process_controller.py 37.2 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

Daniel Scheffler's avatar
Daniel Scheffler committed
123
    def add_local_availability(self, datasets):
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
124
        # type: (List[OrderedDict]) -> List[OrderedDict]
125
        # TODO revise this function
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
126
        def check_single_dataset(dataset):
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
            # 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']))

            # 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 = []
Daniel Scheffler's avatar
Daniel Scheffler committed
146
                else:
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
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
                    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)
Daniel Scheffler's avatar
Daniel Scheffler committed
192

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
193
                                dataset['proc_level'] = ProcL
194

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
                            # 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:
                                    dataset['proc_level'] = ProcL

                        else:
                            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 '')
219

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
220
221
222
223
            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
224

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
225
226
            else:
                dataset['proc_level'] = None
227

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
228
229
230
231
            return dataset

        datasets = MAP(check_single_dataset, datasets)

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
232
233
234
        #############################################################
        # check availability of all subsets per scene an proc_level #
        #############################################################
235

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
236
237
        datasets_validated = []
        datasets_grouped = HLP_F.group_dicts_by_key(datasets, key='scene_ID')
238

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
239
240
        for ds_group in datasets_grouped:
            proc_lvls = [ds['proc_level'] for ds in ds_group]
241

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
242
243
            if not len(list(set(proc_lvls))) == 1:
                # reset processing level of those scenes where not all subsystems are available
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
244
245
                self.logger.info('%s: Found already processed subsystems at differnt processing levels %s. '
                                 'Dataset has to be reprocessed to avoid errors'
Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
246
                                 % (ds_group[0]['entity_ID'], proc_lvls))
247

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
248
249
250
251
252
                for ds in ds_group:
                    ds['proc_level'] = None
                    datasets_validated += ds
            else:
                datasets_validated.extend(ds_group)
253

Daniel Scheffler's avatar
Bugfix.    
Daniel Scheffler committed
254
        return datasets_validated
255

256
257
    @staticmethod
    def _is_inMEM(GMS_objects, dataset):
Daniel Scheffler's avatar
Daniel Scheffler committed
258
        # type: (list, OrderedDict) -> bool
259
260
261
262
        """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()
        """
263
        # check if the scene ID of the given dataset is in the scene IDs of the previously processed datasets
264
265
266
        return dataset['scene_ID'] in [obj.scene_ID for obj in GMS_objects]

    def _get_processor_data_list(self, procLvl, prevLvl_objects=None):
267
        """Returns a list of datasets that have to be read from disk and then processed by a specific processor.
268
269
270
271
272

        :param procLvl:
        :param prevLvl_objects:
        :return:
        """
Daniel Scheffler's avatar
Daniel Scheffler committed
273
274
        def is_procL_lower(dataset):
            return HLP_F.is_proc_level_lower(dataset['proc_level'], target_lvl=procLvl)
275
276

        if prevLvl_objects is None:
Daniel Scheffler's avatar
Daniel Scheffler committed
277
            return [dataset for dataset in self.config.data_list if is_procL_lower(dataset)]  # TODO generator?
278
        else:
Daniel Scheffler's avatar
Daniel Scheffler committed
279
            return [dataset for dataset in self.config.data_list if is_procL_lower(dataset) and
280
                    not self._is_inMEM(prevLvl_objects + self.failed_objects, dataset)]
281
282
283
284
285
286
287
288
289
290
291
292
293

    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
294
        if procLvl == 'L1A':
295
296
297
            return []
        else:
            # handle input parameters
Daniel Scheffler's avatar
Daniel Scheffler committed
298
299
            parallLev = parallLev or self.config.parallelization_level
            blocksize = blocksize or self.config.tiling_block_size_XY
300
            prevLvl = proc_chain[proc_chain.index(procLvl) - 1]  # TODO replace by enum
301
302

            # get GMSfile list
303
            dataset_dicts = self._get_processor_data_list(procLvl, prevLvl_objects)
304
305
306
307
308
            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
309
                work = [[GMS, ['cube', None]] for GMS in GMSfile_list_prevLvl_inDB]
310
311
            else:
                # define tile positions and size
312
                def get_tilepos_list(GMSfile):
313
314
                    return get_array_tilebounds(array_shape=INP_R.GMSfile2dict(GMSfile)['shape_fullArr'],
                                                tile_shape=blocksize)
315
316
317

                # get input parameters for creating GMS objects as blocks
                work = [[GMSfile, ['block', tp]] for GMSfile in GMSfile_list_prevLvl_inDB
318
                        for tp in get_tilepos_list(GMSfile)]
319

