-
-
Notifications
You must be signed in to change notification settings - Fork 6
/
jobs.py
1598 lines (1316 loc) · 52.1 KB
/
jobs.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
'''
Job resource input/output control using Redis as a locking layer
Copyright 2016-2023 Dr. Josiah Lee Carlson, Ph.D.
This library licensed under the GNU LGPL v2.1
The initial library requirements and implementation were done for OpenMail LLC.
(now System1 LLC.) jobs.py (this library) was more or less intended to offer
input and output control like Luigi and/or Airflow (both Python packages), with
fewer hard integration requirements. In fact, jobs.py has been used successfully
as part of jobs running in a cron schedule via Jenkins, in build chains in
Jenkins, inside individual rpqueue tasks, and even inside individual Flask web
requests for some high-value data (jobs.py is backed by Redis, so job locking
overhead *can* be low, even when you need to keep data safe).
Note: any company that I (Josiah Carlson, the author) introduce or use this
library, automatically has a license to use / deploy / ship jobs.py as part of a
docker layer, or other similar virtualization component for dev / staging / test
/ or production use. Basically I'm trying to say to any lawyers / lawyer-type
people: if I worked for the company, or the company you bought, and they are
using this library, it's fine.
If I *didn't* work for your company, and you want the *same* rights as ^^^, just
buy a commercial license. Monthly, annual, and *lifetime* rates are very
reasonable, give you the warm and fuzzies for supporting open-source software,
and give you actual *rights* to some level of support from the author.
Source: https://github.com/josiahcarlson/jobs/
PyPI: https://pypi.python.org/pypi/jobspy/
Docs: https://pythonhosted.org/jobspy/
Features
========
Input/output locking on multiple *named* keys, called "inputs" and "outputs".
* All keys are case-sensitive
* Multiple readers on input keys
* Exclusive single writer on output keys (no readers or other writers)
* All inputs must have been an output previously
* Optional global and per-job history of sanitized input/output edges (enabled
by default)
* Lock multiple inputs and outputs simultaneously, e.g. to produce outputs Y and
Z, I need to consume inputs A, B, C.
How to use
==========
* Install jobs.py::
$ sudo pip install jobspy
* Import jobs.py and configure the Redis connection *required* (maybe put this
in some configuration file)::
# in myjob.py or local_config.py
import jobs
jobs.CONN = redis.Redis(...)
* Use as a decorator on a function (must explicitly .start() the job, .stop()
performed automatically if left uncalled)::
# in myjob.py
@jobs.resource_manager(['input1', 'input2'], ['output1', 'output2'], duration=300, wait=900)
def some_job(job):
job.add_inputs('input6', 'input7')
job.add_outputs(...)
job.start()
# At this point, all inputs and outputs are locked according to the
# locking semantics specified in the documentation.
# If you call job.stop(failed=True), then the outputs will not be
# "written"
#job.stop(failed=True)
# If you call job.stop(), then the outputs will be "written"
job.stop()
# Alternating job.stop() with job.start() is okay! You will drop the
# locks in the .stop(), but will (try to) get them again with the
# .start()
job.start()
# But if you need the lock for longer than the requested duration, you
# can also periodically refresh the lock. The lock is only actually
# refreshed once per second at most, and you can only refresh an already
# started lock.
job.refresh()
# If an exception is raised and not caught before the decorator catches
# it, the job will be stopped by the decorator, as though failed=True:
raise Exception("Oops!")
