process_controller.py 39.4 KB
Newer Older
1
2
# -*- coding: utf-8 -*-

3
from __future__ import (division, print_function, unicode_literals, absolute_import)
4
5
6

import numpy as np
from pandas import DataFrame
7
8
9
10
import datetime
import os
import time
from itertools import chain
11
import signal
12
import re
13
from typing import TYPE_CHECKING
14
import shutil
15

16
17
from ..io import output_writer as OUT_W
from ..io import input_reader as INP_R
18
19
20
21
22
from ..misc import database_tools as DB_T
from ..misc import helper_functions as HLP_F
from ..misc.path_generator import path_generator
from ..misc.logging import GMS_logger, shutdown_loggers
from ..algorithms import L1A_P, L1B_P, L1C_P, L2A_P, L2B_P, L2C_P
23
from ..model.metadata import get_LayerBandsAssignment
24
from ..model.gms_object import failed_GMS_object, GMS_object
25
from .pipeline import (L1A_map, L1A_map_1, L1A_map_2, L1A_map_3, L1B_map, L1C_map,
26
                       L2A_map, L2B_map, L2C_map)
27
from ..options.config import set_config
28
from .multiproc import MAP, imap_unordered
29
from ..misc.definition_dicts import proc_chain, db_jobs_statistics_def
30
from ..misc.locks import release_unclosed_locks
31
from ..version import __version__, __versionalias__
32

33
34
from py_tools_ds.numeric.array import get_array_tilebounds

35
if TYPE_CHECKING:
Daniel Scheffler's avatar
Daniel Scheffler committed
36
    from collections import OrderedDict  # noqa F401  # flake8 issue
37
38
    from typing import List  # noqa F401  # flake8 issue
    from ..options.config import GMS_config  # noqa F401  # flake8 issue
39
40
41
42


__author__ = 'Daniel Scheffler'

43
44

class process_controller(object):
45
    def __init__(self, job_ID, **config_kwargs):
46
        """gms_preprocessing process controller
47

48
49
        :param job_ID:          job ID belonging to a valid database record within table 'jobs'
        :param config_kwargs:   keyword arguments to be passed to gms_preprocessing.set_config()
50
51
52
        """

        # assertions
53
54
        if not isinstance(job_ID, int):
            raise ValueError("'job_ID' must be an integer value. Got %s." % type(job_ID))
55

56
57
        # set GMS configuration
        config_kwargs.update(dict(reset_status=True))
58
        self.config = set_config(job_ID, **config_kwargs)  # type: GMS_config
59
60

        # defaults
61
        self._logger = None
62
        self._DB_job_record = None
63
        self.profiler = None
64
65
66
67
68

        self.failed_objects = []
        self.L1A_newObjects = []
        self.L1B_newObjects = []
        self.L1C_newObjects = []
69
        self.L2A_newObjects = []
70
        self.L2A_tiles = []
71
72
73
74
        self.L2B_newObjects = []
        self.L2C_newObjects = []

        self.summary_detailed = None
75
        self.summary_quick = None
76

77
78
        # check if process_controller is executed by debugger
        # isdebugging = 1 if True in [frame[1].endswith("pydevd.py") for frame in inspect.stack()] else False
79
        # if isdebugging:  # override the existing settings in order to get write access everywhere
80
81
        #    pass

82
        # called_from_iPyNb = 1 if 'ipykernel/__main__.py' in sys.argv[0] else 0
83

84
85
86
87
        path_job_logfile = os.path.join(self.config.path_job_logs, '%s.log' % self.config.ID)
        if os.path.exists(path_job_logfile):
            HLP_F.silentremove(path_job_logfile)

88
        self.logger.info("Executing gms_preprocessing, version: %s (%s)" % (__version__, __versionalias__))
89
        self.logger.info('Process Controller initialized for job ID %s (comment: %s).'
90
                         % (self.config.ID, self.DB_job_record.comment))
91

92
        if self.config.delete_old_output:
93
94
            self.logger.info('Deleting previously processed data...')
            self.DB_job_record.delete_procdata_of_entire_job(force=True)
95

96
97
98
99
100
    @property
    def logger(self):
        if self._logger and self._logger.handlers[:]:
            return self._logger
        else:
101
102
            self._logger = GMS_logger('log__%s' % self.config.ID, fmt_suffix='ProcessController',
                                      log_level=self.config.log_level, append=True, log_to_joblog=True)
103
104
105
106
107
108
109
110
111
112
            return self._logger

    @logger.setter
    def logger(self, logger):
        self._logger = logger

    @logger.deleter
    def logger(self):
        if self._logger not in [None, 'not set']:
            self.logger.close()
113
            self.logger = None  # FIXME _logger = None?
114
115

    @property
116
117
118
119
    def DB_job_record(self):
        if self._DB_job_record:
            return self._DB_job_record
        else:
120
121
            self._DB_job_record = DB_T.GMS_JOB(self.config.conn_database)
            self._DB_job_record.from_job_ID(self.config.ID)
122
            return self._DB_job_record
123

124
125
126
    @DB_job_record.setter
    def DB_job_record(self, value):
        self._DB_job_record = value
127

128
129
130
    @property
    def sceneids_failed(self):
        return [obj.scene_ID for obj in self.failed_objects]
131

