process_controller.py 37.1 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
Bugfix.    
Daniel Scheffler committed
13
from typing import TYPE_CHECKING, List
14

15
16
from ..io import output_writer as OUT_W
from ..io import input_reader as INP_R
17
18
19
20
21
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
22
from ..model.metadata import get_LayerBandsAssignment
23
24
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,
25
                       L2A_map, L2B_map, L2C_map)
26
from ..options.config import set_config, GMS_config
27
from .multiproc import MAP
28
from ..misc.definition_dicts import proc_chain, db_jobs_statistics_def
29

30
31
from py_tools_ds.numeric.array import get_array_tilebounds

32
if TYPE_CHECKING:
Daniel Scheffler's avatar
Daniel Scheffler committed
33
    from collections import OrderedDict  # noqa F401  # flake8 issue
34
35
36
37


__author__ = 'Daniel Scheffler'

38
39

class process_controller(object):
40
    def __init__(self, job_ID, **config_kwargs):
41
        """gms_preprocessing process controller
42

43
44
        :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()
45
46
47
        """

        # assertions
48
49
        if not isinstance(job_ID, int):
            raise ValueError("'job_ID' must be an integer value. Got %s." % type(job_ID))
50

51
52
53
54
55
56
        # set GMS configuration
        config_kwargs.update(dict(reset_status=True))
        set_config(job_ID, **config_kwargs)
        self.config = GMS_config  # type: GMS_config

        # defaults
57
        self._logger = None
58
        self._DB_job_record = None
59
        self.profiler = None
60
61
62
63
64

        self.failed_objects = []
        self.L1A_newObjects = []
        self.L1B_newObjects = []
        self.L1C_newObjects = []
65
        self.L2A_tiles = []
66
67
68
69
        self.L2B_newObjects = []
        self.L2C_newObjects = []

        self.summary_detailed = None
70
        self.summary_quick = None
71

72
73
        # 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
74
        # if isdebugging:  # override the existing settings in order to get write access everywhere
75
76
        #    pass

77
        # called_from_iPyNb = 1 if 'ipykernel/__main__.py' in sys.argv[0] else 0
78

79
        self.logger.info('Process Controller initialized for job ID %s (comment: %s).'
80
                         % (self.config.ID, self.DB_job_record.comment))
81

82
        if self.config.delete_old_output:
83
84
            self.logger.info('Deleting previously processed data...')
            self.DB_job_record.delete_procdata_of_entire_job(force=True)
85

86
87
88
89
90
    @property
    def logger(self):
        if self._logger and self._logger.handlers[:]:
            return self._logger
        else:
91
92
93
            self._logger = GMS_logger('log__%s' % self.config.ID,
                                      path_logfile=os.path.join(self.config.path_job_logs, '%s.log' % self.config.ID),
                                      log_level=self.config.log_level, append=False)
94
95
96
97
98
99
100
101
102
103
104
105
106
            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
107
108
109
110
    def DB_job_record(self):
        if self._DB_job_record:
            return self._DB_job_record
        else:
111
112
            self._DB_job_record = DB_T.GMS_JOB(self.config.conn_database)
            self._DB_job_record.from_job_ID(self.config.ID)
113
            return self._DB_job_record
114

115
116
117
    @DB_job_record.setter
    def DB_job_record(self, value):
        self._DB_job_record = value
118

119
120
121
    @property
    def sceneids_failed(self):
        return [obj.scene_ID for obj in self.failed_objects]
122

123
124
    def _add_local_availability_single_dataset(self, dataset):
        # type: (OrderedDict) -> OrderedDict
125
        # TODO revise this function
126
127
128
129
        # query the database and get the last written processing level and LayerBandsAssignment
        DB_match = DB_T.get_info_from_postgreSQLdb(
            GMS_config.conn_database, 'scenes_proc', ['proc_level', 'layer_bands_assignment'],
            dict(sceneid=dataset['scene_ID']))
Daniel Scheffler's avatar
Daniel Scheffler committed
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
        # get the corresponding logfile
        path_logfile = path_generator(
            dataset).get_path_logfile()  # FIXME this always returns the logfile for the subsystem.

