mirror of
				https://gitlab.ub.uni-bielefeld.de/sfb1288inf/nopaque.git
				synced 2025-11-03 20:02:47 +00:00 
			
		
		
		
	Big Corpus analysis update
This commit is contained in:
		@@ -1,4 +1,5 @@
 | 
			
		||||
from app import db
 | 
			
		||||
from flask import current_app
 | 
			
		||||
from time import sleep
 | 
			
		||||
from .corpus_utils import CheckCorporaMixin
 | 
			
		||||
from .job_utils import CheckJobsMixin
 | 
			
		||||
@@ -8,6 +9,11 @@ import docker
 | 
			
		||||
class Daemon(CheckCorporaMixin, CheckJobsMixin):
 | 
			
		||||
    def __init__(self):
 | 
			
		||||
        self.docker = docker.from_env()
 | 
			
		||||
        self.docker.login(
 | 
			
		||||
            username=current_app.config['NOPAQUE_DOCKER_REGISTRY_USERNAME'],
 | 
			
		||||
            password=current_app.config['NOPAQUE_DOCKER_REGISTRY_PASSWORD'],
 | 
			
		||||
            registry=current_app.config['NOPAQUE_DOCKER_REGISTRY']
 | 
			
		||||
        )
 | 
			
		||||
 | 
			
		||||
    def run(self):
 | 
			
		||||
        while True:
 | 
			
		||||
 
 | 
			
		||||
@@ -8,21 +8,19 @@ 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:
 | 
			
		||||
        for corpus in (x for x in corpora if x.status == 'submitted'):
 | 
			
		||||
            self.create_build_corpus_service(corpus)
 | 
			
		||||
        for corpus in queued_corpora + running_corpora:
 | 
			
		||||
        for corpus in (x for x in corpora if x.status == 'queued' or x.status == 'running'):  # noqa
 | 
			
		||||
            self.checkout_build_corpus_service(corpus)
 | 
			
		||||
        for corpus in start_analysis_corpora:
 | 
			
		||||
            self.create_cqpserver_container(corpus)
 | 
			
		||||
        for corpus in analysing_corpora:
 | 
			
		||||
        for corpus in (x for x in corpora if x.status == 'prepared' and x.num_analysis_sessions > 0):  # noqa
 | 
			
		||||
            corpus.status = 'start analysis'
 | 
			
		||||
        for corpus in (x for x in corpora if x.status == 'analysing' and x.num_analysis_sessions == 0):  # noqa
 | 
			
		||||
            corpus.status = 'stop analysis'
 | 
			
		||||
        for corpus in (x for x in corpora if x.status == 'analysing'):
 | 
			
		||||
            self.checkout_analysing_corpus_container(corpus)
 | 
			
		||||
        for corpus in stop_analysis_corpora:
 | 
			
		||||
        for corpus in (x for x in corpora if x.status == 'start analysis'):
 | 
			
		||||
            self.create_cqpserver_container(corpus)
 | 
			
		||||
        for corpus in (x for x in corpora if x.status == 'stop analysis'):
 | 
			
		||||
            self.remove_cqpserver_container(corpus)
 | 
			
		||||
 | 
			
		||||
    def create_build_corpus_service(self, corpus):
 | 
			
		||||
@@ -32,7 +30,7 @@ class CheckCorporaMixin:
 | 
			
		||||
        ''' ## Constraints ## '''
 | 
			
		||||
        constraints = ['node.role==worker']
 | 
			
		||||
        ''' ## Image ## '''
 | 
			
		||||
        image = current_app.config['DOCKER_IMAGE_PREFIX'] + 'cqpserver:latest'
 | 
			
		||||
        image = f'{current_app.config["NOPAQUE_DOCKER_IMAGE_PREFIX"]}cqpserver:r1674'  # noqa
 | 
			
		||||
        ''' ## Labels ## '''
 | 
			
		||||
        labels = {
 | 
			
		||||
            'origin': current_app.config['SERVER_NAME'],
 | 
			
		||||
@@ -43,27 +41,24 @@ class CheckCorporaMixin:
 | 
			
		||||
        ''' ### Corpus file mount ### '''
 | 
			
		||||
        corpus_file_source = os.path.join(corpus.path, 'merged', 'corpus.vrt')
 | 
			