132
133
    def _add_local_availability_single_dataset(self, dataset):
        # type: (OrderedDict) -> OrderedDict
134
        # TODO revise this function
135
136
        # query the database and get the last written processing level and LayerBandsAssignment
        DB_match = DB_T.get_info_from_postgreSQLdb(
137
            self.config.conn_database, 'scenes_proc', ['proc_level', 'layer_bands_assignment'],
138
            dict(sceneid=dataset['scene_ID']))
Daniel Scheffler's avatar
Daniel Scheffler committed
139

140
        # get the corresponding logfile
141
142
        path_logfile = path_generator(dataset).get_path_logfile(merged_subsystems=False)
        path_logfile_merged_ss = path_generator(dataset).get_path_logfile(merged_subsystems=True)
143
144
145
146

        def get_AllWrittenProcL_dueLog(path_log):  # TODO replace this by database query + os.path.exists
            """Returns all processing level that have been successfully written according to logfile."""

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
147
148
149
150
            if not os.path.exists(path_log):
                if path_log == path_logfile:  # path_logfile_merged_ss has already been searched
                    self.logger.info("No logfile named '%s' found for %s at %s. Dataset has to be reprocessed."
                                     % (os.path.basename(path_log), dataset['entity_ID'], os.path.dirname(path_log)))
151
152
153
154
                AllWrittenProcL_dueLog = []
            else:
                logfile = open(path_log, 'r').read()
                AllWrittenProcL_dueLog = re.findall(":*(\S*\s*) data successfully saved.", logfile, re.I)
155
                if not AllWrittenProcL_dueLog and path_logfile == path_logfile_merged_ss:  # AllWrittenProcL_dueLog = []
156
157
158
159
160
161
162
163
164
165
                    self.logger.info('%s: According to logfile no completely processed data exist at any '
                                     'processing level. Dataset has to be reprocessed.' % dataset['entity_ID'])
                else:
                    AllWrittenProcL_dueLog = HLP_F.sorted_nicely(list(set(AllWrittenProcL_dueLog)))
            return AllWrittenProcL_dueLog

        # check if there are not multiple database records for this dataset
        if len(DB_match) == 1 or DB_match == [] or DB_match == 'database connection fault':

            # get all processing level that have been successfully written
166
167
            # NOTE: first check for merged subsystem datasets because they have hiver processing levels
            AllWrittenProcL = get_AllWrittenProcL_dueLog(path_logfile_merged_ss)
168
            if not AllWrittenProcL:
169
                AllWrittenProcL = get_AllWrittenProcL_dueLog(path_logfile)
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
170
171
172
            else:
                # A L2A+ dataset with merged subsystems has been found. Use that logfile.
                path_logfile = path_logfile_merged_ss
173

174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
            dataset['proc_level'] = None  # default (dataset has to be reprocessed)

            # loop through all the found proc. levels and find the one that fulfills all requirements
            for ProcL in reversed(AllWrittenProcL):
                if dataset['proc_level']:
                    break  # proc_level found; no further searching for lower proc_levels
                assumed_path_GMS_file = '%s_%s.gms' % (os.path.splitext(path_logfile)[0], ProcL)

                # check if there is also a corresponding GMS_file on disk
                if os.path.isfile(assumed_path_GMS_file):
                    GMS_file_dict = INP_R.GMSfile2dict(assumed_path_GMS_file)
                    target_LayerBandsAssignment = \
                        get_LayerBandsAssignment(dict(
                            image_type=dataset['image_type'],
                            Satellite=dataset['satellite'],
                            Sensor=dataset['sensor'],
190
                            Subsystem=dataset['subsystem'] if path_logfile != path_logfile_merged_ss else '',
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
                            proc_level=ProcL,  # must be respected because LBA changes after atm. Corr.
                            dataset_ID=dataset['dataset_ID'],
                            logger=None), nBands=(1 if dataset['sensormode'] == 'P' else None))

                    # check if the LayerBandsAssignment of the written dataset on disk equals the
                    # desired LayerBandsAssignment
                    if target_LayerBandsAssignment == GMS_file_dict['LayerBandsAssignment']:

                        # update the database record if the dataset could not be found in database
                        if DB_match == [] or DB_match == 'database connection fault':
                            self.logger.info('The dataset %s is not included in the database of processed data but'
                                             ' according to logfile %s has been written successfully. Recreating '
                                             'missing database entry.' % (dataset['entity_ID'], ProcL))
                            DB_T.data_DB_updater(GMS_file_dict)

                            dataset['proc_level'] = ProcL

                        # if the dataset could be found in database
                        elif len(DB_match) == 1:
                            try:
                                self.logger.info('Found a matching %s dataset for %s. Processing skipped until %s.'
                                                 % (ProcL, dataset['entity_ID'],
                                                    proc_chain[proc_chain.index(ProcL) + 1]))
                            except IndexError:
                                self.logger.info('Found a matching %s dataset for %s. Processing already done.'
                                                 % (ProcL, dataset['entity_ID']))

                            if DB_match[0][0] == ProcL:
                                dataset['proc_level'] = DB_match[0][0]
                            else:
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
221
                                dataset['proc_level'] = ProcL
222