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

        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."""

            if not os.path.exists(path_log):
                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)))
                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 = []
                    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
            AllWrittenProcL = get_AllWrittenProcL_dueLog(path_logfile)
            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'],
                            Subsystem=dataset['subsystem'],
                            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
207
                                dataset['proc_level'] = ProcL
208

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
209
                    else:
210
211
212
213
214
215
216
217
                        self.logger.info('Found a matching dataset for %s but with a different '
                                         'LayerBandsAssignment. Dataset has to be reprocessed.'
                                         % dataset['entity_ID'])
                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 '')
218

219
220
221
222
        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
223

224
225
        else:
            dataset['proc_level'] = None
226

227
        return dataset
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
228

229
230
231
    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
232

233
234
235
236
237
238
        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]
239

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
240
241
        datasets_validated = []
        datasets_grouped = HLP_F.group_dicts_by_key(datasets, key='scene_ID')
242

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
243
244
        for ds_group in datasets_grouped:
            proc_lvls = [ds['proc_level'] for ds in ds_group]
245

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
246
247
            if not len(list(set(proc_lvls))) == 1:
                # reset processing level of those scenes where not all subsystems are available
248
                self.logger.info('%s: Found already processed subsystems at different processing levels %s. '
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
249
                                 'Dataset has to be reprocessed to avoid errors'
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
250
                                 % (ds_group[0]['entity_ID'], proc_lvls))
251

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
252
253
254
255
256
                for ds in ds_group:
                    ds['proc_level'] = None
                    datasets_validated += ds
            else:
                datasets_validated.extend(ds_group)
257

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
258
        return datasets_validated
259

260
261
    @staticmethod
    def _is_inMEM(GMS_objects, dataset):
Daniel Scheffler's avatar
Daniel Scheffler committed
262
        # type: (list, OrderedDict) -> bool
263
264
265
266
        """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()
        """
267
        # check if the scene ID of the given dataset is in the scene IDs of the previously processed datasets
268
269
270
        return dataset['scene_ID'] in [obj.scene_ID for obj in GMS_objects]

    def _get_processor_data_list(self, procLvl, prevLvl_objects=None):
271
        """Returns a list of datasets that have to be read from disk and then processed by a specific processor.
272
273
274
275
276

        :param procLvl:
        :param prevLvl_objects:
        :return:
        """
Daniel Scheffler's avatar
Daniel Scheffler committed
277
278
        def is_procL_lower(dataset):
            return HLP_F.is_proc_level_lower(dataset['proc_level'], target_lvl=procLvl)
279
280

        if prevLvl_objects is None:
Daniel Scheffler's avatar
Daniel Scheffler committed
281
            return [dataset for dataset in self.config.data_list if is_procL_lower(dataset)]  # TODO generator?
282
        else:
Daniel Scheffler's avatar
Daniel Scheffler committed
283
            return [dataset for dataset in self.config.data_list if is_procL_lower(dataset) and
284
                    not self._is_inMEM(prevLvl_objects + self.failed_objects, dataset)]
285
286
287
288
289
290
291
292
293
294
295
296
297

    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
298
        if procLvl == 'L1A':
299
300
301
            return []
        else:
            # handle input parameters
Daniel Scheffler's avatar
Daniel Scheffler committed
302
303
            parallLev = parallLev or self.config.parallelization_level
            blocksize = blocksize or self.config.tiling_block_size_XY
304
            prevLvl = proc_chain[proc_chain.index(procLvl) - 1]  # TODO replace by enum
305
306