# will stop the job the same as
# job.stop(failed=True)
# ... where the exception will bubble up out of the decorator.
* Or use as a context manager for automatic start/stop calling, with the same
exception handling semantics as the decorator::
def multi_step_job(arg1, arg2, ...):
with jobs.ResourceManager([arg1], [arg2], duration=30, wait=60, overwrite=True) as job:
for something in loop:
# do something
job.refresh()
if bad_condition:
raise Exception("Something bad happened, don't mark arg2 as available")
elif other_bad_condition:
# stop the job, not setting
job.stop(failed=True)
# arg2 should exist after it was an output, and we didn't get an
# exception... though if someone else is writing to it immediately in
# another call, then this may block...
with jobs.ResourceManager([arg2], ['output.x'], duration=60, wait=900, overwrite=True):
# something else
pass
# output.x should be written if the most recent ResourceManager stopped
# cleanly.
return
More examples
-------------
* Scheduled at 1AM UTC (5/6PM Pacific, depending on DST)::
import datetime
FMT = '%Y-%m-%d'
def yesterday():
return (datetime.datetime.utcnow().date() - datetime.timedelta(days=1)).strftime(FMT)
@jobs.resource_manager([jobs.NG.reporting.events], (), 300, 900)
def aggregate_daily_events(job):
yf = yesterday()
# outputs 'reporting.events_by_partner.YYYY-MM-DD'
# we can add job inputs and outputs inside a decorated function before
# we call .start()
job.add_outputs(jobs.NG.reporting.events_by_partner[yf])
job.start()
# actually aggregate events
* Scheduled the next day around the time when we expect upstream reporting to
be available::
@jobs.resource_manager((), (), 300, 900)
def fetch_daily_revenue(job):
yf = yesterday()
job.add_outputs(jobs.NG.reporting.upsteam_revenue[yf])
job.start()
# actually fetch daily revenue
* Executed downstream of fetch_daily_revenue()::
@jobs.resource_manager((), (), 300, 900)
def send_reports(job):
yf = yesterday()
# having jobs inputs here ensures that both of the *expected* upstream
# flows were *actual*
job.add_inputs(
jobs.NG.reporting.events_by_partner[yf],
jobs.NG.reporting.upstream_revenue[yf]
)
job.add_outputs(jobs.NG.reporting.report_by_partner[yf])
job.start()
# inputs are available, go ahead and generate the reports!
* And in other contexts...::
def make_recommendations(partners):
yf = yesterday()
for partner in partners:
with jobs.ResourceManager([jobs.NG.reporting.report_by_partner[yf]],
[jobs.NG.reporting.recommendations_by_partner[yf][partner]], 300, 900):
# job is already started
# generate the recommendations for the partner
pass
Configuration options
=====================
All configuration options are available as options on the jobs.py module itself,
though you *can* override the connection explicitly on a per-job basis. See the
'Connection configuration' section below for more details.::
# The Redis connection, REQUIRED!
jobs.CONN = redis.Redis()
# Sets a prefix to be used on all keys stored in Redis (optional)
jobs.GLOBAL_PREFIX = ''
# Keep a sanitized ZSET of inputs and outputs, available for traversal
# later. Note: sanitization runs the following on all edges before storage:
# edge = re.sub('[0-9][0-9-]*', '*', edge)
# ... which allows you to get a compact flow graph even in cases where you
# have day-parameterized builds.
jobs.GRAPH_HISTORY = True
# Sometimes you don't want your outputs to last forever (sometimes history
# should be forgotten, right?), and jobs.py gives you the chance to say as
# much.
# By default, a `None` duration means that outputs will last forever. Any
# other value will be used in a call to `expire` on the associated output
# keys after they are set on a job's successful completion. This value is in
# seconds.
jobs.OUTPUT_DURATION = None
# To use a logger that doesn't print to standard output, set the logging
# object at the module level (see below). By default, the built-in "default
# logger" prints to standard output.
jobs.DEFAULT_LOGGER = logging.getLogger(...)