320
321
322
323
            # 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
324
            def init_GMS_obj(): return HLP_F.parentObjDict[prevLvl](*HLP_F.initArgsDict[prevLvl])
325
326
            DB_objs = [init_GMS_obj().from_disk(tuple_GMS_subset=w) for w in work]  # init

327
328
329
330
331
            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
332
    def run_all_processors_OLD(self, custom_data_list=None):
333
334
335
        """
        Run all processors at once.
        """
336

337
        signal.signal(signal.SIGINT, self.stop)  # enable clean shutdown possibility
338

339
        # noinspection PyBroadException
340
        try:
341
            if self.config.profiling:
342
343
344
345
                from pyinstrument import Profiler
                self.profiler = Profiler()  # or Profiler(use_signal=False), see below
                self.profiler.start()

346
            self.logger.info('Execution of entire GeoMultiSens pre-processing chain started for job ID %s...'
347
                             % self.config.ID)
348
            self.DB_job_record.reset_job_progress()  # updates attributes of DB_job_record and related DB entry
349
            self.config.status = 'running'
350
            self.update_DB_job_record()  # TODO implement that into job.status.setter
351
352
353
354
355

            self.failed_objects = []

            # get list of datasets to be processed
            if custom_data_list:
356
                self.config.data_list = custom_data_list
357
358

            # add local availability
Daniel Scheffler's avatar
Daniel Scheffler committed
359
            self.config.data_list = self.add_local_availability(self.config.data_list)
360
            self.update_DB_job_statistics(self.config.data_list)
361
362
363
364
365
366
367
368
369
370
371
372

            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.')
373
            # TODO implement failed_with_warnings:
374
375
376
377
            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)
378
379
380
381

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

382
            if self.config.profiling:
383
384
385
386
387
                self.profiler.stop()
                print(self.profiler.output_text(unicode=True, color=True))

            shutdown_loggers()

388
        except Exception:  # noqa E722  # bare except
389
            if self.config.profiling:
390
391
392
                self.profiler.stop()
                print(self.profiler.output_text(unicode=True, color=True))

393
            self.config.status = 'failed'
Daniel Scheffler's avatar
Daniel Scheffler committed
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
            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'
465
466
            self.update_DB_job_record()

467
            if not self.config.disable_exception_handler:
468
                self.logger.error('Execution failed with an error:', exc_info=True)
469
470
                shutdown_loggers()
            else:
471
                self.logger.error('Execution failed with an error:')
472
473
                shutdown_loggers()
                raise
474

475
476
    def stop(self, signum, frame):
        """Interrupt the running process controller gracefully."""
477

478
        self.config.status = 'canceled'
479
480
        self.update_DB_job_record()

481
        self.logger.warning('Process controller stopped by user.')
482
483
        del self.logger
        shutdown_loggers()
484

485
        raise KeyboardInterrupt  # terminate execution and show traceback
486

487
488
489
490
    def benchmark(self):
        """
        Run a benchmark.
        """
491
        data_list_bench = self.config.data_list
492
493
494
495
        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]
496
497
                if os.path.exists(self.config.path_database):
                    os.remove(self.config.path_database)
498
499
500
501
502
503
504
505
506
507
                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):
        """
508
        Run Level 1A processing: Data import and metadata homogenization
509
        """
510
        if self.config.exec_L1AP[0]:
511
            self.logger.info('\n\n##### Level 1A Processing started - raster format and metadata homogenization ####\n')
512

513
514
            datalist_L1A_P = self._get_processor_data_list('L1A')

515
            if self.config.parallelization_level == 'scenes':
516
                # map
517
                L1A_resObjects = MAP(L1A_map, datalist_L1A_P, CPUs=12)
518
            else:  # tiles
519
520
                all_L1A_tiles_map1 = MAP(L1A_map_1, datalist_L1A_P,
                                         flatten_output=True)  # map_1 # merge results to new list of splits
521

522
523
524
                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
525

526
                L1A_objects = MAP(L1A_P.L1A_object().from_tiles, grouped_L1A_Tiles)  # reduce
527

528
                L1A_resObjects = MAP(L1A_map_3, L1A_objects)  # map_3
529

530
            self.L1A_newObjects = [obj for obj in L1A_resObjects if isinstance(obj, L1A_P.L1A_object)]
531
            self.failed_objects += [obj for obj in L1A_resObjects if isinstance(obj, failed_GMS_object) and
532
533
534
535
536
537
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L1A_newObjects

    def L1B_processing(self):
        """
538
        Run Level 1B processing: calculation of geometric shifts
539
540
541
542
        """
        # TODO implement check for running spatial index mediator server
        # run on full cubes

543
        if self.config.exec_L1BP[0]:
544
            self.logger.info('\n\n####### Level 1B Processing started - detection of geometric displacements #######\n')
545

