def update_code():
data = request_data()
provider = FileProvider(_write_session)
file = provider.by_id(data['file_id'])
content = data['content'].encode('utf-8')
md5 = hashlib.md5(content).hexdigest()
if md5 == file.md5:
return
if file.dag != data['dag']:
new_file = File(md5=md5,
content=content,
project=file.project,
dag=data['dag'])
provider.add(new_file)
storage = DagStorageProvider(_write_session).by_id(data['storage'])
storage.file = new_file.id
provider.commit()
return {'file': new_file.id}
else:
file.content = content
file.md5 = md5
provider.commit()
return {'file': file.id}
def create_providers(self):
self.log_info('create_providers')
self.dag_provider = DagProvider(self.session)
self.task_provider = TaskProvider(self.session)
self.file_provider = FileProvider(self.session)
self.dag_storage_provider = DagStorageProvider(self.session)
def __init__(self, session: Session, logger=None,
component: ComponentType = None,
max_file_size: int = 10 ** 5, max_count=10 ** 3):
self.file_provider = FileProvider(session)
self.provider = DagStorageProvider(session)
self.task_provider = TaskProvider(session)
self.library_provider = DagLibraryProvider(session)
self.dag_provider = DagProvider(session)
self.logger = logger
self.component = component
self.max_file_size = max_file_size
self.max_count = max_count
def code():
id = request_data()
res = OrderedDict()
parents = dict()
for s, f in DagStorageProvider(_read_session).by_dag(id):
s.path = s.path.strip()
parent = os.path.dirname(s.path)
name = os.path.basename(s.path)
if name == '':
continue
if s.is_dir:
node = {
'name': name,
'children': [],
'id': s.id,
'dag': id,
'storage': s.id
}
if not parent:
res[name] = node
parents[s.path] = node
else:
# noinspection PyUnresolvedReferences
parents[parent]['children'].append(node)
parents[os.path.join(parent, name)] = node
else:
node = {'name': name, 'id': f.id, 'dag': id, 'storage': s.id}
try:
node['content'] = f.content.decode('utf-8')
except UnicodeDecodeError:
node['content'] = ''
if not parent:
res[name] = node
else:
# noinspection PyUnresolvedReferences
parents[parent]['children'].append(node)
def sort_key(x):
if 'children' in x and len(x['children']) > 0:
return '_' * 5 + x['name']
return x['name']
def sort(node: dict):
if 'children' in node and len(node['children']) > 0:
node['children'] = sorted(node['children'], key=sort_key)
for c in node['children']:
sort(c)
res = sorted(list(res.values()), key=sort_key)
for r in res:
sort(r)
return {'items': res}
class Storage:
def __init__(self,
session: Session,
logger=None,
component: ComponentType = None,
max_file_size: int = 10**5,
max_count=10**3):
self.file_provider = FileProvider(session)
self.provider = DagStorageProvider(session)
self.task_provider = TaskProvider(session)
self.library_provider = DagLibraryProvider(session)
self.dag_provider = DagProvider(session)
self.logger = logger
self.component = component
self.max_file_size = max_file_size
self.max_count = max_count
def log_info(self, message: str):
if self.logger:
self.logger.info(message, self.component)
def copy_from(self, src: int, dag: Dag):
storages = self.provider.query(DagStorage). \
filter(DagStorage.dag == src). \
all()
libraries = self.library_provider.query(DagLibrary). \
filter(DagLibrary.dag == src). \
all()
s_news = []
for s in storages:
s_new = DagStorage(dag=dag.id,
file=s.file,
path=s.path,
is_dir=s.is_dir)
s_news.append(s_new)
l_news = []
for l in libraries:
l_new = DagLibrary(dag=dag.id,
library=l.library,
version=l.version)
l_news.append(l_new)
self.provider.add_all(s_news)
self.library_provider.add_all(l_news)
def _build_spec(self, folder: str):
ignore_file = os.path.join(folder, 'file.ignore.txt')
if not os.path.exists(ignore_file):
ignore_patterns = []
else:
ignore_patterns = read_lines(ignore_file)
ignore_patterns.extend(
['log', '/data', '/models', '__pycache__', '*.ipynb'])
return pathspec.PathSpec.from_lines(
pathspec.patterns.GitWildMatchPattern, ignore_patterns)
def upload(self, folder: str, dag: Dag, control_reqs: bool = True):
self.log_info('upload started')
hashs = self.file_provider.hashs(dag.project)
self.log_info('hashes are retrieved')
all_files = []
spec = self._build_spec(folder)
files = glob(os.path.join(folder, '**'))
for file in files[:]:
path = os.path.relpath(file, folder)
if spec.match_file(path) or path == '.':
continue
if os.path.isdir(file):
child_files = glob(os.path.join(folder, file, '**'),
recursive=True)
files.extend(child_files)
if self.max_count and len(files) > self.max_count:
raise Exception(f'files count = {len(files)} '
f'But max count = {self.max_count}')
self.log_info('list of files formed')
folders_to_add = []
files_to_add = []
files_storage_to_add = []
total_size_added = 0
for o in files:
path = os.path.relpath(o, folder)
if spec.match_file(path) or path == '.':
continue
if isdir(o):
folder_to_add = DagStorage(dag=dag.id, path=path, is_dir=True)
folders_to_add.append(folder_to_add)
continue
content = open(o, 'rb').read()
size = sys.getsizeof(content)
if self.max_file_size and size > self.max_file_size:
raise Exception(
f'file = {o} has size {size}.'
