process_controller.py 34.9 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
Daniel Scheffler's avatar
Daniel Scheffler committed
13
from typing import TYPE_CHECKING
14

15
16
from ..io import output_writer as OUT_W
from ..io import input_reader as INP_R
17
18
19
20
21
22
from ..misc import database_tools as DB_T
from ..misc import helper_functions as HLP_F
from ..misc import environment as ENV
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
25
from ..model.gms_object import failed_GMS_object
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
28
from ..config import set_config, GMS_config
from .multiproc import MAP
29
from ..misc.definition_dicts import proc_chain, db_jobs_statistics_def
30

31
32
from py_tools_ds.numeric.array import get_array_tilebounds

33
if TYPE_CHECKING:
Daniel Scheffler's avatar
Daniel Scheffler committed
34
    from collections import OrderedDict  # noqa F401  # flake8 issue
35
36
37
38


__author__ = 'Daniel Scheffler'

39
40

class process_controller(object):
41
    def __init__(self, job_ID, call_type='webapp', exec_mode='Python', db_host='localhost',
42
                 parallelization_level='scenes', delete_old_output=False, job_config_kwargs=None):
43
44
        # type: (int, str, str, str, str, bool) -> None

45
        """gms_preprocessing process controller
46
47
48
49
50

        :param job_ID:                  <int> a job ID belonging to a valid database record within table 'jobs'
        :param call_type:               <str> choices: 'webapp' and 'console'
        :param exec_mode:               <str> choices: 'Python' - writes all intermediate data to disk
                                                       'Flink'  - keeps all intermediate data in memory
51
        :param db_host:                 <str> hostname of the host where database is hosted
52
53
        :param parallelization_level:   <str> choices: 'scenes' - parallelization on scene-level
                                                       'tiles'  - parallelisation on tile-level
54
55
        :param delete_old_output:       <bool> whether to delete previously created output of the given job ID
                                        before running the job (default = False)
56
57
58
        """

        # assertions
59
60
61
62
63
64
        if not isinstance(job_ID, int):
            raise ValueError("'job_ID' must be an integer value. Got %s." % type(job_ID))
        if call_type not in ['webapp', 'console']:
            raise ValueError("Unexpected call_type '%s'!" % call_type)
        if exec_mode not in ['Python', 'Flink']:
            raise ValueError("Unexpected exec_mode '%s'!" % exec_mode)
65
        if parallelization_level not in ['scenes', 'tiles']:
66
            raise ValueError("Unexpected parallelization_level '%s'!" % parallelization_level)
67

68
69
70
        self.call_type = call_type
        self.parallLev = parallelization_level
        self._logger = None
71
        self._DB_job_record = None
72
        self.profiler = None
73
74
75
76
77

        self.failed_objects = []
        self.L1A_newObjects = []
        self.L1B_newObjects = []
        self.L1C_newObjects = []
78
        self.L2A_tiles = []
79
80
81
82
        self.L2B_newObjects = []
        self.L2C_newObjects = []

        self.summary_detailed = None
83
        self.summary_quick = None
84
85

        # set GMS configuration
86
87
        set_config(call_type=call_type, exec_mode=exec_mode, job_ID=job_ID, db_host=db_host, reset=True,
                   job_kwargs=job_config_kwargs)
88
        self.job = GMS_config.job
89
90
91
        self.usecase = GMS_config.usecase

        # check environment
92
93
        self.GMSEnv = ENV.GMSEnvironment(self.logger)
        self.GMSEnv.check_dependencies()
94
        self.GMSEnv.check_read_write_permissions()
95

96
97
        # 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
98
        # if isdebugging:  # override the existing settings in order to get write access everywhere
99
100
        #    pass

101
        # called_from_iPyNb = 1 if 'ipykernel/__main__.py' in sys.argv[0] else 0
102

103
104
105
        self.logger.info('Process Controller initialized for job ID %s (comment: %s).'
                         % (self.job.ID, self.DB_job_record.comment))

