#12275

pending completion

Build # #12275

Build Type

push

travis-ci

Committed by

web-flow

Commit Message

<a href="https://github.com/datajoint/datajoint-python/commit/<a class=hub.com/datajoint/datajoint-python/commit/823acaaba855a2e58996452679b4b6a7c376fe29">823acaaba<a href="https://github.com/datajoint/datajoint-python/commit/823acaaba855a2e58996452679b4b6a7c376fe29">">Merge </a><a class="double-link" href="https://github.com/datajoint/datajoint-python/commit/<a class="double-link" href="https://github.com/datajoint/datajoint-python/commit/719ea38c2238b4b89df25e06c3797b1114b9104a">719ea38c2</a>">719ea38c2</a><a href="https://github.com/datajoint/datajoint-python/commit/823acaaba855a2e58996452679b4b6a7c376fe29"> into 0ff34f283">0ff34f283</a>

Run Details

6 of 6 new or added lines in 1 file covered. (100.0%)

2991 of 3312 relevant lines covered (90.31%)

0.9 hits per line

Source File
Press 'n' to go to next uncovered line, 'b' for previous

81.2

/datajoint/autopopulate.py

"""This module defines class dj.AutoPopulate"""
import logging
import datetime
import traceback
import random
import inspect
from tqdm import tqdm
from .expression import QueryExpression, AndList
from .errors import DataJointError, LostConnectionError
import signal
import multiprocessing as mp
import contextlib

# noinspection PyExceptionInherit,PyCallingNonCallable

logger = logging.getLogger(__name__)


# --- helper functions for multiprocessing --


def _initialize_populate(table, jobs, populate_kwargs):
    """
    Initialize the process for mulitprocessing.
    Saves the unpickled copy of the table to the current process and reconnects.
    """
    process = mp.current_process()
    process.table = table
    process.jobs = jobs
    process.populate_kwargs = populate_kwargs
    table.connection.connect()  # reconnect


def _call_populate1(key):
    """
    Call current process' table._populate1()
    :key - a dict specifying job to compute
    :return: key, error if error, otherwise None
    """
    process = mp.current_process()
    return process.table._populate1(key, process.jobs, **process.populate_kwargs)


class AutoPopulate:
    """
    AutoPopulate is a mixin class that adds the method populate() to a Relation class.
    Auto-populated relations must inherit from both Relation and AutoPopulate,
    must define the property `key_source`, and must define the callback method `make`.
    """

    _key_source = None
    _allow_insert = False

    @property
    def key_source(self):
        """
        :return: the query expression that yields primary key values to be passed,
        sequentially, to the ``make`` method when populate() is called.
        The default value is the join of the parent tables references from the primary key.
        Subclasses may override they key_source to change the scope or the granularity
        of the make calls.
        """

        def _rename_attributes(table, props):
            return (
                table.proj(
                    **{
                        attr: ref
                        for attr, ref in props["attr_map"].items()
                        if attr != ref
                    }
                )
                if props["aliased"]
                else table.proj()
            )

        if self._key_source is None:
            parents = self.target.parents(
                primary=True, as_objects=True, foreign_key_info=True
            )
            if not parents:
                raise DataJointError(
                    "A table must have dependencies "
                    "from its primary key for auto-populate to work"
                )
            self._key_source = _rename_attributes(*parents[0])
            for q in parents[1:]:
                self._key_source *= _rename_attributes(*q)
        return self._key_source

    def make(self, key):
        """
        Derived classes must implement method `make` that fetches data from tables
        above them in the dependency hierarchy, restricting by the given key,
        computes secondary attributes, and inserts the new tuples into self.
        """
        raise NotImplementedError(
            "Subclasses of AutoPopulate must implement the method `make`"
        )