f' But max size is set to {self.max_file_size}')
md5 = hashlib.md5(content).hexdigest()
all_files.append(o)
if md5 not in hashs:
file = File(md5=md5,
content=content,
project=dag.project,
dag=dag.id,
created=now())
hashs[md5] = file
files_to_add.append(file)
total_size_added += size
file_storage = DagStorage(dag=dag.id,
path=path,
file=hashs[md5],
is_dir=False)
files_storage_to_add.append(file_storage)
self.log_info('inserting DagStorage folders')
if len(folders_to_add) > 0:
self.provider.bulk_save_objects(folders_to_add)
self.log_info('inserting Files')
if len(files_to_add) > 0:
self.file_provider.bulk_save_objects(files_to_add,
return_defaults=True)
self.log_info('inserting DagStorage Files')
if len(files_storage_to_add) > 0:
for file_storage in files_storage_to_add:
if isinstance(file_storage.file, File):
# noinspection PyUnresolvedReferences
file_storage.file = file_storage.file.id
self.provider.bulk_save_objects(files_storage_to_add)
dag.file_size += total_size_added
self.dag_provider.update()
if INSTALL_DEPENDENCIES and control_reqs:
reqs = control_requirements(folder, files=all_files)
for name, rel, version in reqs:
self.library_provider.add(
DagLibrary(dag=dag.id, library=name, version=version))
def download_dag(self, dag: int, folder: str):
os.makedirs(folder, exist_ok=True)
items = self.provider.by_dag(dag)
items = sorted(items, key=lambda x: x[1] is not None)
for item, file in items:
path = os.path.join(folder, item.path)
if item.is_dir:
os.makedirs(path, exist_ok=True)
else:
with open(path, 'wb') as f:
f.write(file.content)
def download(self, task: int):
task = self.task_provider.by_id(
task, joinedload(Task.dag_rel, innerjoin=True))
folder = join(TASK_FOLDER, str(task.id))
self.download_dag(task.dag, folder)
config = Config.from_yaml(task.dag_rel.config)
info = config['info']
try:
data_folder = os.path.join(DATA_FOLDER, info['project'])
os.makedirs(data_folder, exist_ok=True)
os.symlink(data_folder,
os.path.join(folder, 'data'),
target_is_directory=True)
except FileExistsError:
pass
try:
model_folder = os.path.join(MODEL_FOLDER, info['project'])
os.makedirs(model_folder, exist_ok=True)
os.symlink(model_folder,
os.path.join(folder, 'models'),
target_is_directory=True)
except FileExistsError:
pass
sys.path.insert(0, folder)
return folder
def import_executor(self,
folder: str,
base_folder: str,
executor: str,
libraries: List[Tuple] = None):
sys.path.insert(0, base_folder)
spec = self._build_spec(folder)
was_installation = False
folders = [
p for p in glob(f'{folder}/*', recursive=True)
if os.path.isdir(p) and not spec.match_file(p)
]
folders += [folder]
library_names = set(n for n, v in (libraries or []))
library_versions = {n: v for n, v in (libraries or [])}
for n in library_names:
try:
version = pkg_resources.get_distribution(n).version
need_install = library_versions[n] != version
except Exception:
need_install = True
if INSTALL_DEPENDENCIES and need_install:
os.system(f'pip install {n}=={library_versions[n]}')
was_installation = True
def is_valid_class(cls: pyclbr.Class):
return cls.name == executor or \
cls.name.lower() == executor or \
to_snake(cls.name) == executor
def relative_name(path: str):
rel = os.path.relpath(path, base_folder)
parts = [str(p).split('.')[0] for p in rel.split(os.sep)]
return '.'.join(parts)
for (module_loader, module_name,
ispkg) in pkgutil.iter_modules(folders):
module = module_loader.find_module(module_name)
rel_path = os.path.relpath(
os.path.splitext(module.path)[0],
base_folder).replace('/', '.')
