v3bwfile.py 41.3 KB
Newer Older
juga  's avatar
juga committed
1
2
3
4
# -*- coding: utf-8 -*-
"""Classes and functions that create the bandwidth measurements document
(v3bw) used by bandwidth authorities."""

juga  's avatar
juga committed
5
import copy
juga  's avatar
juga committed
6
import logging
7
import math
juga  's avatar
juga committed
8
import os
9
from itertools import combinations
juga  's avatar
juga committed
10
from statistics import median, mean
11
from stem.descriptor import parse_file
12

13
from sbws import __version__
14
from sbws.globals import (SPEC_VERSION, BW_LINE_SIZE, SBWS_SCALE_CONSTANT,
15
16
                          TORFLOW_SCALING, SBWS_SCALING, TORFLOW_BW_MARGIN,
                          TORFLOW_OBS_LAST, TORFLOW_OBS_MEAN,
17
                          PROP276_ROUND_DIG, MIN_REPORT, MAX_BW_DIFF_PERC)
juga  's avatar
juga committed
18
from sbws.lib.resultdump import ResultSuccess, _ResultType
19
from sbws.util.filelock import DirectoryLock
juga  's avatar
juga committed
20
21
from sbws.util.timestamp import (now_isodt_str, unixts_to_isodt_str,
                                 now_unixts)
22
from sbws.util.state import State
juga  's avatar
juga committed
23
24
25

log = logging.getLogger(__name__)

26
LINE_SEP = '\n'
27
28
KEYVALUE_SEP_V1 = '='
KEYVALUE_SEP_V2 = ' '
29
30
# List of the extra KeyValues accepted by the class
EXTRA_ARG_KEYVALUES = ['software', 'software_version', 'file_created',
31
                       'earliest_bandwidth', 'generator_started',
32
                       'scanner_country', 'destinations_countries']
juga  's avatar
juga committed
33
34
35
STATS_KEYVALUES = ['number_eligible_relays', 'minimum_number_eligible_relays',
                   'number_consensus_relays', 'percent_eligible_relays',
                   'minimum_percent_eligible_relays']
36
KEYVALUES_INT = STATS_KEYVALUES
37
# List of all unordered KeyValues currently being used to generate the file
38
39
UNORDERED_KEYVALUES = EXTRA_ARG_KEYVALUES + STATS_KEYVALUES + \
                      ['latest_bandwidth']
40
41
# List of all the KeyValues currently being used to generate the file
ALL_KEYVALUES = ['version'] + UNORDERED_KEYVALUES
42
TERMINATOR = '====='
43
44
# Num header lines in v1.X.X using all the KeyValues
NUM_LINES_HEADER_V1 = len(ALL_KEYVALUES) + 2
45
46
LINE_TERMINATOR = TERMINATOR + LINE_SEP

juga  's avatar
juga committed
47
# KeyValue separator in Bandwidth Lines
48
BW_KEYVALUE_SEP_V1 = ' '
49
50
51
52
53
# not inclding in the files the extra bws for now
BW_KEYVALUES_BASIC = ['node_id', 'bw']
BW_KEYVALUES_FILE = BW_KEYVALUES_BASIC + \
                    ['master_key_ed25519', 'nick', 'rtt', 'time',
                     'success', 'error_stream', 'error_circ', 'error_misc']
54
BW_KEYVALUES_EXTRA_BWS = ['bw_median', 'bw_mean', 'desc_bw_avg', 'desc_bw_bur',
55
56
57
                          'desc_bw_obs_last', 'desc_bw_obs_mean',
                          'consensus_bandwidth',
                          'consensus_bandwidth_is_unmeasured']
58
BW_KEYVALUES_EXTRA = BW_KEYVALUES_FILE + BW_KEYVALUES_EXTRA_BWS
59
BW_KEYVALUES_INT = ['bw', 'rtt', 'success', 'error_stream',
60
61
                    'error_circ', 'error_misc'] + BW_KEYVALUES_EXTRA_BWS
BW_KEYVALUES = BW_KEYVALUES_BASIC + BW_KEYVALUES_EXTRA
juga  's avatar
juga committed
62
63