    @property
    def target(self):
        """
        :return: table to be populated.
        In the typical case, dj.AutoPopulate is mixed into a dj.Table class by
        inheritance and the target is self.
        """
        return self

    def _job_key(self, key):
        """
        :param key:  they key returned for the job from the key source
        :return: the dict to use to generate the job reservation hash
        This method allows subclasses to control the job reservation granularity.
        """
        return key

    def _jobs_to_do(self, restrictions):
        """
        :return: the relation containing the keys to be computed (derived from self.key_source)
        """
        if self.restriction:
            raise DataJointError(
                "Cannot call populate on a restricted table. "
                "Instead, pass conditions to populate() as arguments."
            )
        todo = self.key_source

        # key_source is a QueryExpression subclass -- trigger instantiation
        if inspect.isclass(todo) and issubclass(todo, QueryExpression):
            todo = todo()

        if not isinstance(todo, QueryExpression):
            raise DataJointError("Invalid key_source value")

        try:
            # check if target lacks any attributes from the primary key of key_source
            raise DataJointError(
                "The populate target lacks attribute %s "
                "from the primary key of key_source"
                % next(
                    name
                    for name in todo.heading.primary_key
                    if name not in self.target.heading
                )
            )
        except StopIteration:
            pass
        return (todo & AndList(restrictions)).proj()

    def populate(
        self,
        *restrictions,
        suppress_errors=False,
        return_exception_objects=False,
        reserve_jobs=False,
        order="original",
        limit=None,
        max_calls=None,
        display_progress=False,
        processes=1,
        make_kwargs=None
    ):
        """
        ``table.populate()`` calls ``table.make(key)`` for every primary key in
        ``self.key_source`` for which there is not already a tuple in table.

        :param restrictions: a list of restrictions each restrict
            (table.key_source - target.proj())
        :param suppress_errors: if True, do not terminate execution.
        :param return_exception_objects: return error objects instead of just error messages
        :param reserve_jobs: if True, reserve jobs to populate in asynchronous fashion
        :param order: "original"|"reverse"|"random"  - the order of execution
        :param limit: if not None, check at most this many keys
        :param max_calls: if not None, populate at most this many keys
        :param display_progress: if True, report progress_bar
        :param processes: number of processes to use. Set to None to use all cores
        :param make_kwargs: Keyword arguments which do not affect the result of computation
            to be passed down to each ``make()`` call. Computation arguments should be
            specified within the pipeline e.g. using a `dj.Lookup` table.
        :type make_kwargs: dict, optional
        """
        if self.connection.in_transaction:
            raise DataJointError("Populate cannot be called during a transaction.")

        valid_order = ["original", "reverse", "random"]
        if order not in valid_order:
            raise DataJointError(
                "The order argument must be one of %s" % str(valid_order)
            )
        jobs = (
            self.connection.schemas[self.target.database].jobs if reserve_jobs else None
        )

        # define and set up signal handler for SIGTERM:
        if reserve_jobs:

            def handler(signum, frame):
                logger.info("Populate terminated by SIGTERM")
                raise SystemExit("SIGTERM received")

            old_handler = signal.signal(signal.SIGTERM, handler)

        keys = (self._jobs_to_do(restrictions) - self.target).fetch("KEY", limit=limit)
        if order == "reverse":
            keys.reverse()
        elif order == "random":
            random.shuffle(keys)

        logger.info("Found %d keys to populate" % len(keys))

        keys = keys[:max_calls]
        nkeys = len(keys)
        if not nkeys:
            return

        processes = min(_ for _ in (processes, nkeys, mp.cpu_count()) if _)

        error_list = []
        populate_kwargs = dict(
            suppress_errors=suppress_errors,
            return_exception_objects=return_exception_objects,
            make_kwargs=make_kwargs,
        )