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
223
                    else:
Daniel Scheffler's avatar
Daniel Scheffler committed
224
                        self.logger.info('Found a matching %s dataset for %s but with a different '
Daniel Scheffler's avatar
Daniel Scheffler committed
225
                                         'LayerBandsAssignment (desired: %s; found %s). Dataset has to be reprocessed.'
226
227
                                         % (ProcL, dataset['entity_ID'],
                                            target_LayerBandsAssignment, GMS_file_dict['LayerBandsAssignment']))
228
229
230
231
232
                else:
                    self.logger.info('%s for dataset %s has been written due to logfile but no corresponding '
                                     'dataset has been found.' % (ProcL, dataset['entity_ID']) +
                                     ' Searching for lower processing level...'
                                     if AllWrittenProcL.index(ProcL) != 0 else '')
233

234
235
236
237
        elif len(DB_match) > 1:
            self.logger.info('According to database there are multiple matches for the dataset %s. Dataset has to '
                             'be reprocessed.' % dataset['entity_ID'])
            dataset['proc_level'] = None
238

239
240
        else:
            dataset['proc_level'] = None
241

242
        return dataset
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
243

244
245
246
    def add_local_availability(self, datasets):
        # type: (List[OrderedDict]) -> List[OrderedDict]
        """Check availability of all subsets per scene and processing level.
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
247

248
249
250
251
252
253
        NOTE: The processing level of those scenes, where not all subsystems are available in the same processing level
              is reset.

        :param datasets:    List of one OrderedDict per subsystem as generated by CFG.data_list
        """
        datasets = [self._add_local_availability_single_dataset(ds) for ds in datasets]
254

255
256
257
        ######################################################################################################
        # validate that all subsystems of the same sceneid are at the same processing level; otherwise reset #
        ######################################################################################################
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
258
259
        datasets_validated = []
        datasets_grouped = HLP_F.group_dicts_by_key(datasets, key='scene_ID')
260

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
261
262
        for ds_group in datasets_grouped:
            proc_lvls = [ds['proc_level'] for ds in ds_group]
263

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
264
265
            if not len(list(set(proc_lvls))) == 1:
                # reset processing level of those scenes where not all subsystems are available
266
                self.logger.info('%s: Found already processed subsystems at different processing levels %s. '
267
                                 'Dataset has to be reprocessed to avoid errors.'
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
268
                                 % (ds_group[0]['entity_ID'], proc_lvls))
269

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
270
271
                for ds in ds_group:
                    ds['proc_level'] = None
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
272
                    datasets_validated.append(ds)
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
273
274
            else:
                datasets_validated.extend(ds_group)
275

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
276
        return datasets_validated
277

278
279
    @staticmethod
    def _is_inMEM(GMS_objects, dataset):
Daniel Scheffler's avatar
Daniel Scheffler committed
280
        # type: (list, OrderedDict) -> bool
281
282
283
284
        """Checks whether a dataset within a dataset list has been processed in the previous processing level.
        :param GMS_objects: <list> a list of GMS objects that has been recently processed
        :param dataset:     <collections.OrderedDict> as generated by L0A_P.get_data_list_of_current_jobID()
        """
285
        # check if the scene ID of the given dataset is in the scene IDs of the previously processed datasets
286
287
288
        return dataset['scene_ID'] in [obj.scene_ID for obj in GMS_objects]

    def _get_processor_data_list(self, procLvl, prevLvl_objects=None):
289
        """Returns a list of datasets that have to be read from disk and then processed by a specific processor.
290
291
292
293
294

        :param procLvl:
        :param prevLvl_objects:
        :return:
        """
Daniel Scheffler's avatar
Daniel Scheffler committed
295
296
        def is_procL_lower(dataset):
            return HLP_F.is_proc_level_lower(dataset['proc_level'], target_lvl=procLvl)
297
298

        if prevLvl_objects is None:
Daniel Scheffler's avatar
Daniel Scheffler committed
299
            return [dataset for dataset in self.config.data_list if is_procL_lower(dataset)]  # TODO generator?
300
        else:
Daniel Scheffler's avatar
Daniel Scheffler committed
301
            return [dataset for dataset in self.config.data_list if is_procL_lower(dataset) and
302
                    not self._is_inMEM(prevLvl_objects + self.failed_objects, dataset)]
303
304
305
306
307
308
309
310
311
312
313
314
315

    def get_DB_objects(self, procLvl, prevLvl_objects=None, parallLev=None, blocksize=None):
        """
        Returns a list of GMS objects for datasets available on disk that have to be processed by the current processor.