		||||
        corpus_file_target = '/root/files/corpus.vrt'
 | 
			
		||||
        corpus_file_mount = \
 | 
			
		||||
            corpus_file_source + ':' + corpus_file_target + ':ro'
 | 
			
		||||
        corpus_file_mount = f'{corpus_file_source}:{corpus_file_target}:ro'
 | 
			
		||||
        ''' ### Corpus data mount ### '''
 | 
			
		||||
        corpus_data_source = os.path.join(corpus.path, 'data')
 | 
			
		||||
        corpus_data_target = '/corpora/data'
 | 
			
		||||
        corpus_data_mount = \
 | 
			
		||||
            corpus_data_source + ':' + corpus_data_target + ':rw'
 | 
			
		||||
        corpus_data_mount = f'{corpus_data_source}:{corpus_data_target}:rw'
 | 
			
		||||
        # Make sure that their is no data in the corpus data directory
 | 
			
		||||
        shutil.rmtree(corpus_data_source, ignore_errors=True)
 | 
			
		||||
        os.mkdir(corpus_data_source)
 | 
			
		||||
        ''' ### Corpus registry mount ### '''
 | 
			
		||||
        corpus_registry_source = os.path.join(corpus.path, 'registry')
 | 
			
		||||
        corpus_registry_target = '/usr/local/share/cwb/registry'
 | 
			
		||||
        corpus_registry_mount = \
 | 
			
		||||
            corpus_registry_source + ':' + corpus_registry_target + ':rw'
 | 
			
		||||
        corpus_registry_mount = f'{corpus_registry_source}:{corpus_registry_target}:rw'  # noqa
 | 
			
		||||
        # Make sure that their is no data in the corpus registry directory
 | 
			
		||||
        shutil.rmtree(corpus_registry_source, ignore_errors=True)
 | 
			
		||||
        os.mkdir(corpus_registry_source)
 | 
			
		||||
        mounts = [corpus_file_mount, corpus_data_mount, corpus_registry_mount]
 | 
			
		||||
        ''' ## Name ## '''
 | 
			
		||||
        name = 'build-corpus_{}'.format(corpus.id)
 | 
			
		||||
        name = f'build-corpus_{corpus.id}'
 | 
			
		||||
        ''' ## Restart policy ## '''
 | 
			
		||||
        restart_policy = docker.types.RestartPolicy()
 | 
			
		||||
        try:
 | 
			
		||||
@@ -78,57 +73,48 @@ class CheckCorporaMixin:
 | 
			
		||||
            )
 | 
			
		||||
        except docker.errors.APIError as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Create "{}" service raised '.format(name)
 | 
			
		||||
                + '"docker.errors.APIError" The server returned an error. '
 | 
			
		||||
                + 'Details: {}'.format(e)
 | 
			
		||||
                f'Create service "{name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.APIError": {e}'
 | 
			
		||||
            )
 | 
			
		||||
        else:
 | 
			
		||||
            corpus.status = 'queued'
 | 
			
		||||
            return
 | 
			
		||||
        corpus.status = 'queued'
 | 
			
		||||
 | 
			
		||||
    def checkout_build_corpus_service(self, corpus):
 | 
			
		||||
        service_name = 'build-corpus_{}'.format(corpus.id)
 | 
			
		||||
        service_name = f'build-corpus_{corpus.id}'
 | 
			
		||||
        try:
 | 
			
		||||
            service = self.docker.services.get(service_name)
 | 
			
		||||
        except docker.errors.NotFound:
 | 
			
		||||
        except docker.errors.NotFound as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Get "{}" service raised '.format(service_name)
 | 
			
		||||
                + '"docker.errors.NotFound" The service does not exist. '
 | 
			
		||||
                + '(corpus.status: {} -> failed)'.format(corpus.status)
 | 
			
		||||
                f'Get service "{service_name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.NotFound": {e}'
 | 
			
		||||
            )
 | 
			
		||||
            corpus.status = 'failed'
 | 
			
		||||
            return
 | 
			
		||||
        except docker.errors.APIError as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Get "{}" service raised '.format(service_name)
 | 
			
		||||
                + '"docker.errors.APIError" The server returned an error. '
 | 
			
		||||
                + 'Details: {}'.format(e)
 | 
			
		||||
            )
 | 
			
		||||
        except docker.errors.InvalidVersion:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Get "{}" service raised '.format(service_name)
 | 
			