546
547
            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
548

549
            L1B_resObjects = MAP(L1B_map, L1A_Instances)
550

551
552
553
            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]
554
555
556
557
558

        return self.L1B_newObjects

    def L1C_processing(self):
        """
559
        Run Level 1C processing: atmospheric correction
560
        """
561
        if self.config.exec_L1CP[0]:
562
            self.logger.info('\n\n############## Level 1C Processing started - atmospheric correction ##############\n')
563

564
            if self.config.parallelization_level == 'scenes':
565
566
567
568
                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)
569
                grouped_L1B_Instances = HLP_F.group_objects_by_attributes(L1B_Instances, 'scene_ID')
570

571
572
                L1C_resObjects = MAP(L1C_map, grouped_L1B_Instances, flatten_output=True,
                                     CPUs=15)  # FIXME CPUs set to 15 for testing
573

574
            else:  # tiles
575
576
                raise NotImplementedError("Tiled processing is not yet completely implemented for L1C processor. Use "
                                          "parallelization level 'scenes' instead!")
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
                # 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)]
596
            self.failed_objects += [obj for obj in L1C_resObjects if isinstance(obj, failed_GMS_object) and
597
598
599
600
601
602
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L1C_newObjects

    def L2A_processing(self):
        """
603
        Run Level 2A processing: geometric homogenization
604
        """
605
        if self.config.exec_L2AP[0]:
606
607
            self.logger.info(
                '\n\n#### Level 2A Processing started - shift correction / geometric homogenization ####\n')
608

609
            """combine newly and earlier processed L1C data"""
610
611
            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
612
613
614
615

            # 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')

616
            L2A_resTiles = MAP(L2A_map, grouped_L1C_Instances, flatten_output=True)
617

618
            self.L2A_tiles = [obj for obj in L2A_resTiles if isinstance(obj, L2A_P.L2A_object)]
619
            self.failed_objects += [obj for obj in L2A_resTiles if isinstance(obj, failed_GMS_object) and
620
621
622
623
624
625
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L2A_tiles

    def L2B_processing(self):
        """
626
        Run Level 2B processing: spectral homogenization
627
        """
628
        if self.config.exec_L2BP[0]:
629
            self.logger.info('\n\n############# Level 2B Processing started - spectral homogenization ##############\n')
630

631
            if self.config.parallelization_level == 'scenes':
632
                # don't know if scenes makes sense in L2B processing because full objects are very big!
633
                """if newly processed L2A objects are present: merge them to scenes"""
634
635
                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
636
                # L2A_newObjects  = MAP(HLP_F.merge_GMS_tiles_to_GMS_obj, grouped_L2A_Tiles)
637
                L2A_newObjects = [L2A_P.L2A_object().from_tiles(tileList) for tileList in grouped_L2A_Tiles]
638

639
                """combine newly and earlier processed L2A data"""
640
641
                L2A_DBObjects = self.get_DB_objects('L2B', self.L2A_tiles)
                L2A_Instances = L2A_newObjects + L2A_DBObjects  # combine newly and earlier processed L2A data
642

643
                L2B_resObjects = MAP(L2B_map, L2A_Instances)
644
645

            else:  # tiles
646
                L2A_newTiles = self.L2A_tiles  # tiles have the block size specified in L2A_map_2
647
648

                """combine newly and earlier processed L2A data"""
649
650
651
                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
652

653
                L2B_tiles = MAP(L2B_map, L2A_tiles)
654
655

                grouped_L2B_Tiles = \
656
657
                    HLP_F.group_objects_by_attributes(L2B_tiles,
                                                      'scene_ID')  # group results # FIXME nötig an dieser Stelle?
658
659
                [L2B_tiles_group[0].delete_tempFiles() for L2B_tiles_group in grouped_L2B_Tiles]

660
                L2B_resObjects = [L2B_P.L2B_object().from_tiles(tileList) for tileList in grouped_L2B_Tiles]
661

662
            self.L2B_newObjects = [obj for obj in L2B_resObjects if isinstance(obj, L2B_P.L2B_object)]
663
            self.failed_objects += [obj for obj in L2B_resObjects if isinstance(obj, failed_GMS_object) and
664
665
666
667
668
669
                                    obj.scene_ID not in self.sceneids_failed]

        return self.L2B_newObjects

    def L2C_processing(self):
        """
670
        Run Level 2C processing: accurracy assessment and MGRS tiling
671
        """
672
        # FIXME only parallelization_level == 'scenes' implemented
673
        if self.config.exec_L2CP[0]:
674
            self.logger.info('\n\n########## Level 2C Processing started - calculation of quality layers ###########\n')
675

676
            """combine newly and earlier processed L2A data"""
677
678
            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
679