        :param procLvl:         <str> processing level oof the current processor
        :param prevLvl_objects: <list> of in-mem GMS objects produced by the previous processor
        :param parallLev:       <str> parallelization level ('scenes' or 'tiles')
                                -> defines if full cubes or blocks are to be returned
        :param blocksize:       <tuple> block size in case blocks are to be returned, e.g. (2000,2000)
        :return:
        """
        # TODO get prevLvl_objects automatically from self
316
        if procLvl == 'L1A':
317
318
319
            return []
        else:
            # handle input parameters
Daniel Scheffler's avatar
Daniel Scheffler committed
320
321
            parallLev = parallLev or self.config.parallelization_level
            blocksize = blocksize or self.config.tiling_block_size_XY
322
            prevLvl = proc_chain[proc_chain.index(procLvl) - 1]  # TODO replace by enum
323
324

            # get GMSfile list
325
            dataset_dicts = self._get_processor_data_list(procLvl, prevLvl_objects)
326
327
328
329
330
            GMSfile_list_prevLvl_inDB = INP_R.get_list_GMSfiles(dataset_dicts, prevLvl)

            # create GMS objects from disk with respect to parallelization level and block size
            if parallLev == 'scenes':
                # get input parameters for creating GMS objects as full cubes
331
                work = [[GMS, ['cube', None]] for GMS in GMSfile_list_prevLvl_inDB]
332
333
            else:
                # define tile positions and size
334
                def get_tilepos_list(GMSfile):
335
336
                    return get_array_tilebounds(array_shape=INP_R.GMSfile2dict(GMSfile)['shape_fullArr'],
                                                tile_shape=blocksize)
337
338
339

                # get input parameters for creating GMS objects as blocks
                work = [[GMSfile, ['block', tp]] for GMSfile in GMSfile_list_prevLvl_inDB
340
                        for tp in get_tilepos_list(GMSfile)]
341

342
343
344
345
            # create GMS objects for the found files on disk
            # NOTE: DON'T multiprocess that with MAP(GMS_object(*initargs).from_disk, work)
            # in case of multiple subsystems GMS_object(*initargs) would always point to the same object in memory
            # -> subsystem attribute will be overwritten each time
346
            DB_objs = [HLP_F.parentObjDict[prevLvl].from_disk(tuple_GMS_subset=w) for w in work]
347

348
349
350
351
352
            if DB_objs:
                DB_objs = list(chain.from_iterable(DB_objs)) if list in [type(i) for i in DB_objs] else list(DB_objs)

            return DB_objs

Daniel Scheffler's avatar
Daniel Scheffler committed
353
    def run_all_processors_OLD(self, custom_data_list=None):
354
355
356
        """
        Run all processors at once.
        """
357

358
        signal.signal(signal.SIGINT, self.stop)  # enable clean shutdown possibility
359

360
        # noinspection PyBroadException
361
        try:
362
            if self.config.profiling:
363
364
365
366
                from pyinstrument import Profiler
                self.profiler = Profiler()  # or Profiler(use_signal=False), see below
                self.profiler.start()

367
            self.logger.info('Execution of entire GeoMultiSens pre-processing chain started for job ID %s...'
368
                             % self.config.ID)
369
            self.DB_job_record.reset_job_progress()  # updates attributes of DB_job_record and related DB entry
370
            self.config.status = 'running'
371
            self.update_DB_job_record()  # TODO implement that into job.status.setter
372
373
374
375
376

            self.failed_objects = []

            # get list of datasets to be processed
            if custom_data_list:
377
                self.config.data_list = custom_data_list
378
379

            # add local availability
Daniel Scheffler's avatar
Daniel Scheffler committed
380
            self.config.data_list = self.add_local_availability(self.config.data_list)
381
            self.update_DB_job_statistics(self.config.data_list)
382
383
384
385
386
387
388
389
390
391
392
393

            self.L1A_processing()
            self.L1B_processing()
            self.L1C_processing()
            self.L2A_processing()
            self.L2B_processing()
            self.L2C_processing()

            # create summary
            self.create_job_summary()

            self.logger.info('Execution finished.')
394
            # TODO implement failed_with_warnings:
395
396
397
398
            self.config.status = 'finished' if not self.failed_objects else 'finished_with_errors'
            self.config.end_time = datetime.datetime.now()
            self.config.computation_time = self.config.end_time - self.config.start_time
            self.logger.info('Time for execution: %s' % self.config.computation_time)
399
400
401
402

            # update database entry of current job
            self.update_DB_job_record()

403
            if self.config.profiling:
404
405
406
                self.profiler.stop()
                print(self.profiler.output_text(unicode=True, color=True))

407
            self.shutdown()
408

409
        except Exception:  # noqa E722  # bare except
410
            if self.config.profiling:
411
412
413
                self.profiler.stop()
                print(self.profiler.output_text(unicode=True, color=True))

414
            self.config.status = 'failed'
Daniel Scheffler's avatar
Daniel Scheffler committed
415
416
417
418
            self.update_DB_job_record()

            if not self.config.disable_exception_handler:
                self.logger.error('Execution failed with an error:', exc_info=True)
419
                self.shutdown()
Daniel Scheffler's avatar
Daniel Scheffler committed
420
421
            else:
                self.logger.error('Execution failed with an error:')
422
                self.shutdown()
Daniel Scheffler's avatar
Daniel Scheffler committed
423
424
425
426
427
428
429
430
431
432
433
434
435
436
                raise

    def run_all_processors(self, custom_data_list=None):
        signal.signal(signal.SIGINT, self.stop)  # enable clean shutdown possibility

        # noinspection PyBroadException
        try:
            if self.config.profiling:
                from pyinstrument import Profiler
                self.profiler = Profiler()  # or Profiler(use_signal=False), see below
                self.profiler.start()

            self.logger.info('Execution of entire GeoMultiSens pre-processing chain started for job ID %s...'
                             % self.config.ID)
437