try:
classes = pyclbr.readmodule(rel_path, path=[base_folder])
except Exception:
continue
for k, v in classes.items():
if is_valid_class(v):
importlib.import_module(relative_name(module.path))
return True, was_installation
return False, was_installation
class DagCopyBuilder:
def __init__(self,
session: Session,
dag: int,
file_changes: str = '',
dag_suffix: str = '',
logger=None,
component: ComponentType = None):
self.dag = dag
self.file_changes = file_changes
self.session = session
self.logger = logger
self.component = component
self.dag_suffix = dag_suffix
self.dag_db = None
self.dag_provider = None
self.task_provider = None
self.file_provider = None
self.dag_storage_provider = None
def log_info(self, message: str):
if self.logger:
self.logger.info(message, self.component)
def create_providers(self):
self.log_info('create_providers')
self.dag_provider = DagProvider(self.session)
self.task_provider = TaskProvider(self.session)
self.file_provider = FileProvider(self.session)
self.dag_storage_provider = DagStorageProvider(self.session)
def create_dag(self):
dag = self.dag_provider.by_id(self.dag)
name = dag.name
if self.dag_suffix:
name += ' ' + self.dag_suffix
dag_new = Dag(name=name,
created=now(),
config=dag.config,
project=dag.project,
docker_img=dag.docker_img,
img_size=0,
file_size=0,
type=dag.type)
self.dag_provider.add(dag_new)
self.dag_db = dag_new
def find_replace(self, changes: dict, path: str):
for k, v in changes.items():
if not re.match(k, path):
continue
return v
def create_tasks(self):
tasks = self.task_provider.by_dag(self.dag)
tasks_new = []
tasks_old = []
for t in tasks:
if t.parent:
continue
task = Task(
name=t.name,
status=TaskStatus.NotRan.value,
computer=t.computer,
gpu=t.gpu,
gpu_max=t.gpu_max,
cpu=t.cpu,
executor=t.executor,
memory=t.memory,
steps=t.steps,
dag=self.dag_db.id,
debug=t.debug,
type=t.type,
)
task.additional_info = t.additional_info
tasks_new.append(task)
tasks_old.append(t)
self.task_provider.bulk_save_objects(tasks_new, return_defaults=True)
old2new = {
t_old.id: t_new.id
for t_new, t_old in zip(tasks_new, tasks_old)
}
dependencies = self.task_provider.get_dependencies(self.dag)
dependencies_new = []
for d in dependencies:
d_new = TaskDependence(task_id=old2new[d.task_id],
depend_id=old2new[d.depend_id])
dependencies_new.append(d_new)
self.task_provider.bulk_save_objects(dependencies_new,
return_defaults=False)
changes = yaml_load(self.file_changes)
storages = self.dag_storage_provider.by_dag(self.dag)
storages_new = []
for s, f in storages:
if not isinstance(changes, dict):
continue
replace = self.find_replace(changes, s.path)
if replace is not None and f:
content = f.content.decode('utf-8')
if s.path.endswith('.yml'):
data = yaml_load(content)
data = merge_dicts_smart(data, replace)
content = yaml_dump(data)
else:
for k, v in replace:
if k not in content:
raise Exception(f'{k} is not in the content')
content = content.replace(k, v)
content = content.encode('utf-8')
md5 = hashlib.md5(content).hexdigest()
f = self.file_provider.by_md5(md5)
if not f:
f = File(content=content,
created=now(),
project=self.dag_db.project,
md5=md5,
dag=self.dag_db.id)
self.file_provider.add(f)
s_new = DagStorage(dag=self.dag_db.id,
file=f.id,
path=s.path,
is_dir=s.is_dir)
storages_new.append(s_new)
self.dag_storage_provider.bulk_save_objects(storages_new,
return_defaults=False)
def build(self):
self.create_providers()
self.create_dag()
self.create_tasks()
def __init__(self, session: Session):
self.file_provider = FileProvider(session)
self.provider = DagStorageProvider(session)
self.task_provider = TaskProvider(session)
self.library_provider = DagLibraryProvider(session)
class Storage:
def __init__(self, session: Session):
self.file_provider = FileProvider(session)
self.provider = DagStorageProvider(session)
self.task_provider = TaskProvider(session)
self.library_provider = DagLibraryProvider(session)
def copy_from(self, src: int, dag: Dag):
storages = self.provider.query(DagStorage). \
filter(DagStorage.dag == src). \
all()
libraries = self.library_provider.query(DagLibrary). \
filter(DagLibrary.dag == src). \
all()
s_news = []