106
107
108
        if delete_old_output:
            self.logger.info('Deleting previously processed data...')
            self.DB_job_record.delete_procdata_of_entire_job(force=True)
109

110
111
112
113
114
    @property
    def logger(self):
        if self._logger and self._logger.handlers[:]:
            return self._logger
        else:
115
116
            self._logger = GMS_logger('log__%s' % self.job.ID,
                                      path_logfile=os.path.join(self.job.path_job_logs, '%s.log' % self.job.ID),
117
                                      log_level=self.job.log_level, append=False)
118
119
120
121
122
123
124
125
126
127
128
129
130
            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()
            self.logger = None

    @property
131
132
133
134
135
136
137
    def DB_job_record(self):
        if self._DB_job_record:
            return self._DB_job_record
        else:
            self._DB_job_record = DB_T.GMS_JOB(self.job.conn_database)
            self._DB_job_record.from_job_ID(self.job.ID)
            return self._DB_job_record
138

139
140
141
    @DB_job_record.setter
    def DB_job_record(self, value):
        self._DB_job_record = value
142

143
144
145
    @property
    def sceneids_failed(self):
        return [obj.scene_ID for obj in self.failed_objects]
146

147
148
149
    @staticmethod
    def add_local_availability(dataset):
        # TODO revise this function
150
151
152
        # TODO this does not respect that all subsystems of the same scene ID must be available at the same proc level!

        # query the database and get the last written processing level and LayerBandsAssignment
153
        if GMS_config.job.call_type == 'webapp':
154
155
156
            DB_match = DB_T.get_info_from_postgreSQLdb(
                GMS_config.job.conn_database, 'scenes_proc', ['proc_level', 'layer_bands_assignment'],
                dict(sceneid=dataset['scene_ID']))
157

158
        else:  # call_type == 'console'
159
160
            DB_match = DB_T.get_info_from_SQLdb(
                GMS_config.job.path_database, 'processed_data', ['proc_level', 'LayerBandsAssignment'],
161
162
163
164
165
166
                dict(image_type=dataset['image_type'],
                     satellite=dataset['satellite'],
                     sensor=dataset['sensor'],
                     subsystem=dataset['subsystem'],
                     sensormode=dataset['sensormode'],
                     entity_ID=dataset['entity_ID']))
167
168

        # get the corresponding logfile
169
170
        path_logfile = path_generator(
            dataset).get_path_logfile()  # FIXME this always returns the logfile for the subsystem.
171

172
173
        # FIXME -> merged logfiles (L2A+) are ignored
        # FIXME -> for subsystems the highest start procL is L2A
174

175
        def get_AllWrittenProcL_dueLog(path_log):  # TODO replace this by database query + os.path.exists
176
177
            """Returns all processing level that have been successfully written according to logfile."""

178
            if not os.path.exists(path_log):
179
                print("No logfile named '%s' found for %s at %s. Dataset has to be reprocessed."
180
181
182
183
184
185
                      % (os.path.basename(path_log), dataset['entity_ID'], os.path.dirname(path_log)))
                AllWrittenProcL_dueLog = []
            else:
                logfile = open(path_log, 'r').read()
                AllWrittenProcL_dueLog = re.findall(":*(\S*\s*) data successfully saved.", logfile, re.I)
                if not AllWrittenProcL_dueLog:  # AllWrittenProcL_dueLog = []
186
                    print('%s: According to logfile no completely processed data exist at any processing level. '
187
188
189
190
191
                          'Dataset has to be reprocessed.' % dataset['entity_ID'])
                else:
                    AllWrittenProcL_dueLog = HLP_F.sorted_nicely(list(set(AllWrittenProcL_dueLog)))
            return AllWrittenProcL_dueLog

192
        # check if there are not multiple database records for this dataset
193
        if len(DB_match) == 1 or DB_match == [] or DB_match == 'database connection fault':
194
195

            # get all processing level that have been successfully written
196
197
            AllWrittenProcL = get_AllWrittenProcL_dueLog(path_logfile)
            dataset['proc_level'] = None  # default (dataset has to be reprocessed)
198
199