            # get GMSfile list
307
            dataset_dicts = self._get_processor_data_list(procLvl, prevLvl_objects)
308
309
310
311
312
            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
313
                work = [[GMS, ['cube', None]] for GMS in GMSfile_list_prevLvl_inDB]
314
315
            else:
                # define tile positions and size
316
                def get_tilepos_list(GMSfile):
317
318
                    return get_array_tilebounds(array_shape=INP_R.GMSfile2dict(GMSfile)['shape_fullArr'],
                                                tile_shape=blocksize)
319
320
321

                # get input parameters for creating GMS objects as blocks
                work = [[GMSfile, ['block', tp]] for GMSfile in GMSfile_list_prevLvl_inDB
322
                        for tp in get_tilepos_list(GMSfile)]
323

324
325
326
327
            # 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
328
            def init_GMS_obj(): return HLP_F.parentObjDict[prevLvl](*HLP_F.initArgsDict[prevLvl])
329
330
            DB_objs = [init_GMS_obj().from_disk(tuple_GMS_subset=w) for w in work]  # init

331
332
333
334
335
            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
336
    def run_all_processors_OLD(self, custom_data_list=None):
337
338
339
        """
        Run all processors at once.
        """
340

341
        signal.signal(signal.SIGINT, self.stop)  # enable clean shutdown possibility
342

343
        # noinspection PyBroadException
344
        try:
345
            if self.config.profiling:
346
347
348
349
                from pyinstrument import Profiler
                self.profiler = Profiler()  # or Profiler(use_signal=False), see below
                self.profiler.start()

350
            self.logger.info('Execution of entire GeoMultiSens pre-processing chain started for job ID %s...'
351
                             % self.config.ID)
352
            self.DB_job_record.reset_job_progress()  # updates attributes of DB_job_record and related DB entry
353
            self.config.status = 'running'
354
            self.update_DB_job_record()  # TODO implement that into job.status.setter
355
356
357
358
359

            self.failed_objects = []

            # get list of datasets to be processed
            if custom_data_list:
360
                self.config.data_list = custom_data_list
361
362

            # add local availability
Daniel Scheffler's avatar
Daniel Scheffler committed
363
            self.config.data_list = self.add_local_availability(self.config.data_list)
364
            self.update_DB_job_statistics(self.config.data_list)
365
366
367
368
369
370
371
372
373
374
375
376

            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.')
377
            # TODO implement failed_with_warnings:
378
379
380
381
            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)
382
383
384
385

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

386
            if self.config.profiling:
387
388
389
390
391
                self.profiler.stop()
                print(self.profiler.output_text(unicode=True, color=True))

            shutdown_loggers()

392
        except Exception:  # noqa E722  # bare except
393
            if self.config.profiling:
394
395
396
                self.profiler.stop()
                print(self.profiler.output_text(unicode=True, color=True))

397
            self.config.status = 'failed'
Daniel Scheffler's avatar
Daniel Scheffler committed
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
            self.update_DB_job_record()

            if not self.config.disable_exception_handler:
                self.logger.error('Execution failed with an error:', exc_info=True)
                shutdown_loggers()
            else:
                self.logger.error('Execution failed with an error:')
                shutdown_loggers()
                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)
            self.DB_job_record.reset_job_progress()  # updates attributes of DB_job_record and related DB entry
            self.config.status = 'running'
            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')

            from .pipeline import run_complete_preprocessing
            GMS_objs = MAP(run_complete_preprocessing, dataset_groups)

            # separate results into successful and failed objects
            self.L2C_newObjects = [obj for obj in GMS_objs if isinstance(obj, L2C_P.L2C_object)]
            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.')
            # TODO implement failed_with_warnings:
            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)

            # 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))

            shutdown_loggers()

        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'
469
470
            self.update_DB_job_record()

471
            if not self.config.disable_exception_handler:
472
                self.logger.error('Execution failed with an error:', exc_info=True)
473
474
                shutdown_loggers()
            else:
475
                self.logger.error('Execution failed with an error:')
476
477
                shutdown_loggers()
                raise
478