Daniel Scheffler's avatar
Daniel Scheffler committed
438
439
            self.DB_job_record.reset_job_progress()  # updates attributes of DB_job_record and related DB entry
            self.config.status = 'running'
440
            GMS_object.proc_status_all_GMSobjs.clear()  # reset
Daniel Scheffler's avatar
Daniel Scheffler committed
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
            self.update_DB_job_record()  # TODO implement that into config.status.setter

            self.failed_objects = []

            # get list of datasets to be processed
            if custom_data_list:
                self.config.data_list = custom_data_list

            # add local availability
            self.config.data_list = self.add_local_availability(self.config.data_list)
            self.update_DB_job_statistics(self.config.data_list)

            # group dataset dicts by sceneid
            dataset_groups = HLP_F.group_dicts_by_key(self.config.data_list, key='scene_ID')

456
            # RUN PREPROCESSING
Daniel Scheffler's avatar
Daniel Scheffler committed
457
            from .pipeline import run_complete_preprocessing
458
            GMS_objs = imap_unordered(run_complete_preprocessing, dataset_groups, flatten_output=True)
Daniel Scheffler's avatar
Daniel Scheffler committed
459
460

            # separate results into successful and failed objects
461
462
463
464
465
466
467
468
469
            def assign_attr(tgt_procL):
                return [obj for obj in GMS_objs if isinstance(obj, GMS_object) and obj.proc_level == tgt_procL]

            self.L1A_newObjects = assign_attr('L1A')
            self.L1B_newObjects = assign_attr('L1B')
            self.L1C_newObjects = assign_attr('L1C')
            self.L2A_newObjects = assign_attr('L2A')
            self.L2B_newObjects = assign_attr('L2B')
            self.L2C_newObjects = assign_attr('L2C')
Daniel Scheffler's avatar
Daniel Scheffler committed
470
471
472
473
474
475
            self.failed_objects = [obj for obj in GMS_objs if isinstance(obj, failed_GMS_object)]

            # create summary
            self.create_job_summary()

            self.logger.info('Execution finished.')
476
            # TODO implement failed_with_warnings
477
            self.config.status = 'finished' if not self.failed_objects else 'finished_with_errors'
Daniel Scheffler's avatar
Daniel Scheffler committed
478
479
480
481
482
483
484
485
486
487
488
            self.config.end_time = datetime.datetime.now()
            self.config.computation_time = self.config.end_time - self.config.start_time
            self.logger.info('Time for execution: %s' % self.config.computation_time)

            # update database entry of current job
            self.update_DB_job_record()

            if self.config.profiling:
                self.profiler.stop()
                print(self.profiler.output_text(unicode=True, color=True))

489
            self.shutdown()
Daniel Scheffler's avatar
Daniel Scheffler committed
490
491
492
493
494
495
496

        except Exception:  # noqa E722  # bare except
            if self.config.profiling:
                self.profiler.stop()
                print(self.profiler.output_text(unicode=True, color=True))

            self.config.status = 'failed'
497
498
            self.update_DB_job_record()

499
            if not self.config.disable_exception_handler:
500
                self.logger.error('Execution failed with an error:', exc_info=True)
501
                self.shutdown()
502
            else:
503
                self.logger.error('Execution failed with an error:')
504
                self.shutdown()
505
                raise
506

507
508
    def stop(self, signum, frame):
        """Interrupt the running process controller gracefully."""
509

510
        self.config.status = 'canceled'
511
512
        self.update_DB_job_record()

513
        self.shutdown()
514
        self.logger.warning('Process controller stopped by user.')
515
516
517
518
519
520

        raise KeyboardInterrupt  # terminate execution and show traceback

    def shutdown(self):
        """Shutdown the process controller instance (loggers, remove temporary directories, ...)."""

521
522
523
        # release unclosed locks
        release_unclosed_locks(self.logger)

524
        # clear any temporary files
525
        tempdir = os.path.join(self.config.path_tempdir)
526
        self.logger.info('Deleting temporary directory %s.' % tempdir)
527
528
        if os.path.exists(tempdir):
            shutil.rmtree(tempdir)
529

530
531
532
        del self.logger
        shutdown_loggers()

533
534
535
536
    def benchmark(self):
        """
        Run a benchmark.
        """
537
        data_list_bench = self.config.data_list
538
539
540
541
        for count_datasets in range(len(data_list_bench)):
            t_processing_all_runs, t_IO_all_runs = [], []
            for count_run in range(10):
                current_data_list = data_list_bench[0:count_datasets + 1]
542
543
                if os.path.exists(self.config.path_database):
                    os.remove(self.config.path_database)
544
545
546
547
548
549
550
551
552
553
                t_start = time.time()
                self.run_all_processors(current_data_list)
                t_processing_all_runs.append(time.time() - t_start)
                t_IO_all_runs.append(globals()['time_IO'])

            assert current_data_list, 'Empty data list.'
            OUT_W.write_global_benchmark_output(t_processing_all_runs, t_IO_all_runs, current_data_list)

    def L1A_processing(self):
        """
554
        Run Level 1A processing: Data import and metadata homogenization
555
        """
556
        if self.config.exec_L1AP[0]:
557
            self.logger.info('\n\n##### Level 1A Processing started - raster format and metadata homogenization ####\n')
558

