current_app.logger.error(f'Get container "{container_name}" failed: {e}')
|
||||
corpus.num_analysis_sessions = 0
|
||||
corpus.status = CorpusStatus.BUILT
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Get container "{container_name}" failed: {e}')
|
||||
|
||||
def _remove_cqpserver_container(corpus):
|
||||
container_name = f'cqpserver_{corpus.id}'
|
||||
try:
|
||||
container = docker_client.containers.get(container_name)
|
||||
except docker.errors.NotFound:
|
||||
corpus.status = CorpusStatus.BUILT
|
||||
return
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Get container "{container_name}" failed: {e}')
|
||||
return
|
||||
try:
|
||||
container.remove(force=True)
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Remove container "{container_name}" failed: {e}')
|
||||
239
app/jobs/handle_jobs.py
Normal file
239
app/jobs/handle_jobs.py
Normal file
@@ -0,0 +1,239 @@
|
||||
from app import db, docker_client, hashids, scheduler
|
||||
from app.models import (
|
||||
Job,
|
||||
JobResult,
|
||||
JobStatus,
|
||||
TesseractOCRPipelineModel,
|
||||
SpaCyNLPPipelineModel
|
||||
)
|
||||
from datetime import datetime
|
||||
from flask import current_app
|
||||
from werkzeug.utils import secure_filename
|
||||
import docker
|
||||
import json
|
||||
import os
|
||||
import shutil
|
||||
|
||||
|
||||
def job():
|
||||
with scheduler.app.app_context():
|
||||
_handle_jobs()
|
||||
|
||||
def _handle_jobs():
|
||||
jobs = Job.query.all()
|
||||
for job in [x for x in jobs if x.status == JobStatus.SUBMITTED]:
|
||||
_create_job_service(job)
|
||||
for job in [x for x in jobs if x.status in [JobStatus.QUEUED, JobStatus.RUNNING]]:
|
||||
_checkout_job_service(job)
|
||||
for job in [x for x in jobs if x.status == JobStatus.CANCELING]:
|
||||
_remove_job_service(job)
|
||||
db.session.commit()
|
||||
|
||||
def _create_job_service(job):
|
||||
''' # Docker service settings # '''
|
||||
''' ## Service specific settings ## '''
|
||||
if job.service == 'file-setup-pipeline':
|
||||
mem_mb = 512
|
||||
n_cores = 2
|
||||
executable = 'file-setup-pipeline'
|
||||
image = f'{current_app.config["NOPAQUE_DOCKER_IMAGE_PREFIX"]}file-setup-pipeline:v{job.service_version}'
|
||||
elif job.service == 'tesseract-ocr-pipeline':
|
||||
mem_mb = 1024
|
||||
n_cores = 4
|
||||
executable = 'tesseract-ocr-pipeline'
|
||||
image = f'{current_app.config["NOPAQUE_DOCKER_IMAGE_PREFIX"]}tesseract-ocr-pipeline:v{job.service_version}'
|
||||
elif job.service == 'transkribus-htr-pipeline':
|
||||
mem_mb = 1024
|
||||
n_cores = 4
|
||||
executable = 'transkribus-htr-pipeline'
|
||||
image = f'{current_app.config["NOPAQUE_DOCKER_IMAGE_PREFIX"]}transkribus-htr-pipeline:v{job.service_version}'
|
||||
elif job.service == 'spacy-nlp-pipeline':
|
||||
mem_mb = 1024
|
||||
n_cores = 1
|
||||
executable = 'spacy-nlp-pipeline'
|
||||
image = f'{current_app.config["NOPAQUE_DOCKER_IMAGE_PREFIX"]}spacy-nlp-pipeline:v{job.service_version}'
|
||||
''' ## Command ## '''
|
||||
command = f'{executable} -i /input -o /output'
|
||||
command += ' --log-dir /logs'
|
||||
command += f' --mem-mb {mem_mb}'
|
||||
command += f' --n-cores {n_cores}'
|
||||
if job.service == 'spacy-nlp-pipeline':
|
||||
model_id = hashids.decode(job.service_args['model'])
|
||||
model = SpaCyNLPPipelineModel.query.get(model_id)
|
||||
if model is None:
|
||||
job.status = JobStatus.FAILED
|
||||
return
|
||||
command += f' -m {model.pipeline_name}'
|
||||
if 'encoding_detection' in job.service_args and job.service_args['encoding_detection']:
|
||||
command += ' --check-encoding'
|
||||
elif job.service == 'tesseract-ocr-pipeline':
|
||||
command += f' -m {job.service_args["model"]}'
|
||||