479
480
    def stop(self, signum, frame):
        """Interrupt the running process controller gracefully."""
481

482
        self.config.status = 'canceled'
483
484
        self.update_DB_job_record()

485
        self.logger.warning('Process controller stopped by user.')
486
487
        del self.logger
        shutdown_loggers()
488

489
        raise KeyboardInterrupt  # terminate execution and show traceback
490

491
492
493
494
    def benchmark(self):
        """
        Run a benchmark.
        """
495
        data_list_bench = self.config.data_list
496
497
498
499
        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]
500
501
                if os.path.exists(self.config.path_database):
                    os.remove(self.config.path_database)
502
503
504
505
506
507
508
509
510
511
                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):
        """
512
        Run Level 1A processing: Data import and metadata homogenization
513
        """
514
        if self.config.exec_L1AP[0]:
515
            self.logger.info('\n\n##### Level 1A Processing started - raster format and metadata homogenization ####\n')
516

517
518
            datalist_L1A_P = self._get_processor_data_list('L1A')

519
            if self.config.parallelization_level == 'scenes':
520
                # map
521
                L1A_resObjects = MAP(L1A_map, datalist_L1A_P, CPUs=12)
522
            else:  # tiles
523
524
                all_L1A_tiles_map1 = MAP(L1A_map_1, datalist_L1A_P,
                                         flatten_output=True)  # map_1 # merge results to new list of splits
525

526
527
528
                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
529

530
                L1A_objects = MAP(L1A_P.L1A_object().from_tiles, grouped_L1A_Tiles)  # reduce
531

532
                L1A_resObjects = MAP(L1A_map_3, L1A_objects)  # map_3
533

534
            self.L1A_newObjects = [obj for obj in L1A_resObjects if isinstance(obj, L1A_P.L1A_object)]
535
            self.failed_objects += [obj for obj in L1A_resObjects if isinstance(obj, failed_GMS_object) and
536
537
538
539
540
541
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L1A_newObjects

    def L1B_processing(self):
        """
542
        Run Level 1B processing: calculation of geometric shifts
543
544
545
546
        """
        # TODO implement check for running spatial index mediator server
        # run on full cubes

547
        if self.config.exec_L1BP[0]:
548
            self.logger.info('\n\n####### Level 1B Processing started - detection of geometric displacements #######\n')
549

550
551
            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
552

553
            L1B_resObjects = MAP(L1B_map, L1A_Instances)
554

555
556
557
            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]
558
559
560
561
562

        return self.L1B_newObjects

    def L1C_processing(self):
        """
563
        Run Level 1C processing: atmospheric correction
564
        """
565
        if self.config.exec_L1CP[0]:
566
            self.logger.info('\n\n############## Level 1C Processing started - atmospheric correction ##############\n')
567

568
            if self.config.parallelization_level == 'scenes':
569
570
571
572
                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)
573
                grouped_L1B_Instances = HLP_F.group_objects_by_attributes(L1B_Instances, 'scene_ID')
574

575
576
                L1C_resObjects = MAP(L1C_map, grouped_L1B_Instances, flatten_output=True,
                                     CPUs=15)  # FIXME CPUs set to 15 for testing
577

578
            else:  # tiles
579
580
                raise NotImplementedError("Tiled processing is not yet completely implemented for L1C processor. Use "
                                          "parallelization level 'scenes' instead!")
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
                # 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)]
600
            self.failed_objects += [obj for obj in L1C_resObjects if isinstance(obj, failed_GMS_object) and
601
602
603
604
605
606
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L1C_newObjects

    def L2A_processing(self):
        """
607
        Run Level 2A processing: geometric homogenization
608
        """
609
        if self.config.exec_L2AP[0]:
610
611
            self.logger.info(
                '\n\n#### Level 2A Processing started - shift correction / geometric homogenization ####\n')
612

613
            """combine newly and earlier processed L1C data"""
614
615
            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
616
617
618
619