            # loop through all the found proc. levels and find the one that fulfills all requirements
200
            for ProcL in reversed(AllWrittenProcL):
201
202
                if dataset['proc_level']:
                    break  # proc_level found; no further searching for lower proc_levels
203
                assumed_path_GMS_file = '%s_%s.gms' % (os.path.splitext(path_logfile)[0], ProcL)
204
205

                # check if there is also a corresponding GMS_file on disk
206
207
                if os.path.isfile(assumed_path_GMS_file):
                    GMS_file_dict = INP_R.GMSfile2dict(assumed_path_GMS_file)
208
                    target_LayerBandsAssignment = \
209
                        get_LayerBandsAssignment(dict(
210
211
212
213
214
                            image_type=dataset['image_type'],
                            Satellite=dataset['satellite'],
                            Sensor=dataset['sensor'],
                            Subsystem=dataset['subsystem'],
                            proc_level=ProcL,  # must be respected because LBA changes after atm. Corr.
215
                            dataset_ID=dataset['dataset_ID'],
216
217
218
219
                            logger=None), nBands=(1 if dataset['sensormode'] == 'P' else None))

                    # check if the LayerBandsAssignment of the written dataset on disk equals the
                    # desired LayerBandsAssignment
220
                    if target_LayerBandsAssignment == GMS_file_dict['LayerBandsAssignment']:
221
222

                        # update the database record if the dataset could not be found in database
223
                        if DB_match == [] or DB_match == 'database connection fault':
224
225
                            print('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.'
226
227
                                  % (dataset['entity_ID'], ProcL))
                            DB_T.data_DB_updater(GMS_file_dict)
228

229
230
231
                            if GMS_config.job.call_type == 'console':
                                DB_T.SQL_DB_to_csv()
                            dataset['proc_level'] = ProcL
232
233

                        # if the dataset could be found in database
234
235
                        elif len(DB_match) == 1:
                            try:
236
                                print('Found a matching %s dataset for %s. Processing skipped until %s.'
237
                                      % (ProcL, dataset['entity_ID'], proc_chain[proc_chain.index(ProcL) + 1]))
238
                            except IndexError:
239
                                print('Found a matching %s dataset for %s. Processing already done.'
240
                                      % (ProcL, dataset['entity_ID']))
241

242
243
244
245
                            if DB_match[0][0] == ProcL:
                                dataset['proc_level'] = DB_match[0][0]
                            else:
                                dataset['proc_level'] = ProcL
246

247
                    else:
248
                        print('Found a matching dataset for %s but with a different LayerBandsAssignment. '
249
250
251
252
253
                              'Dataset has to be reprocessed.' % dataset['entity_ID'])
                else:
                    print('%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 '')
254

255
        elif len(DB_match) > 1:
256
            print('According to database there are multiple matches for the dataset %s. Dataset has to be reprocessed.'
257
258
                  % dataset['entity_ID'])
            dataset['proc_level'] = None
259

260
261
262
263
264
        else:
            dataset['proc_level'] = None

        return dataset

265
266
    @staticmethod
    def _is_inMEM(GMS_objects, dataset):
Daniel Scheffler's avatar
Daniel Scheffler committed
267
        # type: (list, OrderedDict) -> bool
268
269
270
271
        """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()
        """
272
        # check if the scene ID of the given dataset is in the scene IDs of the previously processed datasets
273
274
275
276
        return dataset['scene_ID'] in [obj.scene_ID for obj in GMS_objects]

    @staticmethod
    def _is_already_present(dataset, procLvl):
277
278
279
280
281
282
        """Checks if the given dataset is already available on disk.