Using jobs.py with a custom Redis configuration
===============================================
If you would like to use jobs.py as a script (for the convenient command-line
options), you need to create a wrapper module, which can also act as your
general configuration updates for jobs.py (hack because I needed to release
this as open-source before the end of summer)::
# myjobs.py
import jobs
jobs.CONN = ...
jobs.DEFAULT_LOGGER = ...
jobs.GLOBAL_PREFIX = ...
jobs.GRAPH_HISTORY = ...
jobs.OUTPUT_DURATION = ...
from jobs import *
if __name__ == '__main__':
main()
Then you can use this as::
$ python myjobs.py --help
And you can use ``myjobs.py`` everywhere, which will have all of your
configuration handled.::
# daily_report.py
import myjobs
@myjobs.resource_manager(...)
def daily_reporting(job, ...):
# exactly the same as before.
'''
from __future__ import print_function
import argparse
import atexit
import binascii
from collections import defaultdict, deque
from datetime import datetime, date
import functools
from hashlib import sha1
import json
import logging
import os
import re
import signal
import sys
import threading
import time
import traceback
import redis.exceptions
_all = set(globals())
VERSION = '0.29.0'
# user-settable configuration
CONN = None
GLOBAL_PREFIX = ''
GRAPH_HISTORY = True
OUTPUT_DURATION = None
DEFAULT_LOGGER = None # actually set below, see BullshitLog()
# end user-settable configuration
EDGE_RE = re.compile('[0-9][0-9-]*')
TS_RE = re.compile(r'^(\d{9,}(?:[.]\d*)?)$')
DT_RE = re.compile(r'^(\d{4})-(\d{2})-(\d{2})(?: (\d{2}):(\d{2})(?::(\d{2})(?:[.](?:\d+)?)?)?)?$')
EPOCH = datetime(1970, 1, 1)
PY3K = sys.version_info >= (3, 0, 0)
TEXT_TYPE = str if PY3K else unicode
LOCKED = set()
AUTO_REFRESH = set()
REFRESH_THREAD = None
_GHD = object()
NT = type(None)
_NT = (NT, int)
class ResourceUnavailable(Exception):
'''
Raised when one or more inputs are unavailable, or when one or more outputs
are already locked.
'''
class NG(object):
'''
Convenience object for generating names:
>>> str(NG.foo.bar.baz[1].goo)
foo.bar.baz.1.goo
'''
__slots__ = '_name',
def __init__(self, start=''):
self._name = start.strip('.')
def __getitem__(self, item):
return self.__class__('%s.%s'%(self._name, item))
__getattr__ = __getitem__
def __call__(self, item):
return self[item]
def __str__(self):
return self._name
def __repr__(self):
return repr(str(self))
def __eq__(self, other):
return str(self) == str(other)
def __hash__(self):
return hash(str(self))
TYPE_NG = NG
NG = NG()
@atexit.register
def _signal_handler(*args, **kwargs):
for m in list(LOCKED):
m.stop(failed=True, shutting_down=True)
if args:
# call the old handler, as necessary
if OLD_SIGNAL:
OLD_SIGNAL(*args, **kwargs)
raise SystemExit(128 + args[0])
# register new signal handler, and keep reference to the old one (if any)
OLD_SIGNAL = signal.signal(signal.SIGTERM, _signal_handler)
ATEXIT_SET = False
SIGNAL_SET = False
OLD_SIGNAL = None
def handle_auto_shutdown():
global ATEXIT_SET, SIGNAL_SET, OLD_SIGNAL
if not ATEXIT_SET:
ATEXIT_SET = atexit.register(_signal_handler)
if not SIGNAL_SET and isinstance(threading.current_thread(), threading._MainThread):
SIGNAL_SET, OLD_SIGNAL = True, signal.signal(signal.SIGTERM, _signal_handler)
def resource_manager(inputs, outputs, duration, wait=None, overwrite=True,
conn=None, graph_history=_GHD, suffix=None, output_duration=None):
'''
Arguments:
* inputs - the list of inputs that need to exist to start the job
* outputs - the list of outputs to produce
* duration - how long you want to lock the inputs and outputs from
modification from other jobs
* wait=None - how long to wait for inputs to be available and for
when overwrite=True, how long to wait for other writers to
finish writing