        if processes == 1:
            for key in (
                tqdm(keys, desc=self.__class__.__name__) if display_progress else keys
            ):
                error = self._populate1(key, jobs, **populate_kwargs)
                if error is not None:
                    error_list.append(error)
        else:
            # spawn multiple processes
            self.connection.close()  # disconnect parent process from MySQL server
            del self.connection._conn.ctx  # SSLContext is not pickleable
            with mp.Pool(
                processes, _initialize_populate, (self, jobs, populate_kwargs)
            ) as pool:
                with (
                    tqdm(desc="Processes: ", total=nkeys)
                    if display_progress
                    else contextlib.nullcontext()
                ) as progress_bar:
                    for error in pool.imap(_call_populate1, keys, chunksize=1):
                        if error is not None:
                            error_list.append(error)
                        if display_progress:
                            progress_bar.update()
            self.connection.connect()  # reconnect parent process to MySQL server

        # restore original signal handler:
        if reserve_jobs:
            signal.signal(signal.SIGTERM, old_handler)

        if suppress_errors:
            return error_list

    def _populate1(
        self, key, jobs, suppress_errors, return_exception_objects, make_kwargs=None
    ):
        """
        populates table for one source key, calling self.make inside a transaction.
        :param jobs: the jobs table or None if not reserve_jobs
        :param key: dict specifying job to populate
        :param suppress_errors: bool if errors should be suppressed and returned
        :param return_exception_objects: if True, errors must be returned as objects
        :return: (key, error) when suppress_errors=True, otherwise None
        """
        make = self._make_tuples if hasattr(self, "_make_tuples") else self.make

        if jobs is None or jobs.reserve(self.target.table_name, self._job_key(key)):
            self.connection.start_transaction()
            if key in self.target:  # already populated
                self.connection.cancel_transaction()
                if jobs is not None:
                    jobs.complete(self.target.table_name, self._job_key(key))
            else:
                logger.info("Populating: " + str(key))
                self.__class__._allow_insert = True
                try:
                    make(dict(key), **(make_kwargs or {}))
                except (KeyboardInterrupt, SystemExit, Exception) as error:
                    try:
                        self.connection.cancel_transaction()
                    except LostConnectionError:
                        pass
                    error_message = "{exception}{msg}".format(
                        exception=error.__class__.__name__,
                        msg=": " + str(error) if str(error) else "",
                    )
                    if jobs is not None:
                        # show error name and error message (if any)
                        jobs.error(
                            self.target.table_name,
                            self._job_key(key),
                            error_message=error_message,
                            error_stack=traceback.format_exc(),
                        )
                    if not suppress_errors or isinstance(error, SystemExit):
                        raise
                    else:
                        logger.error(error)
                        return key, error if return_exception_objects else error_message
                else:
                    self.connection.commit_transaction()
                    if jobs is not None:
                        jobs.complete(self.target.table_name, self._job_key(key))
                finally:
                    self.__class__._allow_insert = False

    def progress(self, *restrictions, display=True):
        """
        Report the progress of populating the table.
        :return: (remaining, total) -- numbers of tuples to be populated
        """
        todo = self._jobs_to_do(restrictions)
        total = len(todo)
        remaining = len(todo - self.target)
        if display:
            print(
                "%-20s" % self.__class__.__name__,
                "Completed %d of %d (%2.1f%%)   %s"
                % (
                    total - remaining,
                    total,
                    100 - 100 * remaining / (total + 1e-12),
                    datetime.datetime.strftime(
                        datetime.datetime.now(), "%Y-%m-%d %H:%M:%S"
                    ),
                ),
                flush=True,
            )
        return remaining, total