		||||
                + '"docker.errors.InvalidVersion" One of the arguments is '
 | 
			
		||||
                + 'not supported with the current API version.'
 | 
			
		||||
                f'Get service "{service_name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.APIError": {e}'
 | 
			
		||||
            )
 | 
			
		||||
        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'
 | 
			
		||||
            return
 | 
			
		||||
        elif corpus.status == 'running' and task_state == 'complete':
 | 
			
		||||
            corpus.status = 'prepared'
 | 
			
		||||
        elif corpus.status == 'running' and task_state == 'failed':
 | 
			
		||||
            corpus.status = 'failed'
 | 
			
		||||
        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'
 | 
			
		||||
            elif (corpus.status == 'running'
 | 
			
		||||
                  and task_state in ['complete', 'failed']):
 | 
			
		||||
                try:
 | 
			
		||||
                    service.remove()
 | 
			
		||||
                except docker.errors.APIError as e:
 | 
			
		||||
                    current_app.logger.error(
 | 
			
		||||
                        'Remove "{}" service raised '.format(service_name)
 | 
			
		||||
                        + '"docker.errors.APIError" The server returned an error. '  # noqa
 | 
			
		||||
                        + 'Details: {}'.format(e)
 | 
			
		||||
                    )
 | 
			
		||||
                    return
 | 
			
		||||
                else:
 | 
			
		||||
                    corpus.status = \
 | 
			
		||||
                        'prepared' if task_state == 'complete' else 'failed'
 | 
			
		||||
            return
 | 
			
		||||
        try:
 | 
			
		||||
            service.remove()
 | 
			
		||||
        except docker.errors.APIError as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                f'Remove service "{service_name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.APIError": {e}'
 | 
			
		||||
            )
 | 
			
		||||
 | 
			
		||||
    def create_cqpserver_container(self, corpus):
 | 
			
		||||
        ''' # Docker container settings # '''
 | 
			
		||||
@@ -137,22 +123,20 @@ class CheckCorporaMixin:
 | 
			
		||||
        ''' ## Detach ## '''
 | 
			
		||||
        detach = True
 | 
			
		||||
        ''' ## Image ## '''
 | 
			
		||||
        image = current_app.config['DOCKER_IMAGE_PREFIX'] + 'cqpserver:latest'
 | 
			
		||||
        image = f'{current_app.config["NOPAQUE_DOCKER_IMAGE_PREFIX"]}cqpserver:r1674'  # noqa
 | 
			
		||||
        ''' ## Name ## '''
 | 
			
		||||
        name = 'cqpserver_{}'.format(corpus.id)
 | 
			
		||||
        name = f'cqpserver_{corpus.id}'
 | 
			
		||||
        ''' ## Network ## '''
 | 
			
		||||
        network = 'nopaque_default'
 | 
			
		||||
        ''' ## Volumes ## '''
 | 
			
		||||
        ''' ### Corpus data volume ### '''
 | 
			
		||||
        corpus_data_source = os.path.join(corpus.path, 'data')
 | 
			
		||||
        corpus_data_target = '/corpora/data'
 | 
			
		||||
        corpus_data_volume = \
 | 
			
		||||
            corpus_data_source + ':' + corpus_data_target + ':rw'
 | 
			
		||||
        corpus_data_volume = f'{corpus_data_source}:{corpus_data_target}:rw'
 | 
			
		||||
        ''' ### Corpus registry volume ### '''
 | 
			
		||||
        corpus_registry_source = os.path.join(corpus.path, 'registry')
 | 
			
		||||
        corpus_registry_target = '/usr/local/share/cwb/registry'
 | 
			
		||||
        corpus_registry_volume = \
 | 
			
		||||
            corpus_registry_source + ':' + corpus_registry_target + ':rw'
 | 
			
		||||
        corpus_registry_volume = f'{corpus_registry_source}:{corpus_registry_target}:rw'  # noqa
 | 
			
		||||
        volumes = [corpus_data_volume, corpus_registry_volume]
 | 
			
		||||
        # Check if a cqpserver container already exists. If this is the case,
 | 
			
		||||
        # remove it and create a new one
 | 
			
		||||
@@ -162,9 +146,8 @@ class CheckCorporaMixin:
 | 
			
		||||
            pass
 | 
			
		||||
        except docker.errors.APIError as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Get "{}" container raised '.format(name)
 | 
			