559
560
            datalist_L1A_P = self._get_processor_data_list('L1A')

561
            if self.config.parallelization_level == 'scenes':
562
                # map
563
                L1A_resObjects = MAP(L1A_map, datalist_L1A_P, CPUs=12)
564
            else:  # tiles
565
566
                all_L1A_tiles_map1 = MAP(L1A_map_1, datalist_L1A_P,
                                         flatten_output=True)  # map_1 # merge results to new list of splits
567

568
569
570
                L1A_obj_tiles = MAP(L1A_map_2, all_L1A_tiles_map1)  # map_2
                grouped_L1A_Tiles = HLP_F.group_objects_by_attributes(
                    L1A_obj_tiles, 'scene_ID', 'subsystem')  # group results
571

572
                L1A_objects = MAP(L1A_P.L1A_object.from_tiles, grouped_L1A_Tiles)  # reduce
573

574
                L1A_resObjects = MAP(L1A_map_3, L1A_objects)  # map_3
575

576
            self.L1A_newObjects = [obj for obj in L1A_resObjects if isinstance(obj, L1A_P.L1A_object)]
577
            self.failed_objects += [obj for obj in L1A_resObjects if isinstance(obj, failed_GMS_object) and
578
579
580
581
582
583
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L1A_newObjects

    def L1B_processing(self):
        """
584
        Run Level 1B processing: calculation of geometric shifts
585
586
587
588
        """
        # TODO implement check for running spatial index mediator server
        # run on full cubes

589
        if self.config.exec_L1BP[0]:
590
            self.logger.info('\n\n####### Level 1B Processing started - detection of geometric displacements #######\n')
591

592
593
            L1A_DBObjects = self.get_DB_objects('L1B', self.L1A_newObjects, parallLev='scenes')
            L1A_Instances = self.L1A_newObjects + L1A_DBObjects  # combine newly and earlier processed L1A data
594

595
            L1B_resObjects = MAP(L1B_map, L1A_Instances)
596

597
598
599
            self.L1B_newObjects = [obj for obj in L1B_resObjects if isinstance(obj, L1B_P.L1B_object)]
            self.failed_objects += [obj for obj in L1B_resObjects if isinstance(obj, failed_GMS_object) and
                                    obj.scene_ID not in self.sceneids_failed]
600
601
602
603
604

        return self.L1B_newObjects

    def L1C_processing(self):
        """
605
        Run Level 1C processing: atmospheric correction
606
        """
607
        if self.config.exec_L1CP[0]:
608
            self.logger.info('\n\n############## Level 1C Processing started - atmospheric correction ##############\n')
609

610
            if self.config.parallelization_level == 'scenes':
611
612
613
614
                L1B_DBObjects = self.get_DB_objects('L1C', self.L1B_newObjects)
                L1B_Instances = self.L1B_newObjects + L1B_DBObjects  # combine newly and earlier processed L1B data

                # group by scene ID (all subsystems belonging to the same scene ID must be processed together)
615
                grouped_L1B_Instances = HLP_F.group_objects_by_attributes(L1B_Instances, 'scene_ID')
616

617
618
                L1C_resObjects = MAP(L1C_map, grouped_L1B_Instances, flatten_output=True,
                                     CPUs=15)  # FIXME CPUs set to 15 for testing
619

620
            else:  # tiles
621
622
                raise NotImplementedError("Tiled processing is not yet completely implemented for L1C processor. Use "
                                          "parallelization level 'scenes' instead!")
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
                # blocksize = (5000, 5000)
                # """if newly processed L1A objects are present: cut them into tiles"""
                # L1B_newTiles = []
                # if self.L1B_newObjects:
                #     tuples_obj_blocksize = [(obj, blocksize) for obj in self.L1B_newObjects]
                #     L1B_newTiles = MAP(HLP_F.cut_GMS_obj_into_blocks, tuples_obj_blocksize, flatten_output=True)
                #
                # """combine newly and earlier processed L1B data"""
                # L1B_newDBTiles = self.get_DB_objects('L1C', self.L1B_newObjects, blocksize=blocksize)
                # L1B_tiles = L1B_newTiles + L1B_newDBTiles
                #
                # # TODO merge subsets of S2/Aster in order to provide all bands for atm.correction
                # L1C_tiles = MAP(L1C_map, L1B_tiles)
                # grouped_L1C_Tiles = \
                #     HLP_F.group_objects_by_attributes(L1C_tiles, 'scene_ID', 'subsystem')  # group results
                # [L1C_tiles_group[0].delete_tempFiles() for L1C_tiles_group in grouped_L1C_Tiles]
                # L1C_resObjects = MAP(L1C_P.L1C_object().from_tiles, grouped_L1C_Tiles)  # reduce

            self.L1C_newObjects = [obj for obj in L1C_resObjects if isinstance(obj, L1C_P.L1C_object)]
642
            self.failed_objects += [obj for obj in L1C_resObjects if isinstance(obj, failed_GMS_object) and
643
644
645
646
647
648
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L1C_newObjects

    def L2A_processing(self):
        """
649
        Run Level 2A processing: geometric homogenization
650
        """
651
        if self.config.exec_L2AP[0]:
652
653
            self.logger.info(
                '\n\n#### Level 2A Processing started - shift correction / geometric homogenization ####\n')
654