if 'binarization' in job.service_args and job.service_args['binarization']:
|
||||
command += ' --binarize'
|
||||
if 'ocropus_nlbin_threshold' in job.service_args and job.service_args['ocropus_nlbin_threshold']:
|
||||
value = job.service_args['ocropus_nlbin_threshold']
|
||||
command += f' --ocropus-nlbin-threshold {value}'
|
||||
elif job.service == 'transkribus-htr-pipeline':
|
||||
transkribus_htr_pipeline_model_id = job.service_args['model']
|
||||
command += f' -m {transkribus_htr_pipeline_model_id}'
|
||||
readcoop_username = current_app.config.get('NOPAQUE_READCOOP_USERNAME')
|
||||
command += f' --readcoop-username "{readcoop_username}"'
|
||||
readcoop_password = current_app.config.get('NOPAQUE_READCOOP_PASSWORD')
|
||||
command += f' --readcoop-password "{readcoop_password}"'
|
||||
if 'binarization' in job.service_args and job.service_args['binarization']:
|
||||
command += ' --binarize'
|
||||
''' ## Constraints ## '''
|
||||
constraints = ['node.role==worker']
|
||||
''' ## Labels ## '''
|
||||
labels = {
|
||||
'origin': current_app.config['SERVER_NAME'],
|
||||
'type': 'job',
|
||||
'job_id': str(job.id)
|
||||
}
|
||||
''' ## Mounts ## '''
|
||||
mounts = []
|
||||
''' ### Input mount(s) ### '''
|
||||
input_mount_target_base = '/input'
|
||||
if job.service == 'file-setup-pipeline':
|
||||
input_mount_target_base += f'/{secure_filename(job.title)}'
|
||||
for job_input in job.inputs:
|
||||
input_mount_source = job_input.path
|
||||
input_mount_target = f'{input_mount_target_base}/{job_input.filename}'
|
||||
input_mount = f'{input_mount_source}:{input_mount_target}:ro'
|
||||
mounts.append(input_mount)
|
||||
if job.service == 'tesseract-ocr-pipeline':
|
||||
if isinstance(job.service_args['model'], str):
|
||||
model_id = hashids.decode(job.service_args['model'])
|
||||
elif isinstance(job.service_args['model'], int):
|
||||
model_id = job.service_args['model']
|
||||
else:
|
||||
job.status = JobStatus.FAILED
|
||||
return
|
||||
model = TesseractOCRPipelineModel.query.get(model_id)
|
||||
if model is None:
|
||||
job.status = JobStatus.FAILED
|
||||
return
|
||||
models_mount_source = model.path
|
||||
models_mount_target = f'/usr/local/share/tessdata/{model.id}.traineddata'
|
||||
models_mount = f'{models_mount_source}:{models_mount_target}:ro'
|
||||
mounts.append(models_mount)
|
||||
elif job.service == 'spacy-nlp-pipeline':
|
||||
model_id = hashids.decode(job.service_args['model'])
|
||||
model = SpaCyNLPPipelineModel.query.get(model_id)
|
||||
if model is None:
|
||||
job.status = JobStatus.FAILED
|
||||
return
|
||||
models_mount_source = model.path
|
||||
models_mount_target = f'/usr/local/share/spacy/models/{model.filename}'
|
||||
models_mount = f'{models_mount_source}:{models_mount_target}:ro'
|
||||
mounts.append(models_mount)
|
||||
''' ### Output mount ### '''
|
||||
output_mount_source = os.path.join(job.path, 'results')
|
||||
output_mount_target = '/output'
|
||||
output_mount = f'{output_mount_source}:{output_mount_target}:rw'
|
||||
# Make sure that their is no data in the output directory
|
||||
shutil.rmtree(output_mount_source, ignore_errors=True)
|
||||
os.makedirs(output_mount_source)
|
||||
mounts.append(output_mount)
|
||||
''' ### Pipeline data mount ### '''
|
||||
pyflow_data_mount_source = os.path.join(job.path, 'pipeline_data')
|
||||
pyflow_data_mount_target = '/logs/pyflow.data'
|
||||
pyflow_data_mount = f'{pyflow_data_mount_source}:{pyflow_data_mount_target}:rw'
|
||||
# Make sure that their is no data in the output directory
|
||||
shutil.rmtree(pyflow_data_mount_source, ignore_errors=True)
|
||||
os.makedirs(pyflow_data_mount_source)
|
||||
mounts.append(pyflow_data_mount)
|
||||
''' ## Name ## '''
|
||||
name = f'job_{job.id}'
|
||||