		||||
                + '"docker.errors.APIError" The server returned an error. '
 | 
			
		||||
                + 'Details: {}'.format(e)
 | 
			
		||||
                f'Get container "{name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.APIError": {e}'
 | 
			
		||||
            )
 | 
			
		||||
            return
 | 
			
		||||
        else:
 | 
			
		||||
@@ -172,77 +155,68 @@ class CheckCorporaMixin:
 | 
			
		||||
                container.remove(force=True)
 | 
			
		||||
            except docker.errors.APIError as e:
 | 
			
		||||
                current_app.logger.error(
 | 
			
		||||
                    'Remove "{}" container raised '.format(name)
 | 
			
		||||
                    + '"docker.errors.APIError" The server returned an error. '
 | 
			
		||||
                    + 'Details: {}'.format(e)
 | 
			
		||||
                    f'Remove container "{name}" failed '
 | 
			
		||||
                    + f'due to "docker.errors.APIError": {e}'
 | 
			
		||||
                )
 | 
			
		||||
                return
 | 
			
		||||
        try:
 | 
			
		||||
            self.docker.containers.run(image, command=command, detach=detach,
 | 
			
		||||
                                       volumes=volumes, name=name,
 | 
			
		||||
                                       network=network)
 | 
			
		||||
        except docker.errors.ContainerError:
 | 
			
		||||
            # This case should not occur, because detach is True.
 | 
			
		||||
            self.docker.containers.run(
 | 
			
		||||
                image,
 | 
			
		||||
                command=command,
 | 
			
		||||
                detach=detach,
 | 
			
		||||
                volumes=volumes,
 | 
			
		||||
                name=name,
 | 
			
		||||
                network=network
 | 
			
		||||
            )
 | 
			
		||||
        except docker.errors.ImageNotFound as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Run "{}" container raised '.format(name)
 | 
			
		||||
                + '"docker.errors.ContainerError" The container exits with a '
 | 
			
		||||
                + 'non-zero exit code and detach is False.'
 | 
			
		||||
            )
 | 
			
		||||
            corpus.status = 'failed'
 | 
			
		||||
        except docker.errors.ImageNotFound:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Run "{}" container raised '.format(name)
 | 
			
		||||
                + '"docker.errors.ImageNotFound" The specified image does not '
 | 
			
		||||
                + 'exist.'
 | 
			
		||||
                f'Run container "{name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.ImageNotFound" error: {e}'
 | 
			
		||||
            )
 | 
			
		||||
            corpus.status = 'failed'
 | 
			
		||||
            return
 | 
			
		||||
        except docker.errors.APIError as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Run "{}" container raised '.format(name)
 | 
			
		||||
                + '"docker.errors.APIError" The server returned an error. '
 | 
			
		||||
                + 'Details: {}'.format(e)
 | 
			
		||||
                f'Run container "{name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.APIError" error: {e}'
 | 
			
		||||
            )
 | 
			
		||||
        else:
 | 
			
		||||
            corpus.status = 'analysing'
 | 
			
		||||
            return
 | 
			
		||||
        corpus.status = 'analysing'
 | 
			
		||||
 | 
			
		||||
    def checkout_analysing_corpus_container(self, corpus):
 | 
			
		||||
        container_name = 'cqpserver_{}'.format(corpus.id)
 | 
			
		||||
        container_name = f'cqpserver_{corpus.id}'
 | 
			
		||||
        try:
 | 
			
		||||
            self.docker.containers.get(container_name)
 | 
			
		||||
        except docker.errors.NotFound:
 | 
			
		||||
        except docker.errors.NotFound as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Could not find "{}" but the corpus state is "analysing".'
 | 
			
		||||
                f'Get container "{container_name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.NotFound": {e}'
 | 
			
		||||
            )
 | 
			
		||||
            corpus.num_analysis_sessions = 0
 | 
			
		||||
            corpus.status = 'prepared'
 | 
			
		||||
        except docker.errors.APIError as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Get "{}" container raised '.format(container_name)
 | 
			
		||||
                + '"docker.errors.APIError" The server returned an error. '
 | 
			
		||||
                + 'Details: {}'.format(e)
 | 
			
		||||
                f'Get container "{container_name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.APIError": {e}'
 | 
			