1	"""This module defines class dj.AutoPopulate"""
2	import logging	1✔
3	import datetime	1✔
4	import traceback	1✔
5	import random	1✔
6	import inspect	1✔
7	from tqdm import tqdm	1✔
8	from .expression import QueryExpression, AndList	1✔
9	from .errors import DataJointError, LostConnectionError	1✔
10	import signal	1✔
11	import multiprocessing as mp	1✔
12	import contextlib	1✔
13
14	# noinspection PyExceptionInherit,PyCallingNonCallable
15
16	logger = logging.getLogger(__name__)	1✔
17
18
19	# --- helper functions for multiprocessing --
20
21
22	def _initialize_populate(table, jobs, populate_kwargs):	1✔
23	"""
24	Initialize the process for mulitprocessing.
25	Saves the unpickled copy of the table to the current process and reconnects.
26	"""
27	process = mp.current_process()	×
28	process.table = table	×
29	process.jobs = jobs	×
30	process.populate_kwargs = populate_kwargs	×
31	table.connection.connect() # reconnect	×
32
33
34	def _call_populate1(key):	1✔
35	"""
36	Call current process' table._populate1()
37	:key - a dict specifying job to compute
38	:return: key, error if error, otherwise None
39	"""
40	process = mp.current_process()	×
41	return process.table._populate1(key, process.jobs, **process.populate_kwargs)	×
42
43
44	class AutoPopulate:	1✔
45	"""
46	AutoPopulate is a mixin class that adds the method populate() to a Relation class.
47	Auto-populated relations must inherit from both Relation and AutoPopulate,
48	must define the property `key_source`, and must define the callback method `make`.
49	"""
50
51	_key_source = None	1✔
52	_allow_insert = False	1✔
53
54	@property	1✔
55	def key_source(self):
56	"""
57	:return: the query expression that yields primary key values to be passed,
58	sequentially, to the ``make`` method when populate() is called.
59	The default value is the join of the parent tables references from the primary key.
60	Subclasses may override they key_source to change the scope or the granularity
61	of the make calls.
62	"""
63
64	def _rename_attributes(table, props):	1✔
65	return (	1✔
66	table.proj(
67	**{
68	attr: ref
69	for attr, ref in props["attr_map"].items()
70	if attr != ref
71	}
72	)
73	if props["aliased"]
74	else table.proj()
75	)
76
77	if self._key_source is None:	1✔
78	parents = self.target.parents(	1✔
79	primary=True, as_objects=True, foreign_key_info=True
80	)
81	if not parents:	1✔
82	raise DataJointError(	×
83	"A table must have dependencies "
84	"from its primary key for auto-populate to work"
85	)
86	self._key_source = _rename_attributes(*parents[0])	1✔
87	for q in parents[1:]:	1✔
88	self._key_source = _rename_attributes(q)	1✔
89	return self._key_source	1✔
90
91	def make(self, key):	1✔
92	"""
93	Derived classes must implement method `make` that fetches data from tables
94	above them in the dependency hierarchy, restricting by the given key,
95	computes secondary attributes, and inserts the new tuples into self.
96	"""
97	raise NotImplementedError(	×
98	"Subclasses of AutoPopulate must implement the method `make`"
99	)
100
101	@property	1✔
102	def target(self):
103	"""
104	:return: table to be populated.
105	In the typical case, dj.AutoPopulate is mixed into a dj.Table class by
106	inheritance and the target is self.
107	"""
108	return self	1✔
109
110	def _job_key(self, key):	1✔
111	"""
112	:param key: they key returned for the job from the key source
113	:return: the dict to use to generate the job reservation hash
114	This method allows subclasses to control the job reservation granularity.
115	"""
116	return key	1✔
117
118	def _jobs_to_do(self, restrictions):	1✔
119	"""
120	:return: the relation containing the keys to be computed (derived from self.key_source)
121	"""
122	if self.restriction:	1✔
123	raise DataJointError(	×
