clearml/trains/backend_interface/task/repo/scriptinfo.py

644 lines
25 KiB
Python

import os
import sys
from copy import copy
from tempfile import mkstemp
import attr
import collections
import logging
import json
from furl import furl
from pathlib2 import Path
from threading import Thread, Event
from .util import get_command_output
from ....backend_api import Session
from ....debugging import get_logger
from .detectors import GitEnvDetector, GitDetector, HgEnvDetector, HgDetector, Result as DetectionResult
_logger = get_logger("Repository Detection")
class ScriptInfoError(Exception):
pass
class ScriptRequirements(object):
_max_requirements_size = 512 * 1024
def __init__(self, root_folder):
self._root_folder = root_folder
def get_requirements(self, entry_point_filename=None):
try:
from ....utilities.pigar.reqs import get_installed_pkgs_detail
from ....utilities.pigar.__main__ import GenerateReqs
installed_pkgs = get_installed_pkgs_detail()
gr = GenerateReqs(save_path='', project_path=self._root_folder, installed_pkgs=installed_pkgs,
ignores=['.git', '.hg', '.idea', '__pycache__', '.ipynb_checkpoints',
'site-packages', 'dist-packages'])
reqs, try_imports, guess, local_pks = gr.extract_reqs(
module_callback=ScriptRequirements.add_trains_used_packages, entry_point_filename=entry_point_filename)
return self.create_requirements_txt(reqs, local_pks)
except Exception:
return '', ''
@staticmethod
def add_trains_used_packages(modules):
# hack: forcefully insert storage modules if we have them
# noinspection PyBroadException
try:
import boto3
modules.add('boto3', 'trains.storage', 0)
except Exception:
pass
# noinspection PyBroadException
try:
from google.cloud import storage
modules.add('google_cloud_storage', 'trains.storage', 0)
except Exception:
pass
# noinspection PyBroadException
try:
from azure.storage.blob import ContentSettings
modules.add('azure_storage_blob', 'trains.storage', 0)
except Exception:
pass
# bugfix, replace sklearn with scikit-learn name
if 'sklearn' in modules:
sklearn = modules.pop('sklearn', {})
for fname, lines in sklearn.items():
modules.add('scikit_learn', fname, lines)
# if we have torch and it supports tensorboard, we should add that as well
# (because it will not be detected automatically)
if 'torch' in modules and 'tensorboard' not in modules:
# noinspection PyBroadException
try:
# see if this version of torch support tensorboard
import torch.utils.tensorboard
import tensorboard
modules.add('tensorboard', 'torch', 0)
except Exception:
pass
# remove setuptools, we should not specify this module version. It is installed by default
if 'setuptools' in modules:
modules.pop('setuptools', {})
# add forced requirements:
# noinspection PyBroadException
try:
from ..task import Task
for package, version in Task._force_requirements.items():
modules.add(package, 'trains', 0)
except Exception:
pass
return modules
@staticmethod
def create_requirements_txt(reqs, local_pks=None):
# write requirements.txt
try:
conda_requirements = ''
conda_prefix = os.environ.get('CONDA_PREFIX')
if conda_prefix and not conda_prefix.endswith(os.path.sep):
conda_prefix += os.path.sep
if conda_prefix and sys.executable.startswith(conda_prefix):
conda_packages_json = get_command_output(['conda', 'list', '--json'])
conda_packages_json = json.loads(conda_packages_json)
reqs_lower = {k.lower(): (k, v) for k, v in reqs.items()}
for r in conda_packages_json:
# check if this is a pypi package, if it is, leave it outside
if not r.get('channel') or r.get('channel') == 'pypi':
continue
# check if we have it in our required packages
name = r['name'].lower().replace('-', '_')
# hack support pytorch/torch different naming convention
if name == 'pytorch':
name = 'torch'
k, v = reqs_lower.get(name, (None, None))
if k:
conda_requirements += '{0} {1} {2}\n'.format(k, '==', v.version)
except:
conda_requirements = ''
# add forced requirements:
# noinspection PyBroadException
try:
from ..task import Task
forced_packages = copy(Task._force_requirements)
except Exception:
forced_packages = {}
# python version header
requirements_txt = '# Python ' + sys.version.replace('\n', ' ').replace('\r', ' ') + '\n'
if local_pks:
requirements_txt += '\n# Local modules found - skipping:\n'
for k, v in local_pks.sorted_items():
requirements_txt += '# {0} == {1}\n'.format(k, v.version)
# requirement summary
requirements_txt += '\n'
for k, v in reqs.sorted_items():
version = v.version
if k in forced_packages:
forced_version = forced_packages.pop(k, None)
if forced_version:
version = forced_version
# requirements_txt += ''.join(['# {0}\n'.format(c) for c in v.comments.sorted_items()])
if k == '-e':
requirements_txt += '{0} {1}\n'.format(k, version)
elif v:
requirements_txt += '{0} {1} {2}\n'.format(k, '==', version)
else:
requirements_txt += '{0}\n'.format(k)
# add forced requirements that we could not find installed on the system
for k in sorted(forced_packages.keys()):
if forced_packages[k]:
requirements_txt += '{0} {1} {2}\n'.format(k, '==', forced_packages[k])
else:
requirements_txt += '{0}\n'.format(k)
requirements_txt_packages_only = \
requirements_txt + '\n# Skipping detailed import analysis, it is too large\n'
# requirements details (in comments)
requirements_txt += '\n' + \
'# Detailed import analysis\n' \
'# **************************\n'
if local_pks:
for k, v in local_pks.sorted_items():
requirements_txt += '\n'
requirements_txt += '# IMPORT LOCAL PACKAGE {0}\n'.format(k)
requirements_txt += ''.join(['# {0}\n'.format(c) for c in v.comments.sorted_items()])
for k, v in reqs.sorted_items():
requirements_txt += '\n'
if k == '-e':
requirements_txt += '# IMPORT PACKAGE {0} {1}\n'.format(k, v.version)
else:
requirements_txt += '# IMPORT PACKAGE {0}\n'.format(k)
requirements_txt += ''.join(['# {0}\n'.format(c) for c in v.comments.sorted_items()])
# make sure we do not exceed the size a size limit
return (requirements_txt if len(requirements_txt) < ScriptRequirements._max_requirements_size
else requirements_txt_packages_only,
conda_requirements)
class _JupyterObserver(object):
_thread = None
_exit_event = Event()
_sync_event = Event()
_sample_frequency = 30.
_first_sample_frequency = 3.
@classmethod
def observer(cls, jupyter_notebook_filename):
if cls._thread is not None:
# order of signaling is important!
cls._exit_event.set()
cls._sync_event.set()
cls._thread.join()
cls._sync_event.clear()
cls._exit_event.clear()
cls._thread = Thread(target=cls._daemon, args=(jupyter_notebook_filename, ))
cls._thread.daemon = True
cls._thread.start()
@classmethod
def signal_sync(cls, *_):
cls._sync_event.set()
@classmethod
def close(cls):
if not cls._thread:
return
cls._exit_event.set()
cls._sync_event.set()
cls._thread.join()
cls._thread = None
@classmethod
def _daemon(cls, jupyter_notebook_filename):
from trains import Task
# load jupyter notebook package
# noinspection PyBroadException
try:
from nbconvert.exporters.script import ScriptExporter
_script_exporter = ScriptExporter()
except Exception:
return
# load pigar
# noinspection PyBroadException
try:
from ....utilities.pigar.reqs import get_installed_pkgs_detail, file_import_modules
from ....utilities.pigar.modules import ReqsModules
from ....utilities.pigar.log import logger
logger.setLevel(logging.WARNING)
except Exception:
file_import_modules = None
# load IPython
# noinspection PyBroadException
try:
from IPython import get_ipython
except Exception:
# should not happen
get_ipython = None
# setup local notebook files
if jupyter_notebook_filename:
notebook = Path(jupyter_notebook_filename)
local_jupyter_filename = jupyter_notebook_filename
else:
notebook = None
fd, local_jupyter_filename = mkstemp(suffix='.ipynb')
os.close(fd)
last_update_ts = None
counter = 0
prev_script_hash = None
try:
from ....version import __version__
our_module = cls.__module__.split('.')[0], __version__
except:
our_module = None
try:
import re
replace_ipython_pattern = re.compile('\\n([ \\t]*)get_ipython\(\)')
except:
replace_ipython_pattern = None