		||||
            )
 | 
			
		||||
            return
 | 
			
		||||
 | 
			
		||||
    def remove_cqpserver_container(self, corpus):
 | 
			
		||||
        container_name = 'cqpserver_{}'.format(corpus.id)
 | 
			
		||||
        container_name = f'cqpserver_{corpus.id}'
 | 
			
		||||
        try:
 | 
			
		||||
            container = self.docker.containers.get(container_name)
 | 
			
		||||
        except docker.errors.NotFound:
 | 
			
		||||
            pass
 | 
			
		||||
            corpus.status = 'prepared'
 | 
			
		||||
            return
 | 
			
		||||
        except docker.errors.APIError as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Get "{}" container raised '.format(container_name)
 | 
			
		||||
                + '"docker.errors.APIError" The server returned an error. '
 | 
			
		||||
                + 'Details: {}'.format(e)
 | 
			
		||||
                f'Get container "{container_name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.APIError": {e}'
 | 
			
		||||
            )
 | 
			
		||||
            return
 | 
			
		||||
        else:
 | 
			
		||||
            try:
 | 
			
		||||
                container.remove(force=True)
 | 
			
		||||
            except docker.errors.APIError as e:
 | 
			
		||||
                current_app.logger.error(
 | 
			
		||||
                    'Remove "{}" container raised '.format(container_name)
 | 
			
		||||
                    + '"docker.errors.APIError" The server returned an error. '
 | 
			
		||||
                    + 'Details: {}'.format(e)
 | 
			
		||||
                )
 | 
			
		||||
                return
 | 
			
		||||
        corpus.status = 'prepared'
 | 
			
		||||
        try:
 | 
			
		||||
            container.remove(force=True)
 | 
			
		||||
        except docker.errors.APIError as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                f'Remove container "{container_name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.APIError": {e}'
 | 
			
		||||
            )
 | 
			
		||||
 
 | 
			
		||||
@@ -12,15 +12,11 @@ import shutil
 | 
			
		||||
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:
 | 
			
		||||
        for job in (x for x in jobs if x.status == 'submitted'):
 | 
			
		||||
            self.create_job_service(job)
 | 
			
		||||
        for job in queued_jobs + running_jobs:
 | 
			
		||||
        for job in (x for x in jobs if x.status in ['queued', 'running']):
 | 
			
		||||
            self.checkout_job_service(job)
 | 
			
		||||
        for job in canceling_jobs:
 | 
			
		||||
        for job in (x for x in jobs if x.status == 'canceling'):
 | 
			
		||||
            self.remove_job_service(job)
 | 
			
		||||
 | 
			
		||||
    def create_job_service(self, job):
 | 
			
		||||
@@ -30,26 +26,23 @@ class CheckJobsMixin:
 | 
			
		||||
            mem_mb = 2048
 | 
			
		||||
            n_cores = 2
 | 
			
		||||
            executable = 'file-setup'
 | 
			
		||||
            image = (current_app.config['DOCKER_IMAGE_PREFIX']
 | 
			
		||||
                     + 'file-setup:' + job.service_version)
 | 
			
		||||
            image = f'{current_app.config["NOPAQUE_DOCKER_IMAGE_PREFIX"]}file-setup:{job.service_version}'  # noqa
 | 
			
		||||
        elif job.service == 'ocr':
 | 
			
		||||
            mem_mb = 4096
 | 
			
		||||
            n_cores = 4
 | 
			
		||||
            executable = 'ocr'
 | 
			
		||||
            image = (current_app.config['DOCKER_IMAGE_PREFIX']
 | 
			
		||||
                     + 'ocr:' + job.service_version)
 | 
			
		||||
            image = f'{current_app.config["NOPAQUE_DOCKER_IMAGE_PREFIX"]}ocr:{job.service_version}'  # noqa
 | 
			
		||||
        elif job.service == 'nlp':
 | 
			
		||||
            mem_mb = 2048
 | 
			
		||||
            n_cores = 2
 | 
			
		||||
            executable = 'nlp'
 | 
			
		||||
            image = (current_app.config['DOCKER_IMAGE_PREFIX']
 | 
			
		||||
                     + 'nlp:' + job.service_version)
 | 
			
		||||
            image = f'{current_app.config["NOPAQUE_DOCKER_IMAGE_PREFIX"]}nlp:{job.service_version}'  # noqa
 | 
			