* overwrite=False - whether to overwrite a pre-existing output if it
already exists
* conn=None - a Redis connection to use (provide here, or when
calling .start())
* graph_history=True - whether to keep history of graph edges
* output_duration=None - how long to keep outputs in Redis, defaults
to forever
'''
def wrap(fcn):
@functools.wraps(fcn)
def call(*args, **kwargs):
manager = ResourceManager(inputs, outputs, duration, wait,
overwrite, conn, graph_history, _caller_name(fcn), suffix,
output_duration)
ex = False
try:
return fcn(manager, *args, **kwargs)
except:
ex = True
raise
finally:
manager.stop(failed=ex)
return call
return wrap
class ResourceManager(object):
def __init__(self, inputs, outputs, duration, wait=None, overwrite=True,
conn=None, graph_history=_GHD, identifier=None, suffix=None,
output_duration=None):
'''
Arguments:
* inputs - the list of inputs that need to exist to start the job
* outputs - the list of outputs to produce
* duration - how long you want to lock the inputs and outputs from
modification from other jobs
* wait=None - how long to wait for inputs to be available and for
when overwrite=True, how long to wait for other writers to
finish writing
* overwrite=False - whether to overwrite a pre-existing output if it
already exists
* conn=None - a Redis connection to use (provide here, or when
calling .start())
* graph_history=True - whether to keep history of graph edges
* output_duration=None - how long to keep outputs in Redis, defaults
to forever
'''
assert isinstance(inputs, (list, tuple, set)), inputs
assert isinstance(outputs, (list, tuple, set)), outputs
self.inputs = list(inputs)
self.outputs = list(outputs)
self.duration = max(duration, 0)
self.wait = max(wait or 0, 0)
self.overwrite = overwrite
self.last_refreshed = None
self.prefix_identifier(identifier or _caller_name(_get_caller()))
self.conn = conn
self.graph_history = GRAPH_HISTORY if graph_history is _GHD else graph_history
self.auto_refresh = None
if not isinstance(output_duration, _NT):
raise TypeError("output_duration must be None or integer")
if isinstance(output_duration, int):
if output_duration <= 0:
raise ValueError("output_duration must be > 0")
self.output_duration = output_duration or OUTPUT_DURATION
self._lock = threading.RLock()
# This is a symptom of bad design. But it exists because I need the
# functionality. Practicality beats purity.
self.suffix = suffix
def add_inputs(self, *inputs):
'''
Adds inputs before the job has started.
'''
if self.is_running:
raise RuntimeError("Can't add inputs after starting")
self.inputs.extend(inputs)
def add_outputs(self, *outputs):
'''
Adds outputs before the job as started.
'''
if self.is_running:
raise RuntimeError("Can't add outputs after starting")
self.outputs.extend(outputs)
@property
def identifier(self):
'''
Property to allow for .suffix to be set after job creation.
'''
s = (self.suffix or '').strip('.')
s = ('.' + s) if s else ''
return str(self._identifier) + s
def prefix_identifier(self, base_identifier):
'''
Will set a new identifier derived from the provided base_identifier by
adding a ``.<random string>`` suffix. Ensures that otherwise identical
names from job runners don't get confused about who is running what.
The final identifier to be used will be the base_identifier provided here,
a 48 bit random numeric identifier, followed by an optional .suffix:
``<base_identifier>.<random string>[.suffix]``
'''
if self.is_running:
raise RuntimeError("Can't set the identifier after starting")
# generate a 48 bit identifier using os.urandom, use decimal not hex
self._identifier = NG(base_identifier)[int(binascii.hexlify(os.urandom(6)), 16)]
def can_run(self, conn=None):
'''
Will return whether the job can be run immediately, but does not start
the job.