        :param dataset:     <GMS object>
        :param procLvl:     <str> processing level to be checked
        :return:            <bool>
        """
283
284
285
        return HLP_F.proc_level_already_present(dataset['proc_level'], procLvl)

    def _get_processor_data_list(self, procLvl, prevLvl_objects=None):
286
        """Returns a list of datasets that have to be read from disk and then processed by a specific processor.
287
288
289
290
291

        :param procLvl:
        :param prevLvl_objects:
        :return:
        """
292
293
        def is_already_present(dataset):
            return HLP_F.proc_level_already_present(dataset['proc_level'], target_lvl=procLvl)
294
295

        if prevLvl_objects is None:
296
            return [dataset for dataset in self.usecase.data_list if not is_already_present(dataset)]  # TODO generator?
297
        else:
298
299
            return [dataset for dataset in self.usecase.data_list if not is_already_present(dataset) and
                    not self._is_inMEM(prevLvl_objects + self.failed_objects, dataset)]
300
301
302
303
304
305
306
307
308
309
310
311
312

    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
313
        if procLvl == 'L1A':
314
315
316
317
318
            return []
        else:
            # handle input parameters
            parallLev = parallLev if parallLev else self.parallLev
            blocksize = blocksize if blocksize else self.job.tiling_block_size_XY
319
            prevLvl = proc_chain[proc_chain.index(procLvl) - 1]  # TODO replace by enum
320
321

            # get GMSfile list
322
            dataset_dicts = self._get_processor_data_list(procLvl, prevLvl_objects)
323
324
325
326
327
            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
328
                work = [[GMS, ['cube', None]] for GMS in GMSfile_list_prevLvl_inDB]
329
330
            else:
                # define tile positions and size
331
                def get_tilepos_list(GMSfile):
332
333
                    return get_array_tilebounds(array_shape=INP_R.GMSfile2dict(GMSfile)['shape_fullArr'],
                                                tile_shape=blocksize)
334
335
336

                # get input parameters for creating GMS objects as blocks
                work = [[GMSfile, ['block', tp]] for GMSfile in GMSfile_list_prevLvl_inDB
337
                        for tp in get_tilepos_list(GMSfile)]
338

339
340
341
342
            # 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
343
            def init_GMS_obj(): return HLP_F.parentObjDict[prevLvl](*HLP_F.initArgsDict[prevLvl])
344
345
            DB_objs = [init_GMS_obj().from_disk(tuple_GMS_subset=w) for w in work]  # init

346
347
348
349
350
            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

351
352
353
354
    def run_all_processors(self, custom_data_list=None):
        """
        Run all processors at once.
        """
355

356
        signal.signal(signal.SIGINT, self.stop)  # enable clean shutdown possibility
357

358
        # noinspection PyBroadException
359
360
361
362
363
364
        try:
            if self.job.profiling:
                from pyinstrument import Profiler
                self.profiler = Profiler()  # or Profiler(use_signal=False), see below
                self.profiler.start()

365
366
367
            self.logger.info('Execution of entire GeoMultiSens pre-processing chain started for job ID %s...'
                             % self.job.ID)
            self.DB_job_record.reset_job_progress()  # updates attributes of DB_job_record and related DB entry
368
            self.job.status = 'running'
369
            self.update_DB_job_record()  # TODO implement that into job.status.setter
370
371
372
373
374
375
376
377
378

            self.failed_objects = []

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

            # add local availability
            self.usecase.data_list = MAP(self.add_local_availability, self.usecase.data_list)
379
            self.update_DB_job_statistics(self.usecase.data_list)
380
381
382
383
384
385
386
387
388
389
390
391

            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.')
392
393
394
395
            # TODO implement failed_with_warnings:
            self.job.status = 'finished' if not self.failed_objects else 'finished_with_errors'
            self.job.end_time = datetime.datetime.now()
            self.job.computation_time = self.job.end_time - self.job.start_time
396
397
398
399
400
401
402
403
404
405
406
            self.logger.info('Time for execution: %s' % self.job.computation_time)

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

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

            shutdown_loggers()

407
        except Exception:
408
409
410
411
412
413
414
415
            if self.job.profiling:
                self.profiler.stop()
                print(self.profiler.output_text(unicode=True, color=True))

            self.job.status = 'failed'
            self.update_DB_job_record()

            if not self.job.disable_exception_handler:
416
                self.logger.error('Execution failed with an error:', exc_info=True)
417
418
                shutdown_loggers()
            else:
419
                self.logger.error('Execution failed with an error:')
420
421
                shutdown_loggers()
                raise
422