64
def round_sig_dig(n, digits=PROP276_ROUND_DIG):
65
66
67
68
69
70
71
    """Round n to 'digits' significant digits in front of the decimal point.
       Results less than or equal to 1 are rounded to 1.
       Returns an integer.

       digits must be greater than 0.
       n must be less than or equal to 2**73, to avoid floating point errors.
       """
72
    digits = int(digits)
73
74
75
76
77
78
79
80
81
    assert digits >= 1
    if n <= 1:
        return 1
    digits_in_n = int(math.log10(n)) + 1
    round_digits = max(digits_in_n - digits, 0)
    rounded_n = round(n, -round_digits)
    return int(rounded_n)


82
def kb_round_x_sig_dig(bw_bs, digits=PROP276_ROUND_DIG):
83
84
85
86
87
88
89
90
91
92
93
    """Convert bw_bs from bytes to kilobytes, and round the result to
       'digits' significant digits.
       Results less than or equal to 1 are rounded up to 1.
       Returns an integer.

       digits must be greater than 0.
       n must be less than or equal to 2**82, to avoid floating point errors.
       """
    # avoid double-rounding by using floating-point
    bw_kb = bw_bs / 1000.0
    return round_sig_dig(bw_kb, digits=digits)
94
95


juga  's avatar
juga committed
96
97
98
99
100
101
102
103
104
def num_results_of_type(results, type_str):
    return len([r for r in results if r.type == type_str])


# Better way to use enums?
def result_type_to_key(type_str):
    return type_str.replace('-', '_')


105
class V3BWHeader(object):
juga  's avatar
juga committed
106
107
    """
    Create a bandwidth measurements (V3bw) header
108
    following bandwidth measurements document spec version 1.X.X.
juga  's avatar
juga committed
109

110
    :param str timestamp: timestamp in Unix Epoch seconds of the most recent
111
        generator result.
juga  's avatar
juga committed
112
113
114
    :param str version: the spec version
    :param str software: the name of the software that generates this
    :param str software_version: the version of the software
115
    :param dict kwargs: extra headers. Currently supported:
juga  's avatar
juga committed
116

117
118
119
120
        - earliest_bandwidth: str, ISO 8601 timestamp in UTC time zone
          when the first bandwidth was obtained
        - generator_started: str, ISO 8601 timestamp in UTC time zone
          when the generator started
juga  's avatar
juga committed
121
    """
122
    def __init__(self, timestamp, **kwargs):
123
124
125
126
        assert isinstance(timestamp, str)
        for v in kwargs.values():
            assert isinstance(v, str)
        self.timestamp = timestamp
127
128
129
130
        # KeyValues with default value when not given by kwargs
        self.version = kwargs.get('version', SPEC_VERSION)
        self.software = kwargs.get('software', 'sbws')
        self.software_version = kwargs.get('software_version', __version__)
131
        self.file_created = kwargs.get('file_created', now_isodt_str())
juga  's avatar
juga committed
132
        # latest_bandwidth should not be in kwargs, since it MUST be the
133
        # same as timestamp
juga  's avatar
juga committed
134
        self.latest_bandwidth = unixts_to_isodt_str(timestamp)
135
136
        [setattr(self, k, v) for k, v in kwargs.items()
         if k in EXTRA_ARG_KEYVALUES]
137

138
    def __str__(self):
139
        if self.version.startswith('1.'):
140
141
            return self.strv1
        return self.strv2
142

juga  's avatar
juga committed
143
    @classmethod
144
145
    def from_results(cls, results, scanner_country=None,
                     destinations_countries=None, state_fpath=''):
juga  's avatar
juga committed
146
147
148
        kwargs = dict()
        latest_bandwidth = cls.latest_bandwidth_from_results(results)
        earliest_bandwidth = cls.earliest_bandwidth_from_results(results)
149
        generator_started = cls.generator_started_from_file(state_fpath)
juga  's avatar
juga committed
150
151
152
153
154
        timestamp = str(latest_bandwidth)
        kwargs['latest_bandwidth'] = unixts_to_isodt_str(latest_bandwidth)
        kwargs['earliest_bandwidth'] = unixts_to_isodt_str(earliest_bandwidth)
        if generator_started is not None:
            kwargs['generator_started'] = generator_started
155
156
157
        # To be compatible with older bandwidth files, do not require it.
        if scanner_country is not None:
            kwargs['scanner_country'] = scanner_country
158
159
        if destinations_countries is not None:
            kwargs['destinations_countries'] = destinations_countries
juga  's avatar
juga committed
160
161
162
        h = cls(timestamp, **kwargs)
        return h