# main observer loop, check if we need to exit
while not cls._exit_event.wait(timeout=0.):
# wait for timeout or sync event
cls._sync_event.wait(cls._sample_frequency if counter else cls._first_sample_frequency)
cls._sync_event.clear()
counter += 1
# noinspection PyBroadException
try:
# if there is no task connected, do nothing
task = Task.current_task()
if not task:
continue
# if we have a local file:
if notebook:
if not notebook.exists():
continue
# check if notebook changed
if last_update_ts is not None and notebook.stat().st_mtime - last_update_ts <= 0:
continue
last_update_ts = notebook.stat().st_mtime
else:
# serialize notebook to a temp file
# noinspection PyBroadException
try:
get_ipython().run_line_magic('notebook', local_jupyter_filename)
except Exception as ex:
continue
# get notebook python script
script_code, resources = _script_exporter.from_filename(local_jupyter_filename)
current_script_hash = hash(script_code)
if prev_script_hash and prev_script_hash == current_script_hash:
continue
# remove ipython direct access from the script code
# we will not be able to run them anyhow
if replace_ipython_pattern:
script_code = replace_ipython_pattern.sub('\n# \g<1>get_ipython()', script_code)
requirements_txt = ''
conda_requirements = ''
# parse jupyter python script and prepare pip requirements (pigar)
# if backend supports requirements
if file_import_modules and Session.check_min_api_version('2.2'):
fmodules, _ = file_import_modules(notebook.parts[-1], script_code)
fmodules = ScriptRequirements.add_trains_used_packages(fmodules)
installed_pkgs = get_installed_pkgs_detail()
# make sure we are in installed packages
if our_module and (our_module[0] not in installed_pkgs):
installed_pkgs[our_module[0]] = our_module
reqs = ReqsModules()
for name in fmodules:
if name in installed_pkgs:
pkg_name, version = installed_pkgs[name]
reqs.add(pkg_name, version, fmodules[name])
requirements_txt, conda_requirements = ScriptRequirements.create_requirements_txt(reqs)
# update script
prev_script_hash = current_script_hash
data_script = task.data.script
data_script.diff = script_code
data_script.requirements = {'pip': requirements_txt, 'conda': conda_requirements}
task._update_script(script=data_script)
# update requirements
task._update_requirements(requirements=requirements_txt)
except Exception:
pass
class ScriptInfo(object):
plugins = [GitEnvDetector(), HgEnvDetector(), HgDetector(), GitDetector()]
""" Script info detection plugins, in order of priority """
@classmethod
def _jupyter_install_post_store_hook(cls, jupyter_notebook_filename):
# noinspection PyBroadException
try:
if 'IPython' in sys.modules:
from IPython import get_ipython
if get_ipython():
_JupyterObserver.observer(jupyter_notebook_filename)
get_ipython().events.register('pre_run_cell', _JupyterObserver.signal_sync)
except Exception:
pass
@classmethod
def _get_jupyter_notebook_filename(cls):
if not (sys.argv[0].endswith(os.path.sep + 'ipykernel_launcher.py') or
sys.argv[0].endswith(os.path.join(os.path.sep, 'ipykernel', '__main__.py'))) \
or len(sys.argv) < 3 or not sys.argv[2].endswith('.json'):
return None
# we can safely assume that we can import the notebook package here
# noinspection PyBroadException
try:
from notebook.notebookapp import list_running_servers
import requests
current_kernel = sys.argv[2].split(os.path.sep)[-1].replace('kernel-', '').replace('.json', '')
try:
server_info = next(list_running_servers())
except Exception:
# on some jupyter notebook versions this function can crash on parsing the json file,
# we will parse it manually here
import ipykernel
from glob import glob
import json
for f in glob(os.path.join(os.path.dirname(ipykernel.get_connection_file()), 'nbserver-*.json')):
try:
with open(f, 'r') as json_data:
server_info = json.load(json_data)
except:
server_info = None
if server_info:
break
try:
r = requests.get(
url=server_info['url'] + 'api/sessions',
headers={'Authorization': 'token {}'.format(server_info.get('token', '')), })
except requests.exceptions.SSLError:
# disable SSL check warning
from urllib3.exceptions import InsecureRequestWarning
requests.packages.urllib3.disable_warnings(category=InsecureRequestWarning)
# fire request
r = requests.get(
url=server_info['url'] + 'api/sessions',
headers={'Authorization': 'token {}'.format(server_info.get('token', '')), }, verify=False)
# enable SSL check warning
import warnings
warnings.simplefilter('default', InsecureRequestWarning)
r.raise_for_status()
notebooks = r.json()
cur_notebook = None
for n in notebooks:
if n['kernel']['id'] == current_kernel:
cur_notebook = n
break
notebook_path = cur_notebook['notebook'].get('path', '')
notebook_name = cur_notebook['notebook'].get('name', '')
is_google_colab = False
# check if this is google.colab, then there is no local file
# noinspection PyBroadException
try:
from IPython import get_ipython
if get_ipython() and 'google.colab' in get_ipython().extension_manager.loaded:
is_google_colab = True
except Exception:
pass
if is_google_colab:
script_entry_point = notebook_name
local_ipynb_file = None
else:
# always slash, because this is from uri (so never backslash not even oon windows)
entry_point_filename = notebook_path.split('/')[-1]
# now we should try to find the actual file
entry_point = (Path.cwd() / entry_point_filename).absolute()
if not entry_point.is_file():
entry_point = (Path.cwd() / notebook_path).absolute()
# get local ipynb for observer
local_ipynb_file = entry_point.as_posix()
# now replace the .ipynb with .py
# we assume we will have that file available with the Jupyter notebook plugin
entry_point = entry_point.with_suffix('.py')
script_entry_point = entry_point.as_posix()
# install the post store hook,
# notice that if we do not have a local file we serialize/write every time the entire notebook
cls._jupyter_install_post_store_hook(local_ipynb_file)
return script_entry_point
except Exception:
return None
@classmethod
def _get_entry_point(cls, repo_root, script_path):
repo_root = Path(repo_root).absolute()
try:
# Use os.path.relpath as it calculates up dir movements (../)
entry_point = os.path.relpath(str(script_path), str(Path.cwd()))
except ValueError:
# Working directory not under repository root
entry_point = script_path.relative_to(repo_root)
return Path(entry_point).as_posix()
@classmethod
def _get_working_dir(cls, repo_root):
repo_root = Path(repo_root).absolute()
try:
return Path.cwd().relative_to(repo_root).as_posix()
except ValueError:
# Working directory not under repository root
return os.path.curdir
@classmethod
def _get_script_code(cls, script_path):
# noinspection PyBroadException
try:
with open(script_path, 'r') as f:
script_code = f.read()
return script_code
except Exception:
pass
return ''
@classmethod
def _get_script_info(cls, filepaths, check_uncommitted=True, create_requirements=True, log=None):
jupyter_filepath = cls._get_jupyter_notebook_filename()
if jupyter_filepath:
scripts_path = [Path(os.path.normpath(jupyter_filepath)).absolute()]
else:
scripts_path = [Path(os.path.normpath(f)).absolute() for f in filepaths if f]
if all(not f.is_file() for f in scripts_path):
raise ScriptInfoError(
"Script file {} could not be found".format(scripts_path)
)
scripts_dir = [f.parent for f in scripts_path]
def _log(msg, *args, **kwargs):
if not log:
return
log.warning(
"Failed auto-detecting task repository: {}".format(
msg.format(*args, **kwargs)
)
)
plugin = next((p for p in cls.plugins if any(p.exists(d) for d in scripts_dir)), None)
repo_info = DetectionResult()
script_dir = scripts_dir[0]
script_path = scripts_path[0]
if not plugin:
log.info("No repository found, storing script code instead")
else:
try:
for i, d in enumerate(scripts_dir):
repo_info = plugin.get_info(str(d), include_diff=check_uncommitted)
if not repo_info.is_empty():
script_dir = d
script_path = scripts_path[i]
break
except Exception as ex:
_log("no info for {} ({})", scripts_dir, ex)
else:
if repo_info.is_empty():
_log("no info for {}", scripts_dir)
repo_root = repo_info.root or script_dir
if not plugin:
working_dir = '.'