124	"Cannot call populate on a restricted table. "
125	"Instead, pass conditions to populate() as arguments."
126	)
127	todo = self.key_source	1✔
128
129	# key_source is a QueryExpression subclass -- trigger instantiation
130	if inspect.isclass(todo) and issubclass(todo, QueryExpression):	1✔
131	todo = todo()	1✔
132
133	if not isinstance(todo, QueryExpression):	1✔
134	raise DataJointError("Invalid key_source value")	×
135
136	try:	1✔
137	# check if target lacks any attributes from the primary key of key_source
138	raise DataJointError(	1✔
139	"The populate target lacks attribute %s "
140	"from the primary key of key_source"
141	% next(
142	name
143	for name in todo.heading.primary_key
144	if name not in self.target.heading
145	)
146	)
147	except StopIteration:	1✔
148	pass	1✔
149	return (todo & AndList(restrictions)).proj()	1✔
150
151	def populate(	1✔
152	self,
153	*restrictions,
154	suppress_errors=False,
155	return_exception_objects=False,
156	reserve_jobs=False,
157	order="original",
158	limit=None,
159	max_calls=None,
160	display_progress=False,
161	processes=1,
162	make_kwargs=None
163	):
164	"""
165	``table.populate()`` calls ``table.make(key)`` for every primary key in
166	``self.key_source`` for which there is not already a tuple in table.
167
168	:param restrictions: a list of restrictions each restrict
169	(table.key_source - target.proj())
170	:param suppress_errors: if True, do not terminate execution.
171	:param return_exception_objects: return error objects instead of just error messages
172	:param reserve_jobs: if True, reserve jobs to populate in asynchronous fashion
173	:param order: "original"\|"reverse"\|"random" - the order of execution
174	:param limit: if not None, check at most this many keys
175	:param max_calls: if not None, populate at most this many keys
176	:param display_progress: if True, report progress_bar
177	:param processes: number of processes to use. Set to None to use all cores
178	:param make_kwargs: Keyword arguments which do not affect the result of computation
179	to be passed down to each ``make()`` call. Computation arguments should be
180	specified within the pipeline e.g. using a `dj.Lookup` table.
181	:type make_kwargs: dict, optional
182	"""
183	if self.connection.in_transaction:	1✔
184	raise DataJointError("Populate cannot be called during a transaction.")	×
185
186	valid_order = ["original", "reverse", "random"]	1✔
187	if order not in valid_order:	1✔
188	raise DataJointError(	×
189	"The order argument must be one of %s" % str(valid_order)
190	)
191	jobs = (	1✔
192	self.connection.schemas[self.target.database].jobs if reserve_jobs else None
193	)
194
195	# define and set up signal handler for SIGTERM:
196	if reserve_jobs:	1✔
197
198	def handler(signum, frame):	1✔
199	logger.info("Populate terminated by SIGTERM")	×
200	raise SystemExit("SIGTERM received")	×
201
202	old_handler = signal.signal(signal.SIGTERM, handler)	1✔
203
204	keys = (self._jobs_to_do(restrictions) - self.target).fetch("KEY", limit=limit)	1✔
205	if order == "reverse":	1✔
206	keys.reverse()	×
207	elif order == "random":	1✔
208	random.shuffle(keys)	×
209
210	logger.info("Found %d keys to populate" % len(keys))	1✔
211
212	keys = keys[:max_calls]	1✔
213	nkeys = len(keys)	1✔
214	if not nkeys:	1✔
215	return	1✔
216
217	processes = min(_ for _ in (processes, nkeys, mp.cpu_count()) if _)	1✔
218
219	error_list = []	1✔
220	populate_kwargs = dict(	1✔
221	suppress_errors=suppress_errors,
222	return_exception_objects=return_exception_objects,
223	make_kwargs=make_kwargs,
224	)
225
226	if processes == 1:	1✔
227	for key in (	1✔
228	tqdm(keys, desc=self.__class__.__name__) if display_progress else keys
229	):
230	error = self._populate1(key, jobs, **populate_kwargs)	1✔
231	if error is not None:	1✔