''' ## Resources ## '''
|
||||
resources = docker.types.Resources(
|
||||
cpu_reservation=n_cores * (10 ** 9),
|
||||
mem_reservation=mem_mb * (10 ** 6)
|
||||
)
|
||||
''' ## Restart policy ## '''
|
||||
restart_policy = docker.types.RestartPolicy()
|
||||
try:
|
||||
docker_client.services.create(
|
||||
image,
|
||||
command=command,
|
||||
constraints=constraints,
|
||||
labels=labels,
|
||||
mounts=mounts,
|
||||
name=name,
|
||||
resources=resources,
|
||||
restart_policy=restart_policy,
|
||||
user='0:0'
|
||||
)
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Create service "{name}" failed: {e}')
|
||||
return
|
||||
job.status = JobStatus.QUEUED
|
||||
|
||||
def _checkout_job_service(job):
|
||||
service_name = f'job_{job.id}'
|
||||
try:
|
||||
service = docker_client.services.get(service_name)
|
||||
except docker.errors.NotFound as e:
|
||||
current_app.logger.error(f'Get service "{service_name}" failed: {e}')
|
||||
job.status = JobStatus.FAILED
|
||||
return
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Get service "{service_name}" failed: {e}')
|
||||
return
|
||||
service_tasks = service.tasks()
|
||||
if not service_tasks:
|
||||
return
|
||||
task_state = service_tasks[0].get('Status').get('State')
|
||||
if job.status == JobStatus.QUEUED and task_state != 'pending':
|
||||
job.status = JobStatus.RUNNING
|
||||
return
|
||||
elif job.status == JobStatus.RUNNING and task_state == 'complete':
|
||||
job.status = JobStatus.COMPLETED
|
||||
results_dir = os.path.join(job.path, 'results')
|
||||
with open(os.path.join(results_dir, 'outputs.json')) as f:
|
||||
outputs = json.load(f)
|
||||
for output in outputs:
|
||||
filename = os.path.basename(output['file'])
|
||||
job_result = JobResult(
|
||||
filename=filename,
|
||||
job=job,
|
||||
mimetype=output['mimetype']
|
||||
)
|
||||
if 'description' in output:
|
||||
job_result.description = output['description']
|
||||
db.session.add(job_result)
|
||||
db.session.flush(objects=[job_result])
|
||||
db.session.refresh(job_result)
|
||||
os.rename(
|
||||
os.path.join(results_dir, output['file']),
|
||||
job_result.path
|
||||
)
|
||||
elif job.status == JobStatus.RUNNING and task_state == 'failed':
|
||||
job.status = JobStatus.FAILED
|
||||
else:
|
||||
return
|
||||
job.end_date = datetime.utcnow()
|
||||
try:
|
||||
service.remove()
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Remove service "{service_name}" failed: {e}')
|
||||
|
||||
def _remove_job_service(job):
|
||||
service_name = f'job_{job.id}'
|
||||
try:
|
||||
service = docker_client.services.get(service_name)
|
||||
except docker.errors.NotFound:
|
||||
job.status = JobStatus.CANCELED
|
||||
return
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Get service "{service_name}" failed: {e}')
|
||||
return
|
||||
try:
|
||||
service.update(mounts=None)
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Update service "{service_name}" failed: {e}')
|
||||
return
|
||||
try:
|
||||
service.remove()
|
||||
except docker.errors.DockerException as e:
|
||||
current_app.logger.error(f'Remove "{service_name}" service failed: {e}')
|
||||
@@ -1,72 +0,0 @@
|
||||
from flask import abort, current_app
|
||||
from flask_login import current_user
|
||||
from threading import Thread
|
||||
from app import db
|
||||
from app.decorators import admin_required, content_negotiation
|
||||
from app.models import Job, JobStatus
|
||||
from . import bp
|
||||
|
||||
|
||||
@bp.route('/<hashid:job_id>', methods=['DELETE'])
|
||||
@content_negotiation(produces='application/json')
|
||||
def delete_job(job_id):
|
||||
def _delete_job(app, job_id):
|
||||
with app.app_context():
|
||||
job = Job.query.get(job_id)
|
||||
job.delete()
|
||||