            # 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')

620
            L2A_resTiles = MAP(L2A_map, grouped_L1C_Instances, flatten_output=True)
621

622
            self.L2A_tiles = [obj for obj in L2A_resTiles if isinstance(obj, L2A_P.L2A_object)]
623
            self.failed_objects += [obj for obj in L2A_resTiles if isinstance(obj, failed_GMS_object) and
624
625
626
627
628
629
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L2A_tiles

    def L2B_processing(self):
        """
630
        Run Level 2B processing: spectral homogenization
631
        """
632
        if self.config.exec_L2BP[0]:
633
            self.logger.info('\n\n############# Level 2B Processing started - spectral homogenization ##############\n')
634

635
            if self.config.parallelization_level == 'scenes':
636
                # don't know if scenes makes sense in L2B processing because full objects are very big!
637
                """if newly processed L2A objects are present: merge them to scenes"""
638
639
                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
640
                # L2A_newObjects  = MAP(HLP_F.merge_GMS_tiles_to_GMS_obj, grouped_L2A_Tiles)
641
                L2A_newObjects = [L2A_P.L2A_object().from_tiles(tileList) for tileList in grouped_L2A_Tiles]
642

643
                """combine newly and earlier processed L2A data"""
644
645
                L2A_DBObjects = self.get_DB_objects('L2B', self.L2A_tiles)
                L2A_Instances = L2A_newObjects + L2A_DBObjects  # combine newly and earlier processed L2A data
646

647
                L2B_resObjects = MAP(L2B_map, L2A_Instances)
648
649

            else:  # tiles
650
                L2A_newTiles = self.L2A_tiles  # tiles have the block size specified in L2A_map_2
651
652

                """combine newly and earlier processed L2A data"""
653
654
655
                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
656

657
                L2B_tiles = MAP(L2B_map, L2A_tiles)
658
659

                grouped_L2B_Tiles = \
660
661
                    HLP_F.group_objects_by_attributes(L2B_tiles,
                                                      'scene_ID')  # group results # FIXME nötig an dieser Stelle?
662
663
                [L2B_tiles_group[0].delete_tempFiles() for L2B_tiles_group in grouped_L2B_Tiles]

664
                L2B_resObjects = [L2B_P.L2B_object().from_tiles(tileList) for tileList in grouped_L2B_Tiles]
665

666
            self.L2B_newObjects = [obj for obj in L2B_resObjects if isinstance(obj, L2B_P.L2B_object)]
667
            self.failed_objects += [obj for obj in L2B_resObjects if isinstance(obj, failed_GMS_object) and
668
669
670
671
672
673
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L2B_newObjects

    def L2C_processing(self):
        """
674
        Run Level 2C processing: accurracy assessment and MGRS tiling
675
        """
676
        # FIXME only parallelization_level == 'scenes' implemented
677
        if self.config.exec_L2CP[0]:
678
            self.logger.info('\n\n########## Level 2C Processing started - calculation of quality layers ###########\n')
679

680
            """combine newly and earlier processed L2A data"""
681
682
            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
683

684
            L2C_resObjects = MAP(L2C_map, L2B_Instances, CPUs=8)  # FIXME 8 workers due to heavy IO
685
            # FIXME in case of inmem_serialization mode results are too big to be back-pickled
686
            self.L2C_newObjects = [obj for obj in L2C_resObjects if isinstance(obj, L2C_P.L2C_object)]
687
            self.failed_objects += [obj for obj in L2C_resObjects if isinstance(obj, failed_GMS_object) and
688
689
690
691
692
693
694
695
                                    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').
        """
696
        # TODO move this method to config.Job
697
698
        # update 'failed_sceneids' column of job record within jobs table
        sceneids_failed = list(set([obj.scene_ID for obj in self.failed_objects]))
699
        DB_T.update_records_in_postgreSQLdb(
700
            self.config.conn_database, 'jobs',
701
            {'failed_sceneids': sceneids_failed,  # update 'failed_sceneids' column
702
703
704
             'finishtime': self.config.end_time,  # add job finish timestamp
             'status': self.config.status},  # update 'job_status' column
            {'id': self.config.ID})
705