423
424
    def stop(self, signum, frame):
        """Interrupt the running process controller gracefully."""
425

426
        self.job.status = 'canceled'
427
428
        self.update_DB_job_record()

429
        self.logger.warning('Process controller stopped by user.')
430
431
        del self.logger
        shutdown_loggers()
432

433
        raise KeyboardInterrupt  # terminate execution and show traceback
434

435
436
437
438
    def benchmark(self):
        """
        Run a benchmark.
        """
439
        data_list_bench = self.usecase.data_list
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
        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]
                if os.path.exists(self.job.path_database):
                    os.remove(self.job.path_database)
                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):
        """
456
        Run Level 1A processing: Data import and metadata homogenization
457
        """
458
        if self.job.exec_L1AP[0]:
459
            self.logger.info('\n\n##### Level 1A Processing started - raster format and metadata homogenization ####\n')
460

461
462
463
464
            datalist_L1A_P = self._get_processor_data_list('L1A')

            if self.parallLev == 'scenes':
                # map
465
                L1A_resObjects = MAP(L1A_map, datalist_L1A_P, CPUs=12)
466
            else:  # tiles
467
468
                all_L1A_tiles_map1 = MAP(L1A_map_1, datalist_L1A_P,
                                         flatten_output=True)  # map_1 # merge results to new list of splits
469

470
471
472
                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
473

474
                L1A_objects = MAP(L1A_P.L1A_object().from_tiles, grouped_L1A_Tiles)  # reduce
475

476
                L1A_resObjects = MAP(L1A_map_3, L1A_objects)  # map_3
477

478
            self.L1A_newObjects = [obj for obj in L1A_resObjects if isinstance(obj, L1A_P.L1A_object)]
479
            self.failed_objects += [obj for obj in L1A_resObjects if isinstance(obj, failed_GMS_object) and
480
481
482
483
484
485
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L1A_newObjects

    def L1B_processing(self):
        """
486
        Run Level 1B processing: calculation of geometric shifts
487
488
489
490
        """
        # TODO implement check for running spatial index mediator server
        # run on full cubes

491
        if self.job.exec_L1BP[0]:
492
            self.logger.info('\n\n####### Level 1B Processing started - detection of geometric displacements #######\n')
493

494
495
            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
496

497
            L1B_resObjects = MAP(L1B_map, L1A_Instances)
498

499
500
501
            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]
502
503
504
505
506

        return self.L1B_newObjects

    def L1C_processing(self):
        """
507
        Run Level 1C processing: atmospheric correction
508
        """
509
        if self.job.exec_L1CP[0]:
510
            self.logger.info('\n\n############## Level 1C Processing started - atmospheric correction ##############\n')
511

512
            if self.parallLev == 'scenes':
513
514
515
516
                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)
517
                grouped_L1B_Instances = HLP_F.group_objects_by_attributes(L1B_Instances, 'scene_ID')
518

519
520
                L1C_resObjects = MAP(L1C_map, grouped_L1B_Instances, flatten_output=True,
                                     CPUs=15)  # FIXME CPUs set to 15 for testing
521

522
            else:  # tiles
523
524
                raise NotImplementedError("Tiled processing is not yet completely implemented for L1C processor. Use "
                                          "parallelization level 'scenes' instead!")
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
                # 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)]
544
            self.failed_objects += [obj for obj in L1C_resObjects if isinstance(obj, failed_GMS_object) and
545
546
547
548
549
550
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L1C_newObjects

    def L2A_processing(self):
        """
551
        Run Level 2A processing: geometric homogenization
552
        """
553
        if self.job.exec_L2AP[0]:
554
555
            self.logger.info(
                '\n\n#### Level 2A Processing started - shift correction / geometric homogenization ####\n')
556

557
            """combine newly and earlier processed L1C data"""
558
559
            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
560
561
562
563