680
            L2C_resObjects = MAP(L2C_map, L2B_Instances, CPUs=8)  # FIXME 8 workers due to heavy IO
681
            # FIXME in case of inmem_serialization mode results are too big to be back-pickled
682
            self.L2C_newObjects = [obj for obj in L2C_resObjects if isinstance(obj, L2C_P.L2C_object)]
683
            self.failed_objects += [obj for obj in L2C_resObjects if isinstance(obj, failed_GMS_object) and
684
685
686
687
688
689
690
691
                                    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').
        """
692
        # TODO move this method to config.Job
693
694
        # update 'failed_sceneids' column of job record within jobs table
        sceneids_failed = list(set([obj.scene_ID for obj in self.failed_objects]))
695
        DB_T.update_records_in_postgreSQLdb(
696
            self.config.conn_database, 'jobs',
697
            {'failed_sceneids': sceneids_failed,  # update 'failed_sceneids' column
698
699
700
             'finishtime': self.config.end_time,  # add job finish timestamp
             'status': self.config.status},  # update 'job_status' column
            {'id': self.config.ID})
701

702
703
704
705
706
    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
707
        already_updated_IDs = []
708
        for ds in usecase_datalist:
709
710
            if ds['proc_level'] is not None and ds['scene_ID'] not in already_updated_IDs:
                # update statistics column of jobs table
711
                DB_T.increment_decrement_arrayCol_in_postgreSQLdb(
712
                    self.config.conn_database, 'jobs', 'statistics', cond_dict={'id': self.config.ID},
713
714
715
                    idx_val2decrement=db_jobs_statistics_def['downloaded'],
                    idx_val2increment=db_jobs_statistics_def[ds['proc_level']])

716
717
718
                # avoid double updating in case of subsystems belonging to the same scene ID
                already_updated_IDs.append(ds['scene_ID'])

719
720
721
722
    def create_job_summary(self):
        """
        Create job success summary
        """
723
724
725

        # get objects with highest requested processing level
        highest_procL_Objs = []
726
        for pL in reversed(proc_chain):
727
            if getattr(self.config, 'exec_%sP' % pL)[0]:
728
                highest_procL_Objs = getattr(self, '%s_newObjects' % pL) if pL != 'L2A' else self.L2A_tiles
729
730
                break

731
732
733
734
        gms_objects2summarize = highest_procL_Objs + self.failed_objects
        if gms_objects2summarize:
            # create summaries
            detailed_JS, quick_JS = get_job_summary(gms_objects2summarize)
735
736
737
            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())
738
739

            self.summary_detailed = detailed_JS
740
            self.summary_quick = quick_JS
741
742
743
744
745

        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.")
746
747
748
749
750
751

    def clear_lists_procObj(self):
        self.failed_objects = []
        self.L1A_newObjects = []
        self.L1B_newObjects = []
        self.L1C_newObjects = []
752
        self.L2A_tiles = []
753
        self.L2B_newObjects = []
754
755
756
757
758
        self.L2C_newObjects = []


def get_job_summary(list_GMS_objects):
    # get detailed job summary
759
760
    DJS_cols = ['GMS_object', 'scene_ID', 'entity_ID', 'satellite', 'sensor', 'subsystem', 'image_type', 'proc_level',
                'arr_shape', 'arr_pos', 'failedMapper', 'ExceptionType', 'ExceptionValue', 'ExceptionTraceback']
761
762
763
764
    DJS = DataFrame(columns=DJS_cols)
    DJS['GMS_object'] = list_GMS_objects

    for col in DJS_cols[1:]:
765
766
        def get_val(obj): return getattr(obj, col) if hasattr(obj, col) else None
        DJS[col] = list(DJS['GMS_object'].map(get_val))
767
768

    del DJS['GMS_object']
769
    DJS = DJS.sort_values(by=['satellite', 'sensor', 'entity_ID'])
770
771

    # get quick job summary
772
773
774
775
    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
776
    # count objects with the same satellite/sensor/sceneid combination
777
778
    QJS['count'] = [len(DJS[(DJS['satellite'] == sat) & (DJS['sensor'] == sen)]['scene_ID'].unique())
                    for sat, sen in zip(all_sat, all_sen)]
779
    QJS['proc_successfully'] = [len(DJS[(DJS['satellite'] == sat) &
780
781
                                        (DJS['sensor'] == sen) &
                                        (DJS['failedMapper'].isnull())]['scene_ID'].unique())
782
                                for sat, sen in zip(all_sat, all_sen)]
783
    QJS['proc_failed'] = QJS['count'] - QJS['proc_successfully']
784
785
    QJS = QJS.sort_values(by=['satellite', 'sensor'])
    return DJS, QJS