163
    @classmethod
164
    def from_lines_v1(cls, lines):
165
166
        """
        :param list lines: list of lines to parse
167
        :returns: tuple of V3BWHeader object and non-header lines
168
169
170
171
        """
        assert isinstance(lines, list)
        try:
            index_terminator = lines.index(TERMINATOR)
Matt Traudt's avatar
Matt Traudt committed
172
        except ValueError:
173
174
175
            # is not a bw file or is v100
            log.warn('Terminator is not in lines')
            return None
176
        ts = lines[0]
177
        kwargs = dict([l.split(KEYVALUE_SEP_V1)
178
                       for l in lines[:index_terminator]
179
                       if l.split(KEYVALUE_SEP_V1)[0] in ALL_KEYVALUES])
180
        h = cls(ts, **kwargs)
juga  's avatar
juga committed
181
        # last line is new line
182
        return h, lines[index_terminator + 1:-1]
183
184

    @classmethod
185
    def from_text_v1(self, text):
186
        """
187
        :param str text: text to parse
188
        :returns: tuple of V3BWHeader object and non-header lines
189
190
        """
        assert isinstance(text, str)
191
        return self.from_lines_v1(text.split(LINE_SEP))
192

juga  's avatar
juga committed
193
194
195
196
197
198
199
200
201
202
203
    @classmethod
    def from_lines_v100(cls, lines):
        """
        :param list lines: list of lines to parse
        :returns: tuple of V3BWHeader object and non-header lines
        """
        assert isinstance(lines, list)
        h = cls(lines[0])
        # last line is new line
        return h, lines[1:-1]

204
    @staticmethod
205
    def generator_started_from_file(state_fpath):
206
207
208
209
        '''
        ISO formatted timestamp for the time when the scanner process most
        recently started.
        '''
210
        state = State(state_fpath)
211
212
213
214
        if 'scanner_started' in state:
            return state['scanner_started']
        else:
            return None
215
216

    @staticmethod
juga  's avatar
juga committed
217
    def latest_bandwidth_from_results(results):
218
        return round(max([r.time for fp in results for r in results[fp]]))
219
220
221

    @staticmethod
    def earliest_bandwidth_from_results(results):
222
        return round(min([r.time for fp in results for r in results[fp]]))
223

juga  's avatar
juga committed
224
225
226
227
228
229
230
231
232
233
234
235
236
237
    @property
    def keyvalue_unordered_tuple_ls(self):
        """Return list of KeyValue tuples that do not have specific order."""
        # sort the list to generate determinist headers
        keyvalue_tuple_ls = sorted([(k, v) for k, v in self.__dict__.items()
                                    if k in UNORDERED_KEYVALUES])
        return keyvalue_tuple_ls

    @property
    def keyvalue_tuple_ls(self):
        """Return list of all KeyValue tuples"""
        return [('version', self.version)] + self.keyvalue_unordered_tuple_ls

    @property
238
239
240
    def keyvalue_v1str_ls(self):
        """Return KeyValue list of strings following spec v1.X.X."""
        keyvalues = [self.timestamp] + [KEYVALUE_SEP_V1.join([k, v])
juga  's avatar
juga committed
241
242
243
244
                                        for k, v in self.keyvalue_tuple_ls]
        return keyvalues

    @property
245
246
247
    def strv1(self):
        """Return header string following spec v1.X.X."""
        header_str = LINE_SEP.join(self.keyvalue_v1str_ls) + LINE_SEP + \
juga  's avatar
juga committed
248
249
250
251
            LINE_TERMINATOR
        return header_str