db.session.commit()
|
||||
|
||||
job = Job.query.get_or_404(job_id)
|
||||
if not (job.user == current_user or current_user.is_administrator):
|
||||
abort(403)
|
||||
thread = Thread(
|
||||
target=_delete_job,
|
||||
args=(current_app._get_current_object(), job_id)
|
||||
)
|
||||
thread.start()
|
||||
response_data = {
|
||||
'message': f'Job "{job.title}" marked for deletion'
|
||||
}
|
||||
return response_data, 202
|
||||
|
||||
|
||||
@bp.route('/<hashid:job_id>/log')
|
||||
@admin_required
|
||||
@content_negotiation(produces='application/json')
|
||||
def job_log(job_id):
|
||||
job = Job.query.get_or_404(job_id)
|
||||
if job.status not in [JobStatus.COMPLETED, JobStatus.FAILED]:
|
||||
response = {'errors': {'message': 'Job status is not completed or failed'}}
|
||||
return response, 409
|
||||
with open(job.path / 'pipeline_data' / 'logs' / 'pyflow_log.txt') as log_file:
|
||||
log = log_file.read()
|
||||
response_data = {
|
||||
'jobLog': log
|
||||
}
|
||||
return response_data, 200
|
||||
|
||||
|
||||
@bp.route('/<hashid:job_id>/restart', methods=['POST'])
|
||||
@content_negotiation(produces='application/json')
|
||||
def restart_job(job_id):
|
||||
def _restart_job(app, job_id):
|
||||
with app.app_context():
|
||||
job = Job.query.get(job_id)
|
||||
job.restart()
|
||||
db.session.commit()
|
||||
|
||||
job = Job.query.get_or_404(job_id)
|
||||
if not (job.user == current_user or current_user.is_administrator):
|
||||
abort(403)
|
||||
if job.status == JobStatus.FAILED:
|
||||
response = {'errors': {'message': 'Job status is not "failed"'}}
|
||||
return response, 409
|
||||
thread = Thread(
|
||||
target=_restart_job,
|
||||
args=(current_app._get_current_object(), job_id)
|
||||
)
|
||||
thread.start()
|
||||
response_data = {
|
||||
'message': f'Job "{job.title}" marked for restarting'
|
||||
}
|
||||
return response_data, 202
|
||||
@@ -1,55 +0,0 @@
|
||||
from flask import (
|
||||
abort,
|
||||
redirect,
|
||||
render_template,
|
||||
send_from_directory,
|
||||
url_for
|
||||
)
|
||||
from flask_login import current_user
|
||||
from app.models import Job, JobInput, JobResult
|
||||
from . import bp
|
||||
|
||||
|
||||
@bp.route('')
|
||||
def jobs():
|
||||
return redirect(url_for('main.dashboard', _anchor='jobs'))
|
||||
|
||||
|
||||
@bp.route('/<hashid:job_id>')
|
||||
def job(job_id):
|
||||
job = Job.query.get_or_404(job_id)
|
||||
if not (job.user == current_user or current_user.is_administrator):
|
||||
abort(403)
|
||||
return render_template(
|
||||
'jobs/job.html.j2',
|
||||
title='Job',
|
||||
job=job
|
||||
)
|
||||
|
||||
|
||||
@bp.route('/<hashid:job_id>/inputs/<hashid:job_input_id>/download')
|
||||
def download_job_input(job_id, job_input_id):
|
||||
job_input = JobInput.query.filter_by(job_id=job_id, id=job_input_id).first_or_404()
|
||||
if not (job_input.job.user == current_user or current_user.is_administrator):
|
||||
abort(403)
|
||||
return send_from_directory(
|
||||
job_input.path.parent,
|
||||
job_input.path.name,
|
||||
as_attachment=True,
|
||||
download_name=job_input.filename,
|
||||
mimetype=job_input.mimetype
|
||||
)
|
||||
|
||||
|
||||
@bp.route('/<hashid:job_id>/results/<hashid:job_result_id>/download')
|
||||
def download_job_result(job_id, job_result_id):
|
||||
job_result = JobResult.query.filter_by(job_id=job_id, id=job_result_id).first_or_404()
|
||||
if not (job_result.job.user == current_user or current_user.is_administrator):
|
||||
abort(403)
|
||||
return send_from_directory(
|
||||
job_result.path.parent,
|
||||
job_result.path.name,
|
||||
as_attachment=True,
|
||||
download_name=job_result.filename,
|
||||
mimetype=job_result.mimetype
|
||||
)
|
||||
Reference in New Issue
Block a user