'''
conn = conn or self.conn or CONN
if not conn:
raise RuntimeError("Cannot start a job without a connection to Redis!")
if self.is_running:
raise RuntimeError("Already started!")
return _run_if_possible(conn, self.inputs, self.outputs, self.identifier, 0, self.overwrite)
def refresh(self, lost_lock_fail=False, **kwargs):
'''
For jobs that may take longer than the provided "duration", you should
.refresh() periodically to ensure that someone doesn't overwrite your
inputs or outputs.
Arguments:
* lock_lost_fail - fail if any lock was lost, and raise an exception
Note: will only refresh at most once/second.
'''
inside_auto_refresh = kwargs.get('inside_auto_refresh')
with self._lock:
if self.is_running and time.time() - self.last_refreshed > 1:
DEFAULT_LOGGER.debug("Refreshing job locks")
lost = _refresh_job(self.conn, self.inputs, self.outputs,
self.identifier, self.duration, self.overwrite)
if lost.get('err') or lost.get('temp'):
if lost_lock_fail:
auto = inside_auto_refresh and self.auto_refresh
self.stop(failed=True)
if not auto:
raise ResourceUnavailable(lost.get('err'))
DEFAULT_LOGGER.warning("Lock(s) lost due to timeout: %r", lost)
self.last_refreshed = time.time()
return lost
def start(self, conn=None, auto_refresh=None, **kwargs):
'''
Will attempt to start the run within self.wait seconds, waiting for:
* inputs to be available
* outputs to not be locked for read or write
* outputs to not exist when ``overwrite=False``
If unable to start within self.wait seconds, will raise an exception
showing the bad/missing resources.
If ``auto_refresh`` is provided, and can be considered boolean ``True``,
a background thread will try to call ``job.refresh()`` on this lock
once per second, until the job is explicitly stopped with ``.stop()``
or the process exits, whichever comes first.
'''
try:
with self._lock:
return self._start(conn, auto_refresh, **kwargs)
finally:
if self.is_running and self.auto_refresh:
_start_auto_refresh(self)
def _start(self, conn, auto_refresh, **kwargs):
self.conn = conn or self.conn or CONN
if not self.conn:
raise RuntimeError("Cannot start a job without a connection to Redis!")
if self.is_running:
return
if not self.identifier or not isinstance(self.identifier, (str, TYPE_NG)):
raise RuntimeError("Can't start job without a valid identifier")
if LOCKED and not kwargs.pop('i_really_know_what_i_am_doing_dont_warn_me', None):
DEFAULT_LOGGER.warning("Trying to start job while another job has "
"already started in the same process is a recipe for deadlocks. "
"You should probably stop doing that unless you know what you "
"are doing.")
result = {'ok': False, 'err': {}}
DEFAULT_LOGGER.info("Trying to start job with inputs: %r and outputs: %r",
self.inputs, self.outputs)
def tr():
DEFAULT_LOGGER.debug("Trying to start job")
result = _run_if_possible(self.conn, self.inputs, self.outputs,
self.identifier, self.duration, self.overwrite,
history=self.graph_history)
if result['ok']:
DEFAULT_LOGGER.info("Starting job")
self.last_refreshed = time.time()
self.auto_refresh = bool(auto_refresh)
LOCKED.add(self)
return result, True
else:
DEFAULT_LOGGER.debug("Failed to start job: %r", result)
return result, False
# Report that we are still waiting after waiting for 1 second, the first
# time.
last_reported = time.time() - 29
stop_waiting = time.time() + max(self.wait or 0, 0)
while time.time() < stop_waiting:
result, s = tr()
if s:
return self
if 'output_exists' in result['err']:
# We can't recover from "output exists" errors without
# overwriting the output, and we only get the error when we
# can't overwrite the output. Don't bother waiting any longer.
break
# Only print a message reporting the waiting status once every 30
# seconds
if time.time() - last_reported >= 30:
DEFAULT_LOGGER.info("Still waiting to start job... %r", result['err'])
last_reported = time.time()
# Wait up to 10ms between tests
time.sleep(min(max(stop_waiting - time.time(), 0), .01))
# try one more time before bailing out...
result, s = tr()
if s:
return self
DEFAULT_LOGGER.info("Failed to start job: %r", result['err'])
raise ResourceUnavailable(result['err'])
@property
def is_running(self):
'''
Returns whether or not the job is running.