    @property
252
253
254
    def keyvalue_v2_ls(self):
        """Return KeyValue list of strings following spec v2.X.X."""
        keyvalue = [self.timestamp] + [KEYVALUE_SEP_V2.join([k, v])
juga  's avatar
juga committed
255
256
257
258
                                       for k, v in self.keyvalue_tuple_ls]
        return keyvalue

    @property
259
260
261
    def strv2(self):
        """Return header string following spec v2.X.X."""
        header_str = LINE_SEP.join(self.keyvalue_v2_ls) + LINE_SEP + \
juga  's avatar
juga committed
262
263
264
265
266
267
            LINE_TERMINATOR
        return header_str

    @property
    def num_lines(self):
        return len(self.__str__().split(LINE_SEP))
juga  's avatar
juga committed
268

269
270
271
272
273
    def add_stats(self, **kwargs):
        # Using kwargs because attributes might chage.
        [setattr(self, k, str(v)) for k, v in kwargs.items()
         if k in STATS_KEYVALUES]

juga  's avatar
juga committed
274

juga  's avatar
juga committed
275
class V3BWLine(object):
juga  's avatar
juga committed
276
    """
277
    Create a Bandwidth List line following the spec version 1.X.X.
juga  's avatar
juga committed
278
279
280
281
282
283
284
285
286
287
288
289
290
291

    :param str node_id:
    :param int bw:
    :param dict kwargs: extra headers. Currently supported:

        - nickname, str
        - master_key_ed25519, str
        - rtt, int
        - time, str
        - sucess, int
        - error_stream, int
        - error_circ, int
        - error_misc, int
    """
juga  's avatar
juga committed
292
293
294
    def __init__(self, node_id, bw, **kwargs):
        assert isinstance(node_id, str)
        assert isinstance(bw, int)
295
        assert node_id.startswith('$')
juga  's avatar
juga committed
296
297
298
        self.node_id = node_id
        self.bw = bw
        [setattr(self, k, v) for k, v in kwargs.items()
299
         if k in BW_KEYVALUES_EXTRA]
juga  's avatar
juga committed
300

juga  's avatar
juga committed
301
    def __str__(self):
302
        return self.bw_strv1
juga  's avatar
juga committed
303

juga  's avatar
juga committed
304
    @classmethod
juga  's avatar
juga committed
305
306
    def from_results(cls, results, secs_recent=None, secs_away=None,
                     min_num=0):
307
308
309
        """Convert sbws results to relays' Bandwidth Lines

        ``bs`` stands for Bytes/seconds
310
        ``bw_mean`` means the bw is obtained from the mean of the all the
311
312
313
314
315
        downloads' bandwidth.
        Downloads' bandwidth are calculated as the amount of data received
        divided by the the time it took to received.
        bw = data (Bytes) / time (seconds)
        """
juga  's avatar
juga committed
316
        success_results = [r for r in results if isinstance(r, ResultSuccess)]
317
        # log.debug("Len success_results %s", len(success_results))
juga  's avatar
juga committed
318
319
320
321
322
323
324
        node_id = '$' + results[0].fingerprint
        kwargs = dict()
        kwargs['nick'] = results[0].nickname
        if getattr(results[0], 'master_key_ed25519'):
            kwargs['master_key_ed25519'] = results[0].master_key_ed25519
        kwargs['time'] = cls.last_time_from_results(results)
        kwargs.update(cls.result_types_from_results(results))
325
326
        # useful args for scaling
        if success_results:
327
            results_away = \
328
329
330
                cls.results_away_each_other(success_results, secs_away)
            if not results_away:
                return None
331
332
            # log.debug("Results away from each other: %s",
            #           [unixts_to_isodt_str(r.time) for r in results_away])
333
334
335
            results_recent = cls.results_recent_than(results_away, secs_recent)
            if not results_recent:
                return None
336
337
338
            if not len(results_recent) >= min_num:
                # log.debug('The number of results is less than %s', min_num)
                return None
juga  's avatar
juga committed
339
340
341
            rtt = cls.rtt_from_results(results_recent)
            if rtt:
                kwargs['rtt'] = rtt
342
343
344
            bw = cls.bw_median_from_results(results_recent)
            kwargs['bw_mean'] = cls.bw_mean_from_results(results_recent)
            kwargs['bw_median'] = cls.bw_median_from_results(
345
                results_recent)
346
347
            kwargs['desc_bw_avg'] = \
                cls.desc_bw_avg_from_results(results_recent)
348
349
            kwargs['desc_bw_bur'] = \
                cls.desc_bw_bur_from_results(results_recent)
350
351
352
353
354
            kwargs['consensus_bandwidth'] = \
                cls.consensus_bandwidth_from_results(results_recent)
            kwargs['consensus_bandwidth_is_unmeasured'] = \
                cls.consensus_bandwidth_is_unmeasured_from_results(
                    results_recent)
355
356
357
358
            kwargs['desc_bw_obs_last'] = \
                cls.desc_bw_obs_last_from_results(results_recent)
            kwargs['desc_bw_obs_mean'] = \
                cls.desc_bw_obs_mean_from_results(results_recent)
359
360
361
            bwl = cls(node_id, bw, **kwargs)
            return bwl
        return None
juga  's avatar
juga committed
362
363
364
365
366
367