for s in storages:
s_new = DagStorage(dag=dag.id,
file=s.file,
path=s.path,
is_dir=s.is_dir)
s_news.append(s_new)
l_news = []
for l in libraries:
l_new = DagLibrary(dag=dag.id,
library=l.library,
version=l.version)
l_news.append(l_new)
self.provider.add_all(s_news)
self.library_provider.add_all(l_news)
def _build_spec(self, folder: str):
ignore_file = os.path.join(folder, 'file.ignore.txt')
if not os.path.exists(ignore_file):
ignore_patterns = []
else:
ignore_patterns = read_lines(ignore_file)
ignore_patterns.extend(['log', 'data', 'models', '__pycache__'])
return pathspec.PathSpec.from_lines(
pathspec.patterns.GitWildMatchPattern, ignore_patterns)
def upload(self, folder: str, dag: Dag, control_reqs: bool = True):
hashs = self.file_provider.hashs(dag.project)
files = []
all_files = []
spec = self._build_spec(folder)
for o in glob(os.path.join(folder, '**'), recursive=True):
path = os.path.relpath(o, folder)
if spec.match_file(path) or path == '.':
continue
if isdir(o):
self.provider.add(
DagStorage(dag=dag.id, path=path, is_dir=True))
continue
content = open(o, 'rb').read()
md5 = hashlib.md5(content).hexdigest()
all_files.append(o)
if md5 in hashs:
file_id = hashs[md5]
else:
file = File(md5=md5,
content=content,
project=dag.project,
dag=dag.id,
created=now())
self.file_provider.add(file)
file_id = file.id
hashs[md5] = file.id
files.append(o)
self.provider.add(
DagStorage(dag=dag.id, path=path, file=file_id, is_dir=False))
if INSTALL_DEPENDENCIES and control_reqs:
reqs = control_requirements(folder, files=all_files)
for name, rel, version in reqs:
self.library_provider.add(
DagLibrary(dag=dag.id, library=name, version=version))
def download(self, task: int):
task = self.task_provider.by_id(
task, joinedload(Task.dag_rel, innerjoin=True))
folder = join(TASK_FOLDER, str(task.id))
os.makedirs(folder, exist_ok=True)
items = self.provider.by_dag(task.dag)
items = sorted(items, key=lambda x: x[1] is not None)
for item, file in items:
path = os.path.join(folder, item.path)
if item.is_dir:
os.makedirs(path, exist_ok=True)
else:
with open(path, 'wb') as f:
f.write(file.content)
config = Config.from_yaml(task.dag_rel.config)
info = config['info']
try:
data_folder = os.path.join(DATA_FOLDER, info['project'])
os.makedirs(data_folder, exist_ok=True)
os.symlink(data_folder,
os.path.join(folder, 'data'),
target_is_directory=True)
except FileExistsError:
pass
try:
model_folder = os.path.join(MODEL_FOLDER, info['project'])
os.makedirs(model_folder, exist_ok=True)
os.symlink(model_folder,
os.path.join(folder, 'models'),
target_is_directory=True)
except FileExistsError:
pass
sys.path.insert(0, folder)
return folder
def import_executor(self,
folder: str,
base_folder: str,
executor: str,
libraries: List[Tuple] = None):
sys.path.insert(0, base_folder)
spec = self._build_spec(folder)
was_installation = False
folders = [
p for p in glob(f'{folder}/*', recursive=True)
if os.path.isdir(p) and not spec.match_file(p)
]
folders += [folder]
library_names = set(n for n, v in (libraries or []))
library_versions = {n: v for n, v in (libraries or [])}
for n in library_names:
try:
version = pkg_resources.get_distribution(n).version
need_install = library_versions[n] != version
except Exception:
need_install = True
if INSTALL_DEPENDENCIES and need_install:
os.system(f'pip install {n}=={library_versions[n]}')
was_installation = True
def is_valid_class(cls: pyclbr.Class):
super_names = get_super_names(cls)
if 'Executor' not in super_names:
return False
return cls.name == executor or \
cls.name.lower() == executor or \
to_snake(cls.name) == executor
def relative_name(path: str):
rel = os.path.relpath(path, base_folder)
parts = [str(p).split('.')[0] for p in rel.split(os.sep)]
return '.'.join(parts)
for (module_loader, module_name,
ispkg) in pkgutil.iter_modules(folders):
module = module_loader.find_module(module_name)
module_folder = dirname(module.path)
classes = pyclbr.readmodule(module_name, path=[module_folder])
for k, v in classes.items():
if is_valid_class(v):
importlib.import_module(relative_name(module.path))
return True, was_installation
return False, was_installation