		||||
        ''' ## Command ## '''
 | 
			
		||||
        command = '{} -i /input -o /output'.format(executable)
 | 
			
		||||
        command = f'{executable} -i /input -o /output'
 | 
			
		||||
        command += ' --log-dir /input'
 | 
			
		||||
        command += ' --mem-mb {}'.format(mem_mb)
 | 
			
		||||
        command += ' --n-cores {}'.format(n_cores)
 | 
			
		||||
        command += ' --zip [' + job.service + ']_' + secure_filename(job.title)
 | 
			
		||||
        command += f' --mem-mb {mem_mb}'
 | 
			
		||||
        command += f' --n-cores {n_cores}'
 | 
			
		||||
        command += f' --zip [{job.service}]_{secure_filename(job.title)}'
 | 
			
		||||
        command += ' ' + ' '.join(json.loads(job.service_args))
 | 
			
		||||
        ''' ## Constraints ## '''
 | 
			
		||||
        constraints = ['node.role==worker']
 | 
			
		||||
@@ -64,18 +57,18 @@ class CheckJobsMixin:
 | 
			
		||||
        input_mount_source = job.path
 | 
			
		||||
        input_mount_target = '/input'
 | 
			
		||||
        if job.service == 'file-setup':
 | 
			
		||||
            input_mount_target += '/' + secure_filename(job.title)
 | 
			
		||||
        input_mount = input_mount_source + ':' + input_mount_target + ':rw'
 | 
			
		||||
            input_mount_target += f'/{secure_filename(job.title)}'
 | 
			
		||||
        input_mount = f'{input_mount_source}:{input_mount_target}:rw'
 | 
			
		||||
        ''' ### Output mount ### '''
 | 
			
		||||
        output_mount_source = os.path.join(job.path, 'output')
 | 
			
		||||
        output_mount_target = '/output'
 | 
			
		||||
        output_mount = output_mount_source + ':' + output_mount_target + ':rw'
 | 
			
		||||
        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 = [input_mount, output_mount]
 | 
			
		||||
        ''' ## Name ## '''
 | 
			
		||||
        name = 'job_{}'.format(job.id)
 | 
			
		||||
        name = f'job_{job.id}'
 | 
			
		||||
        ''' ## Resources ## '''
 | 
			
		||||
        resources = docker.types.Resources(
 | 
			
		||||
            cpu_reservation=n_cores * (10 ** 9),
 | 
			
		||||
@@ -96,104 +89,83 @@ class CheckJobsMixin:
 | 
			
		||||
            )
 | 
			
		||||
        except docker.errors.APIError as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Create "{}" service raised '.format(name)
 | 
			
		||||
                + '"docker.errors.APIError" The server returned an error. '
 | 
			
		||||
                + 'Details: {}'.format(e)
 | 
			
		||||
                f'Create service "{name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.APIError": {e}'
 | 
			
		||||
            )
 | 
			
		||||
            return
 | 
			
		||||
        else:
 | 
			
		||||
            job.status = 'queued'
 | 
			
		||||
        job.status = 'queued'
 | 
			
		||||
 | 
			
		||||
    def checkout_job_service(self, job):
 | 
			
		||||
        service_name = 'job_{}'.format(job.id)
 | 
			
		||||
        service_name = f'job_{job.id}'
 | 
			
		||||
        try:
 | 
			
		||||
            service = self.docker.services.get(service_name)
 | 
			
		||||
        except docker.errors.NotFound:
 | 
			
		||||
        except docker.errors.NotFound as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Get "{}" service raised '.format(service_name)
 | 
			
		||||
                + '"docker.errors.NotFound" The service does not exist. '
 | 
			
		||||
                + '(job.status: {} -> failed)'.format(job.status)
 | 
			
		||||
                f'Get service "{service_name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.NotFound": {e}'
 | 
			
		||||
            )
 | 
			
		||||
            job.status = 'failed'
 | 
			
		||||
            return
 | 
			
		||||
        except docker.errors.APIError as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Get "{}" service raised '.format(service_name)
 | 
			
		||||
                + '"docker.errors.APIError" The server returned an error. '
 | 
			
		||||
                + 'Details: {}'.format(e)
 | 
			
		||||
                f'Get service "{service_name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.APIError": {e}'
 | 
			
		||||
            )
 | 
			
		||||
            return
 | 
			
		||||
        except docker.errors.InvalidVersion:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Get "{}" service raised '.format(service_name)
 | 
			