232	error_list.append(error)	1✔
233	else:
234	# spawn multiple processes
235	self.connection.close() # disconnect parent process from MySQL server	1✔
236	del self.connection._conn.ctx # SSLContext is not pickleable	1✔
237	with mp.Pool(	1✔
238	processes, _initialize_populate, (self, jobs, populate_kwargs)
239	) as pool:
240	with (	1✔
241	tqdm(desc="Processes: ", total=nkeys)
242	if display_progress
243	else contextlib.nullcontext()
244	) as progress_bar:
245	for error in pool.imap(_call_populate1, keys, chunksize=1):	1✔
246	if error is not None:	1✔
247	error_list.append(error)	×
248	if display_progress:	1✔
249	progress_bar.update()	×
250	self.connection.connect() # reconnect parent process to MySQL server	1✔
251
252	# restore original signal handler:
253	if reserve_jobs:	1✔
254	signal.signal(signal.SIGTERM, old_handler)	1✔
255
256	if suppress_errors:	1✔
257	return error_list	1✔
258
259	def _populate1(	1✔
260	self, key, jobs, suppress_errors, return_exception_objects, make_kwargs=None
261	):
262	"""
263	populates table for one source key, calling self.make inside a transaction.
264	:param jobs: the jobs table or None if not reserve_jobs
265	:param key: dict specifying job to populate
266	:param suppress_errors: bool if errors should be suppressed and returned
267	:param return_exception_objects: if True, errors must be returned as objects
268	:return: (key, error) when suppress_errors=True, otherwise None
269	"""
270	make = self._make_tuples if hasattr(self, "_make_tuples") else self.make	1✔
271
272	if jobs is None or jobs.reserve(self.target.table_name, self._job_key(key)):	1✔
273	self.connection.start_transaction()	1✔
274	if key in self.target: # already populated	1✔
275	self.connection.cancel_transaction()	×
276	if jobs is not None:	×
277	jobs.complete(self.target.table_name, self._job_key(key))	×
278	else:
279	logger.info("Populating: " + str(key))	1✔
280	self.__class__._allow_insert = True	1✔
281	try:	1✔
282	make(dict(key), **(make_kwargs or {}))	1✔
283	except (KeyboardInterrupt, SystemExit, Exception) as error:	1✔
284	try:	1✔
285	self.connection.cancel_transaction()	1✔
286	except LostConnectionError:	×
287	pass	×
288	error_message = "{exception}{msg}".format(	1✔
289	exception=error.__class__.__name__,
290	msg=": " + str(error) if str(error) else "",
291	)
292	if jobs is not None:	1✔
293	# show error name and error message (if any)
294	jobs.error(	1✔
295	self.target.table_name,
296	self._job_key(key),
297	error_message=error_message,
298	error_stack=traceback.format_exc(),
299	)
300	if not suppress_errors or isinstance(error, SystemExit):	1✔
301	raise	1✔
302	else:
303	logger.error(error)	1✔
304	return key, error if return_exception_objects else error_message	1✔
305	else:
306	self.connection.commit_transaction()	1✔
307	if jobs is not None:	1✔
308	jobs.complete(self.target.table_name, self._job_key(key))	×
309	finally:
310	self.__class__._allow_insert = False	1✔
311
312	def progress(self, *restrictions, display=True):	1✔
313	"""
314	Report the progress of populating the table.
315	:return: (remaining, total) -- numbers of tuples to be populated
316	"""
317	todo = self._jobs_to_do(restrictions)	1✔
318	total = len(todo)	1✔
319	remaining = len(todo - self.target)	1✔
320	if display:	1✔
321	print(	1✔
322	"%-20s" % self.__class__.__name__,
323	"Completed %d of %d (%2.1f%%) %s"
324	% (
325	total - remaining,
326	total,
327	100 - 100 * remaining / (total + 1e-12),
328	datetime.datetime.strftime(
329	datetime.datetime.now(), "%Y-%m-%d %H:%M:%S"
330	),
331	),
332	flush=True,
333	)
334	return remaining, total	1✔

datajoint / datajoint-python / #12275

Source File Press 'n' to go to next uncovered line, 'b' for previous

Source File
Press 'n' to go to next uncovered line, 'b' for previous