'''
return self.last_refreshed is not None
def stop(self, failed=False, shutting_down=False):
'''
Stops a job if running. If the optional "failed" argument is true,
outputs will not be set as available.
'''
if self.is_running:
with self._lock:
if not self.is_running:
# another thread could have changed the status
return
failed = bool(failed)
DEFAULT_LOGGER.info("Stopping job failed = %r", bool(failed))
if shutting_down:
DEFAULT_LOGGER.warning("Stopping job as part of atexit/signal handler exit")
try:
_finish_job(self.conn, self.inputs, self.outputs, self.identifier,
failed=failed, od=self.output_duration)
finally:
self.last_refreshed = None
self.auto_refresh = None
LOCKED.discard(self)
AUTO_REFRESH.discard(self)
def __enter__(self):
return self.start(self.conn)
def __exit__(self, typ, value, tb):
self.stop(bool(typ or value or tb))
def _create_outputs(outputs, conn=None, identifier=None, suffix=None):
'''
Sometimes you just need outputs to exist. These creates outputs.
'''
identifier = NG(identifier or _caller_name(_get_caller()))
if suffix:
identifier = identifier[suffix]
(conn or CONN).mset({str(o):str(identifier) for o in outputs})
def _force_unlock(inputs, outputs, conn=None):
'''
Sometimes you just need to unlock some inputs and outputs. This unlocks
inputs and outputs.
'''
inputs = [i if i.startswith('ilock:') else ('ilock:' + i) for i in inputs]
outputs = [o if o.startswith('olock:') else ('olock:' + o) for o in outputs]
io = inputs + outputs
if io:
return (conn or CONN).delete(*io)
def _check_inputs_and_outputs(fcn):
@functools.wraps(fcn)
def call(conn, inputs, outputs, identifier, *a, **kw):
assert isinstance(inputs, (list, tuple, set)), inputs
assert isinstance(outputs, (list, tuple, set)), outputs
assert '' not in inputs, inputs
assert '' not in outputs, outputs
# this is for actually locking inputs/outputs
inputs, outputs = list(map(str, inputs)), list(map(str, outputs))
locks = inputs + [''] + outputs
if kw.pop('history', None):
igraph = [EDGE_RE.sub('*', inp) for inp in inputs]
ograph = [EDGE_RE.sub('*', out) for out in outputs]
graph_id = EDGE_RE.sub('*', str(identifier))
graph = igraph + [''] + ograph + ['', graph_id]
if all(x.startswith('test.') for x in igraph + ograph):
graph = ['', '']
else:
graph = ['', '']
return fcn(conn, locks, graph, str(identifier), *a, **kw)
return call
def _fix_err(result):
# Translate list of error types to a dictionary of grouped errors.
def _fix(d):
err = defaultdict(list)
for why, key in d:
err[why].append(key)
return dict(err)
if result.get('err'):
result['err'] = _fix(result['err'])
if result.get('temp'):
result['temp'] = _fix(result['temp'])
return result
@_check_inputs_and_outputs
def _run_if_possible(conn, inputs_outputs, graph, identifier, duration, overwrite):
'''
Internal call to run a job if possible, only acquiring the locks if all are
available.
'''
return _fix_err(json.loads(_run_if_possible_lua(conn, keys=inputs_outputs,
args=[json.dumps({
'prefix': GLOBAL_PREFIX,
'id': identifier,
'now': time.time(),
'duration': duration,
'overwrite': bool(overwrite),
'refresh': False,
'edges': graph})]
).decode('latin-1')))
@_check_inputs_and_outputs
def _refresh_job(conn, inputs_outputs, graph, identifier, duration, overwrite):
'''
Internal call to refresh a job that already has a lock.