706
707
708
709
710
    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
711
        already_updated_IDs = []
712
        for ds in usecase_datalist:
713
714
            if ds['proc_level'] is not None and ds['scene_ID'] not in already_updated_IDs:
                # update statistics column of jobs table
715
                DB_T.increment_decrement_arrayCol_in_postgreSQLdb(
716
                    self.config.conn_database, 'jobs', 'statistics', cond_dict={'id': self.config.ID},
717
718
719
                    idx_val2decrement=db_jobs_statistics_def['downloaded'],
                    idx_val2increment=db_jobs_statistics_def[ds['proc_level']])

720
721
722
                # avoid double updating in case of subsystems belonging to the same scene ID
                already_updated_IDs.append(ds['scene_ID'])

723
724
725
726
    def create_job_summary(self):
        """
        Create job success summary
        """
727
728
729

        # get objects with highest requested processing level
        highest_procL_Objs = []
730
        for pL in reversed(proc_chain):
731
            if getattr(self.config, 'exec_%sP' % pL)[0]:
732
                highest_procL_Objs = getattr(self, '%s_newObjects' % pL) if pL != 'L2A' else self.L2A_tiles
733
734
                break

735
736
737
738
        gms_objects2summarize = highest_procL_Objs + self.failed_objects
        if gms_objects2summarize:
            # create summaries
            detailed_JS, quick_JS = get_job_summary(gms_objects2summarize)
739
740
741
            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())
742
743

            self.summary_detailed = detailed_JS
744
            self.summary_quick = quick_JS
745
746
747
748
749

        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.")
750
751
752
753
754
755

    def clear_lists_procObj(self):
        self.failed_objects = []
        self.L1A_newObjects = []
        self.L1B_newObjects = []
        self.L1C_newObjects = []
756
        self.L2A_tiles = []
757
        self.L2B_newObjects = []
758
759
760
761
762
        self.L2C_newObjects = []


def get_job_summary(list_GMS_objects):
    # get detailed job summary
763
764
    DJS_cols = ['GMS_object', 'scene_ID', 'entity_ID', 'satellite', 'sensor', 'subsystem', 'image_type', 'proc_level',
                'arr_shape', 'arr_pos', 'failedMapper', 'ExceptionType', 'ExceptionValue', 'ExceptionTraceback']
765
766
767
768
    DJS = DataFrame(columns=DJS_cols)
    DJS['GMS_object'] = list_GMS_objects

    for col in DJS_cols[1:]:
769
770
        def get_val(obj): return getattr(obj, col) if hasattr(obj, col) else None
        DJS[col] = list(DJS['GMS_object'].map(get_val))
771
772

    del DJS['GMS_object']
773
    DJS = DJS.sort_values(by=['satellite', 'sensor', 'entity_ID'])
774
775

    # get quick job summary
776
777
778
779
    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
780
    # count objects with the same satellite/sensor/sceneid combination
781
782
    QJS['count'] = [len(DJS[(DJS['satellite'] == sat) & (DJS['sensor'] == sen)]['scene_ID'].unique())
                    for sat, sen in zip(all_sat, all_sen)]
783
    QJS['proc_successfully'] = [len(DJS[(DJS['satellite'] == sat) &
784
785
                                        (DJS['sensor'] == sen) &
                                        (DJS['failedMapper'].isnull())]['scene_ID'].unique())
786
                                for sat, sen in zip(all_sat, all_sen)]
787
    QJS['proc_failed'] = QJS['count'] - QJS['proc_successfully']
788
789
    QJS = QJS.sort_values(by=['satellite', 'sensor'])
    return DJS, QJS