            # 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')

564
            L2A_resTiles = MAP(L2A_map, grouped_L1C_Instances, flatten_output=True)
565

566
            self.L2A_tiles = [obj for obj in L2A_resTiles if isinstance(obj, L2A_P.L2A_object)]
567
            self.failed_objects += [obj for obj in L2A_resTiles if isinstance(obj, failed_GMS_object) and
568
569
570
571
572
573
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L2A_tiles

    def L2B_processing(self):
        """
574
        Run Level 2B processing: spectral homogenization
575
        """
576
        if self.job.exec_L2BP[0]:
577
            self.logger.info('\n\n############# Level 2B Processing started - spectral homogenization ##############\n')
578

579
580
            if self.parallLev == 'scenes':
                # don't know if scenes makes sense in L2B processing because full objects are very big!
581
                """if newly processed L2A objects are present: merge them to scenes"""
582
583
                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
584
                # L2A_newObjects  = MAP(HLP_F.merge_GMS_tiles_to_GMS_obj, grouped_L2A_Tiles)
585
                L2A_newObjects = [L2A_P.L2A_object().from_tiles(tileList) for tileList in grouped_L2A_Tiles]
586

587
                """combine newly and earlier processed L2A data"""
588
589
                L2A_DBObjects = self.get_DB_objects('L2B', self.L2A_tiles)
                L2A_Instances = L2A_newObjects + L2A_DBObjects  # combine newly and earlier processed L2A data
590

591
                L2B_resObjects = MAP(L2B_map, L2A_Instances)
592
593

            else:  # tiles
594
                L2A_newTiles = self.L2A_tiles  # tiles have the block size specified in L2A_map_2
595
596

                """combine newly and earlier processed L2A data"""
597
598
599
                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
600

601
                L2B_tiles = MAP(L2B_map, L2A_tiles)
602
603

                grouped_L2B_Tiles = \
604
605
                    HLP_F.group_objects_by_attributes(L2B_tiles,
                                                      'scene_ID')  # group results # FIXME nötig an dieser Stelle?
606
607
                [L2B_tiles_group[0].delete_tempFiles() for L2B_tiles_group in grouped_L2B_Tiles]

608
                L2B_resObjects = [L2B_P.L2B_object().from_tiles(tileList) for tileList in grouped_L2B_Tiles]
609

610
            self.L2B_newObjects = [obj for obj in L2B_resObjects if isinstance(obj, L2B_P.L2B_object)]
611
            self.failed_objects += [obj for obj in L2B_resObjects if isinstance(obj, failed_GMS_object) and
612
613
614
615
616
617
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L2B_newObjects

    def L2C_processing(self):
        """
618
        Run Level 2C processing: accurracy assessment and MGRS tiling
619
        """
620
        # FIXME only parallelization_level == 'scenes' implemented
621
        if self.job.exec_L2CP[0]:
622
            self.logger.info('\n\n########## Level 2C Processing started - calculation of quality layers ###########\n')
623

624
            """combine newly and earlier processed L2A data"""
625
626
            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
627

628
            L2C_resObjects = MAP(L2C_map, L2B_Instances, CPUs=8)  # FIXME 8 workers due to heavy IO
629
            # FIXME in Flink mode results are too big to be back-pickled
630
            self.L2C_newObjects = [obj for obj in L2C_resObjects if isinstance(obj, L2C_P.L2C_object)]
631
            self.failed_objects += [obj for obj in L2C_resObjects if isinstance(obj, failed_GMS_object) and
632
633
634
635
636
637
638
639
                                    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').
        """
640
        # TODO move this method to config.Job
641
642
        # update 'failed_sceneids' column of job record within jobs table
        sceneids_failed = list(set([obj.scene_ID for obj in self.failed_objects]))
643
644
        DB_T.update_records_in_postgreSQLdb(
            self.job.conn_database, 'jobs',
645
646
647
            {'failed_sceneids': sceneids_failed,  # update 'failed_sceneids' column
             'finishtime': self.job.end_time,  # add job finish timestamp
             'status': self.job.status},  # update 'job_status' column
648
            {'id': self.job.ID})
649