entry_point = str(script_path.name)
else:
working_dir = cls._get_working_dir(repo_root)
entry_point = cls._get_entry_point(repo_root, script_path)
if check_uncommitted:
diff = cls._get_script_code(script_path.as_posix()) \
if not plugin or not repo_info.commit else repo_info.diff
else:
diff = ''
# if this is not jupyter, get the requirements.txt
requirements = ''
conda_requirements = ''
# create requirements if backend supports requirements
# if jupyter is present, requirements will be created in the background, when saving a snapshot
if not jupyter_filepath and Session.check_min_api_version('2.2'):
script_requirements = ScriptRequirements(
Path(repo_root).as_posix() if repo_info.url else script_path.as_posix())
if create_requirements:
requirements, conda_requirements = script_requirements.get_requirements()
else:
script_requirements = None
script_info = dict(
repository=furl(repo_info.url).remove(username=True, password=True).tostr(),
branch=repo_info.branch,
version_num=repo_info.commit,
entry_point=entry_point,
working_dir=working_dir,
diff=diff,
requirements={'pip': requirements, 'conda': conda_requirements} if requirements else None,
binary='python{}.{}'.format(sys.version_info.major, sys.version_info.minor),
repo_root=repo_root,
jupyter_filepath=jupyter_filepath,
)
messages = []
if repo_info.modified:
messages.append(
"======> WARNING! UNCOMMITTED CHANGES IN REPOSITORY {} <======".format(
script_info.get("repository", "")
)
)
if not any(script_info.values()):
script_info = None
return (ScriptInfoResult(script=script_info, warning_messages=messages),
script_requirements)
@classmethod
def get(cls, filepaths=None, check_uncommitted=True, create_requirements=True, log=None):
try:
if not filepaths:
filepaths = [sys.argv[0], ]
return cls._get_script_info(
filepaths=filepaths, check_uncommitted=check_uncommitted,
create_requirements=create_requirements, log=log)
except Exception as ex:
if log:
log.warning("Failed auto-detecting task repository: {}".format(ex))
return ScriptInfoResult(), None
@classmethod
def detect_running_module(cls, script_dict):
# noinspection PyBroadException
try:
# If this is jupyter, do not try to detect the running module, we know what we have.
if script_dict.get('jupyter_filepath'):
return script_dict
if '__main__' in sys.modules and vars(sys.modules['__main__'])['__package__']:
argvs = ''
git_root = os.path.abspath(script_dict['repo_root']) if script_dict['repo_root'] else None
for a in sys.argv[1:]:
if git_root and os.path.exists(a):
# check if common to project:
a_abs = os.path.abspath(a)
if os.path.commonpath([a_abs, git_root]) == git_root:
# adjust path relative to working dir inside git repo
a = ' ' + os.path.relpath(a_abs, os.path.join(git_root, script_dict['working_dir']))
argvs += ' {}'.format(a)
# update the script entry point to match the real argv and module call
script_dict['entry_point'] = '-m {}{}'.format(
vars(sys.modules['__main__'])['__package__'], (' ' + argvs) if argvs else '')
except Exception:
pass
return script_dict
@classmethod
def close(cls):
_JupyterObserver.close()
@attr.s
class ScriptInfoResult(object):
script = attr.ib(default=None)
warning_messages = attr.ib(factory=list)