'''
return _fix_err(json.loads(_run_if_possible_lua(conn, keys=inputs_outputs,
args=[json.dumps({
'prefix': GLOBAL_PREFIX,
'id': identifier,
'now': time.time(),
'duration': duration,
'overwrite': bool(overwrite),
'refresh': True,
'edges': []})]
).decode('latin-1')))
@_check_inputs_and_outputs
def _finish_job(conn, inputs_outputs, graph, identifier, failed=False, od=None):
'''
Internal call to finish a job.
'''
args = [identifier, time.time(), not failed, GLOBAL_PREFIX]
if od:
args.append(od)
_finish_job_lua(conn, keys=inputs_outputs,
args=[json.dumps(args)]
)
def _caller_name(code):
if callable(code):
code = code.__code__
return "%s:%s"%(code.co_filename, code.co_name)
def _get_caller():
return sys._getframe(2).f_code
NO_SCRIPT_MESSAGES = ['NOSCRIPT', 'No matching script.']
def _script_load(script):
'''
Variant of my public-domain script load from my book. Also LGPL from my
other library.
https://github.com/josiahcarlson/rom/blob/master/rom/util.py
'''
script = script.encode('utf-8') if isinstance(script, TEXT_TYPE) else script
sha = [None, sha1(script).hexdigest()]
def call(conn, keys=[], args=[], force_eval=False):
keys = tuple(keys)
args = tuple(args)
if not force_eval:
if not sha[0]:
try:
# executing the script implicitly loads it
return conn.execute_command(
'EVAL', script, len(keys), *(keys + args))
finally:
# thread safe by re-using the GIL ;)
del sha[:-1]
try:
return conn.execute_command(
"EVALSHA", sha[0], len(keys), *(keys+args))
except redis.exceptions.ResponseError as msg:
if not any(msg.args[0].startswith(nsm) for nsm in NO_SCRIPT_MESSAGES):
raise
return conn.execute_command(
"EVAL", script, len(keys), *(keys+args))
return call
_run_if_possible_lua = _script_load('''
-- KEYS - list of inputs and outputs to lock, separated by an empty string:
-- {'input', '', 'output'}
-- ARGV - {json.dumps({
-- prefix: key_prefix,
-- id: identifier,
-- now: timestamp,
-- duration: lock_duration_in_seconds,
-- overwrite: overwrite_as_boolean,
-- refresh: refresh_as_boolean,
-- -- If there is a graph history, these edges represent them.
-- edges: [inputs, '', outputs, '', graph_id]
-- })}
local args = cjson.decode(ARGV[1])
local failures = {}
local temp_failures = {}
local is_input = true
local is_refresh = args.refresh
local graph = args.edges
local prefix = args.prefix
redis.call('zremrangebyscore', prefix .. 'jobs:running', '-inf', args.now)
-- make sure input keys are available and output keys are not yet written
for i, kk in ipairs(KEYS) do
local exists = redis.call('exists', prefix .. kk) == 1
local olock = redis.call('get', prefix .. 'olock:' .. kk)
olock = olock and olock ~= args.id
-- always clean out the input lock ZSET
local ilk = prefix .. 'ilock:' .. kk
redis.call('zremrangebyscore', ilk, 0, args.now)
local ilock = redis.call('exists', ilk) == 1
if kk == '' then
is_input = false
elseif is_input then
if olock or not exists then
if is_refresh then
-- lost our input lock
table.insert(failures, {'input_lock_lost', kk})
else
-- input doesn't exist, or input exists but someone is writing to it
table.insert(failures, {'input_missing', kk})
end
elseif is_refresh and not redis.call('zscore', ilk, args.id) then
-- lost our input lock, report the temp failure
table.insert(temp_failures, {'input_lock_lost', kk})
end
else
if exists and not args.overwrite then
-- exists, can't overwrite