650
651
652
653
654
    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
655
        already_updated_IDs = []
656
        for ds in usecase_datalist:
657
658
            if ds['proc_level'] is not None and ds['scene_ID'] not in already_updated_IDs:
                # update statistics column of jobs table
659
660
661
662
663
                DB_T.increment_decrement_arrayCol_in_postgreSQLdb(
                    self.job.conn_database, 'jobs', 'statistics', cond_dict={'id': self.job.ID},
                    idx_val2decrement=db_jobs_statistics_def['downloaded'],
                    idx_val2increment=db_jobs_statistics_def[ds['proc_level']])

664
665
666
                # avoid double updating in case of subsystems belonging to the same scene ID
                already_updated_IDs.append(ds['scene_ID'])

667
668
669
670
    def create_job_summary(self):
        """
        Create job success summary
        """
671
672
673

        # get objects with highest requested processing level
        highest_procL_Objs = []
674
        for pL in reversed(proc_chain):
675
676
            if getattr(self.job, 'exec_%sP' % pL)[0]:
                highest_procL_Objs = getattr(self, '%s_newObjects' % pL) if pL != 'L2A' else self.L2A_tiles
677
678
                break

679
680
681
682
683
        gms_objects2summarize = highest_procL_Objs + self.failed_objects
        if gms_objects2summarize:
            # create summaries
            detailed_JS, quick_JS = get_job_summary(gms_objects2summarize)
            detailed_JS.to_excel(os.path.join(self.job.path_job_logs, '%s_summary.xlsx' % self.job.ID))
684
            detailed_JS.to_csv(os.path.join(self.job.path_job_logs, '%s_summary.csv' % self.job.ID), sep='\t')
685
686
687
            self.logger.info('\nQUICK JOB SUMMARY (ID %s):\n' % self.job.ID + quick_JS.to_string())

            self.summary_detailed = detailed_JS
688
            self.summary_quick = quick_JS
689
690
691
692
693

        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.")
694
695
696
697
698
699

    def clear_lists_procObj(self):
        self.failed_objects = []
        self.L1A_newObjects = []
        self.L1B_newObjects = []
        self.L1C_newObjects = []
700
        self.L2A_tiles = []
701
        self.L2B_newObjects = []
702
703
704
705
706
        self.L2C_newObjects = []


def get_job_summary(list_GMS_objects):
    # get detailed job summary
707
708
    DJS_cols = ['GMS_object', 'scene_ID', 'entity_ID', 'satellite', 'sensor', 'subsystem', 'image_type', 'proc_level',
                'arr_shape', 'arr_pos', 'failedMapper', 'ExceptionType', 'ExceptionValue', 'ExceptionTraceback']
709
710
711
712
    DJS = DataFrame(columns=DJS_cols)
    DJS['GMS_object'] = list_GMS_objects

    for col in DJS_cols[1:]:
713
714
        def get_val(obj): return getattr(obj, col) if hasattr(obj, col) else None
        DJS[col] = list(DJS['GMS_object'].map(get_val))
715
716

    del DJS['GMS_object']
717
    DJS = DJS.sort_values(by=['satellite', 'sensor', 'entity_ID'])
718
719

    # get quick job summary
720
721
722
723
    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
724
    # count objects with the same satellite/sensor/sceneid combination
725
726
    QJS['count'] = [len(DJS[(DJS['satellite'] == sat) & (DJS['sensor'] == sen)]['scene_ID'].unique())
                    for sat, sen in zip(all_sat, all_sen)]
727
    QJS['proc_successfully'] = [len(DJS[(DJS['satellite'] == sat) &
728
729
                                        (DJS['sensor'] == sen) &
                                        (DJS['failedMapper'].isnull())]['scene_ID'].unique())
730
                                for sat, sen in zip(all_sat, all_sen)]
731
    QJS['proc_failed'] = QJS['count'] - QJS['proc_successfully']
732
733
    QJS = QJS.sort_values(by=['satellite', 'sensor'])
    return DJS, QJS