655
            """combine newly and earlier processed L1C data"""
656
657
            L1C_DBObjects = self.get_DB_objects('L2A', self.L1C_newObjects, parallLev='scenes')
            L1C_Instances = self.L1C_newObjects + L1C_DBObjects  # combine newly and earlier processed L1C data
658
659
660
661

            # group by scene ID (all subsystems belonging to the same scene ID must be processed together)
            grouped_L1C_Instances = HLP_F.group_objects_by_attributes(L1C_Instances, 'scene_ID')

662
            L2A_resTiles = MAP(L2A_map, grouped_L1C_Instances, flatten_output=True)
663

664
            self.L2A_tiles = [obj for obj in L2A_resTiles if isinstance(obj, L2A_P.L2A_object)]
665
            self.failed_objects += [obj for obj in L2A_resTiles if isinstance(obj, failed_GMS_object) and
666
667
668
669
670
671
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L2A_tiles

    def L2B_processing(self):
        """
672
        Run Level 2B processing: spectral homogenization
673
        """
674
        if self.config.exec_L2BP[0]:
675
            self.logger.info('\n\n############# Level 2B Processing started - spectral homogenization ##############\n')
676

677
            if self.config.parallelization_level == 'scenes':
678
                # don't know if scenes makes sense in L2B processing because full objects are very big!
679
                """if newly processed L2A objects are present: merge them to scenes"""
680
681
                grouped_L2A_Tiles = HLP_F.group_objects_by_attributes(self.L2A_tiles, 'scene_ID')  # group results
                # reduce # will be too slow because it has to pickle back really large L2A_newObjects
682
                # L2A_newObjects  = MAP(HLP_F.merge_GMS_tiles_to_GMS_obj, grouped_L2A_Tiles)
683
                L2A_newObjects = [L2A_P.L2A_object.from_tiles(tileList) for tileList in grouped_L2A_Tiles]
684

685
                """combine newly and earlier processed L2A data"""
686
687
                L2A_DBObjects = self.get_DB_objects('L2B', self.L2A_tiles)
                L2A_Instances = L2A_newObjects + L2A_DBObjects  # combine newly and earlier processed L2A data
688

689
                L2B_resObjects = MAP(L2B_map, L2A_Instances)
690
691

            else:  # tiles
692
                L2A_newTiles = self.L2A_tiles  # tiles have the block size specified in L2A_map_2
693
694

                """combine newly and earlier processed L2A data"""
695
696
697
                blocksize = (2048, 2048)  # must be equal to the blocksize of L2A_newTiles specified in L2A_map_2
                L2A_newDBTiles = self.get_DB_objects('L2B', self.L2A_tiles, blocksize=blocksize)
                L2A_tiles = L2A_newTiles + L2A_newDBTiles
698

699
                L2B_tiles = MAP(L2B_map, L2A_tiles)
700

701
702
                # group results # FIXME nötig an dieser Stelle?
                grouped_L2B_Tiles = HLP_F.group_objects_by_attributes(L2B_tiles, 'scene_ID')
703
704
                [L2B_tiles_group[0].delete_tempFiles() for L2B_tiles_group in grouped_L2B_Tiles]

705
                L2B_resObjects = [L2B_P.L2B_object.from_tiles(tileList) for tileList in grouped_L2B_Tiles]
706

707
            self.L2B_newObjects = [obj for obj in L2B_resObjects if isinstance(obj, L2B_P.L2B_object)]
708
            self.failed_objects += [obj for obj in L2B_resObjects if isinstance(obj, failed_GMS_object) and
709
710
711
712
713
714
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L2B_newObjects

    def L2C_processing(self):
        """
715
        Run Level 2C processing: accurracy assessment and MGRS tiling
716
        """
717
        # FIXME only parallelization_level == 'scenes' implemented
718
        if self.config.exec_L2CP[0]:
719
            self.logger.info('\n\n########## Level 2C Processing started - calculation of quality layers ###########\n')
720

721
            """combine newly and earlier processed L2A data"""
722
723
            L2B_DBObjects = self.get_DB_objects('L2C', self.L2B_newObjects, parallLev='scenes')
            L2B_Instances = self.L2B_newObjects + L2B_DBObjects  # combine newly and earlier processed L2A data
724

725
            L2C_resObjects = MAP(L2C_map, L2B_Instances, CPUs=8)  # FIXME 8 workers due to heavy IO
726
            # FIXME in case of inmem_serialization mode results are too big to be back-pickled
727
            self.L2C_newObjects = [obj for obj in L2C_resObjects if isinstance(obj, L2C_P.L2C_object)]
728
            self.failed_objects += [obj for obj in L2C_resObjects if isinstance(obj, failed_GMS_object) and
729
730
731
732
733
734
735
736
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L2C_newObjects

    def update_DB_job_record(self):
        """
        Update the database records of the current job (table 'jobs').
        """
737
        # TODO move this method to config.Job
738
739
        # update 'failed_sceneids' column of job record within jobs table
        sceneids_failed = list(set([obj.scene_ID for obj in self.failed_objects]))
740
        DB_T.update_records_in_postgreSQLdb(
741
            self.config.conn_database, 'jobs',
742
            {'failed_sceneids': sceneids_failed,  # update 'failed_sceneids' column
743
744
             'finishtime': self.config.end_time,  # add job finish timestamp
             'status': self.config.status},  # update 'job_status' column
745
            {'id': self.config.ID}, timeout=30000)
746