		||||
                + '"docker.errors.InvalidVersion" One of the arguments is '
 | 
			
		||||
                + 'not supported with the current API version.'
 | 
			
		||||
            )
 | 
			
		||||
        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'
 | 
			
		||||
            return
 | 
			
		||||
        elif job.status == 'running' and task_state == 'complete':
 | 
			
		||||
            job.status = 'complete'
 | 
			
		||||
            results_dir = os.path.join(job.path, 'output')
 | 
			
		||||
            result_files = [x for x in os.listdir(results_dir) if x.endswith('.zip')]  # noqa
 | 
			
		||||
            for result_file in result_files:
 | 
			
		||||
                job_result = JobResult(filename=result_file, job=job)
 | 
			
		||||
                db.session.add(job_result)
 | 
			
		||||
                db.session.flush()
 | 
			
		||||
                db.session.refresh(job_result)
 | 
			
		||||
        elif job.status == 'running' and task_state == 'failed':
 | 
			
		||||
            job.status = 'failed'
 | 
			
		||||
        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'
 | 
			
		||||
            elif job.status == 'running' and task_state in ['complete', 'failed']:  # noqa
 | 
			
		||||
                try:
 | 
			
		||||
                    service.remove()
 | 
			
		||||
                except docker.errors.APIError as e:
 | 
			
		||||
                    current_app.logger.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)
 | 
			
		||||
                    job.end_date = datetime.utcnow()
 | 
			
		||||
                    job.status = task_state
 | 
			
		||||
            return
 | 
			
		||||
        job.end_date = datetime.utcnow()
 | 
			
		||||
        try:
 | 
			
		||||
            service.remove()
 | 
			
		||||
        except docker.errors.APIError as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                f'Remove service "{service_name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.APIError": {e}'
 | 
			
		||||
            )
 | 
			
		||||
 | 
			
		||||
    def remove_job_service(self, job):
 | 
			
		||||
        service_name = 'job_{}'.format(job.id)
 | 
			
		||||
        service_name = f'job_{job.id}'
 | 
			
		||||
        try:
 | 
			
		||||
            service = self.docker.services.get(service_name)
 | 
			
		||||
        except docker.errors.NotFound:
 | 
			
		||||
            job.status = 'canceled'
 | 
			
		||||
            return
 | 
			
		||||
        except docker.errors.APIError as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Get "{}" service raised '.format(service_name)
 | 
			
		||||
                + '"docker.errors.APIError" The server returned an error. '
 | 
			
		||||
                + 'Details: {}'.format(e)
 | 
			
		||||
                f'Get service "{service_name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.APIError": {e}'
 | 
			
		||||
            )
 | 
			
		||||
            return
 | 
			
		||||
        except docker.errors.InvalidVersion:
 | 
			
		||||
        try:
 | 
			
		||||
            service.update(mounts=None)
 | 
			
		||||
        except docker.errors.APIError as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                'Get "{}" service raised '.format(service_name)
 | 
			
		||||
                + '"docker.errors.InvalidVersion" One of the arguments is '
 | 
			
		||||
                + 'not supported with the current API version.'
 | 
			
		||||
                f'Update service "{service_name}" failed '
 | 
			
		||||
                + f'due to "docker.errors.APIError": {e}'
 | 
			
		||||
            )
 | 
			
		||||
            return
 | 
			
		||||
        else:
 | 
			
		||||
            try:
 | 
			
		||||
                service.update(mounts=None)
 | 
			
		||||
            except docker.errors.APIError as e:
 | 
			
		||||
                current_app.logger.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:
 | 
			
		||||
                current_app.logger.error(
 | 
			
		||||
                    'Remove "{}" service raised '.format(service_name)
 | 
			
		||||
                    + '"docker.errors.APIError" The server returned an error. '
 | 
			
		||||
                    + 'Details: {}'.format(e)
 | 
			
		||||
                )
 | 
			
		||||
        try:
 | 
			
		||||
            service.remove()
 | 
			
		||||
        except docker.errors.APIError as e:
 | 
			
		||||
            current_app.logger.error(
 | 
			
		||||
                f'Remove "{service_name}" service failed '
 | 
			
		||||
                + f'due to "docker.errors.APIError": {e}'
 | 
			
		||||
            )
 | 
			
		||||
 
 | 
			
		||||
		Reference in New Issue
	
	Block a user