table.insert(failures, {'output_exists', kk})
elseif olock then
-- the output has been locked by another process
table.insert(failures, {'output_locked', kk})
elseif ilock then
-- the output file is being read by another process
table.insert(failures, {'output_used', kk})
elseif is_refresh and not redis.call('get', prefix .. 'olock:' .. kk) then
-- lost our output lock, reacquire it
table.insert(temp_failures, {'output_lock_lost', kk})
end
end
end
if #failures > 0 then
return cjson.encode({ok=false, err=failures, temp=temp_failures})
end
if args.duration == 0 then
return cjson.encode({ok=true})
end
is_input = true
for i, kk in ipairs(KEYS) do
if kk == '' then
is_input = false
elseif is_input then
local ilock = prefix .. 'ilock:' .. kk
-- add lock for this call
redis.call('zadd', ilock, args.now + args.duration, args.id)
if redis.call('ttl', ilock) < args.duration then
-- ensure that the locks last long enough
redis.call('expire', ilock, args.duration)
end
else
-- lock the output keys to ensure that no one is concurrently writing
local olock = prefix .. 'olock:' .. kk
redis.call('setex', olock, args.duration, args.id)
end
end
redis.call('zadd', prefix .. 'jobs:running', args.now + args.duration, args.id)
redis.call('setex', prefix .. 'jobs:running:' .. args.id, args.duration, cjson.encode(KEYS))
-- keep a record of our input/output graph
if not is_refresh then
is_input = true
local id = table.remove(graph)
table.remove(graph)
for i, kk in ipairs(graph) do
if kk == '' then
is_input = false
elseif is_input then
redis.call('zadd', prefix .. 'jobs:graph:input', args.now, kk .. ' -> ' .. id)
else
redis.call('zadd', prefix .. 'jobs:graph:output', args.now, id .. ' -> ' .. kk)
end
end
end
if #temp_failures > 0 then
return cjson.encode({ok=true, temp=temp_failures})
end
return cjson.encode({ok=true})
''')
_finish_job_lua = _script_load('''
-- KEYS - list of inputs and outputs to finish the job for, same semantics as
-- _run_if_possible_lua()
-- ARGV - {json.dumps([identifier, now, success, prefix, [output_duration]])}
local args = cjson.decode(ARGV[1])
local is_input = true
local prefix = args[4]
local expire = args[5]
for i, kk in ipairs(KEYS) do
if kk == '' then
is_input = false
elseif is_input then
local ilock = prefix .. 'ilock:' .. kk
-- clean out old input locks
redis.call('zremrangebyscore', ilock, 0, args[2])
redis.call('zrem', ilock, args[1])
else
-- clean out old locks that have our identifier
local olock = prefix .. 'olock:' .. kk
if redis.call('get', olock) == args[1] then
redis.call('del', olock)
end
if args[3] then
-- set the output key to the identifier to signify the job is done
if args[5] then
redis.call('setex', prefix .. kk, args[5], args[1])
else
redis.call('set', prefix .. kk, args[1])
end
end
end
end
redis.call('zrem', prefix .. 'jobs:running', args[1])
redis.call('del', prefix .. 'jobs:running:' .. args[1])
''')
_get_job_info_lua = _script_load('''
-- ARGV - {json.dumps([now, prefix])}
local args = cjson.decode(ARGV[1])
local prefix = args[2]
local jobs = {}
local jobl = redis.call('zrangebyscore', prefix .. 'jobs:running', args[1], 'inf', 'withscores')
for i=1, #jobl, 2 do
local job = {}
job.id = jobl[i]
job.exptime = tonumber(jobl[i+1])
job.io = cjson.decode(redis.call('get', prefix .. 'jobs:running:' .. jobl[i]))
table.insert(jobs, job)
end
return cjson.encode(jobs)
''')
class BullshitLog(object):
level = 20