747
748
749
750
751
    def update_DB_job_statistics(self, usecase_datalist):
        """
        Update job statistics of the running job in the database.
        """
        # TODO move this method to config.Job
752
        already_updated_IDs = []
753
        for ds in usecase_datalist:
754
755
            if ds['proc_level'] is not None and ds['scene_ID'] not in already_updated_IDs:
                # update statistics column of jobs table
756
                DB_T.increment_decrement_arrayCol_in_postgreSQLdb(
757
                    self.config.conn_database, 'jobs', 'statistics', cond_dict={'id': self.config.ID},
758
                    idx_val2decrement=db_jobs_statistics_def['pending'],
759
760
                    idx_val2increment=db_jobs_statistics_def[ds['proc_level']],
                    timeout=30000)
761

762
763
764
                # avoid double updating in case of subsystems belonging to the same scene ID
                already_updated_IDs.append(ds['scene_ID'])

765
766
767
768
    def create_job_summary(self):
        """
        Create job success summary
        """
769
770
771

        # get objects with highest requested processing level
        highest_procL_Objs = []
772
        for pL in reversed(proc_chain):
773
            if getattr(self.config, 'exec_%sP' % pL)[0]:
774
775
                highest_procL_Objs = \
                    getattr(self, '%s_newObjects' % pL) if pL != 'L2A' else (self.L2A_tiles or self.L2A_newObjects)
776
777
                break

778
779
780
781
        gms_objects2summarize = highest_procL_Objs + self.failed_objects
        if gms_objects2summarize:
            # create summaries
            detailed_JS, quick_JS = get_job_summary(gms_objects2summarize)
782
783
784
            detailed_JS.to_excel(os.path.join(self.config.path_job_logs, '%s_summary.xlsx' % self.config.ID))
            detailed_JS.to_csv(os.path.join(self.config.path_job_logs, '%s_summary.csv' % self.config.ID), sep='\t')
            self.logger.info('\nQUICK JOB SUMMARY (ID %s):\n' % self.config.ID + quick_JS.to_string())
785
786

            self.summary_detailed = detailed_JS
787
            self.summary_quick = quick_JS
788
789
790
791
792

        else:
            # TODO implement check if proc level with lowest procL has to be processed at all (due to job.exec_L1X)
            # TODO otherwise it is possible that get_job_summary receives an empty list
            self.logger.warning("Job summary skipped because get_job_summary() received an empty list of GMS objects.")
793
794
795
796
797
798

    def clear_lists_procObj(self):
        self.failed_objects = []
        self.L1A_newObjects = []
        self.L1B_newObjects = []
        self.L1C_newObjects = []
799
        self.L2A_tiles = []
800
        self.L2B_newObjects = []
801
802
803
804
805
        self.L2C_newObjects = []


def get_job_summary(list_GMS_objects):
    # get detailed job summary
806
807
    DJS_cols = ['GMS_object', 'scene_ID', 'entity_ID', 'satellite', 'sensor', 'subsystem', 'image_type', 'proc_level',
                'arr_shape', 'arr_pos', 'failedMapper', 'ExceptionType', 'ExceptionValue', 'ExceptionTraceback']
808
809
810
811
    DJS = DataFrame(columns=DJS_cols)
    DJS['GMS_object'] = list_GMS_objects

    for col in DJS_cols[1:]:
812
813
        def get_val(obj): return getattr(obj, col) if hasattr(obj, col) else None
        DJS[col] = list(DJS['GMS_object'].map(get_val))
814
815

    del DJS['GMS_object']
816
    DJS = DJS.sort_values(by=['satellite', 'sensor', 'entity_ID'])
817
818

    # get quick job summary
819
820
821
822
    QJS = DataFrame(columns=['satellite', 'sensor', 'count', 'proc_successfully', 'proc_failed'])
    all_sat, all_sen = zip(*[i.split('__') for i in (np.unique(DJS['satellite'] + '__' + DJS['sensor']))])
    QJS['satellite'] = all_sat
    QJS['sensor'] = all_sen
823
    # count objects with the same satellite/sensor/sceneid combination
824
825
    QJS['count'] = [len(DJS[(DJS['satellite'] == sat) & (DJS['sensor'] == sen)]['scene_ID'].unique())
                    for sat, sen in zip(all_sat, all_sen)]
826
    QJS['proc_successfully'] = [len(DJS[(DJS['satellite'] == sat) &
827
828
                                        (DJS['sensor'] == sen) &
                                        (DJS['failedMapper'].isnull())]['scene_ID'].unique())
829
                                for sat, sen in zip(all_sat, all_sen)]
830
    QJS['proc_failed'] = QJS['count'] - QJS['proc_successfully']
831
832
    QJS = QJS.sort_values(by=['satellite', 'sensor'])
    return DJS, QJS