    @classmethod
    def from_data(cls, data, fingerprint):
        assert fingerprint in data
        return cls.from_results(data[fingerprint])

juga  's avatar
juga committed
368
    @classmethod
369
    def from_bw_line_v1(cls, line):
juga  's avatar
juga committed
370
        assert isinstance(line, str)
371
372
373
        kwargs = dict([kv.split(KEYVALUE_SEP_V1)
                       for kv in line.split(BW_KEYVALUE_SEP_V1)
                       if kv.split(KEYVALUE_SEP_V1)[0] in BW_KEYVALUES])
juga  's avatar
juga committed
374
375
376
        for k, v in kwargs.items():
            if k in BW_KEYVALUES_INT:
                kwargs[k] = int(v)
377
378
379
380
381
        node_id = kwargs['node_id']
        bw = kwargs['bw']
        del kwargs['node_id']
        del kwargs['bw']
        bw_line = cls(node_id, bw, **kwargs)
juga  's avatar
juga committed
382
383
        return bw_line

juga  's avatar
juga committed
384
385
    @staticmethod
    def results_away_each_other(results, secs_away=None):
juga  's avatar
juga committed
386
387
        # log.debug("Checking whether results are away from each other in %s "
        #           "secs.", secs_away)
juga  's avatar
juga committed
388
389
        if secs_away is None or len(results) < 2:
            return results
390
391
392
393
394
395
        for a, b in combinations(results, 2):
            if abs(a.time - b.time) > secs_away:
                return results
        # log.debug("Results are NOT away from each other in at least %ss: %s",
        #           secs_away, [unixts_to_isodt_str(r.time) for r in results])
        return None
juga  's avatar
juga committed
396
397
398
399
400

    @staticmethod
    def results_recent_than(results, secs_recent=None):
        if secs_recent is None:
            return results
401
        results_recent = list(filter(
juga  's avatar
juga committed
402
                            lambda x: (now_unixts() - x.time) < secs_recent,
403
404
405
406
407
408
                            results))
        # if not results_recent:
        #     log.debug("Results are NOT more recent than %ss: %s",
        #               secs_recent,
        #               [unixts_to_isodt_str(r.time) for r in results])
        return results_recent
juga  's avatar
juga committed
409

410
    @staticmethod
411
    def bw_median_from_results(results):
412
413
414
415
        return max(round(median([dl['amount'] / dl['duration']
                                 for r in results for dl in r.downloads])), 1)

    @staticmethod
416
    def bw_mean_from_results(results):
417
418
419
        return max(round(mean([dl['amount'] / dl['duration']
                               for r in results for dl in r.downloads])), 1)

juga  's avatar
juga committed
420
421
422
423
424
425
426
427
    @staticmethod
    def last_time_from_results(results):
        return unixts_to_isodt_str(round(max([r.time for r in results])))

    @staticmethod
    def rtt_from_results(results):
        # convert from miliseconds to seconds
        rtts = [(round(rtt * 1000)) for r in results for rtt in r.rtts]
juga  's avatar
juga committed
428
        rtt = round(median(rtts)) if rtts else None
juga  's avatar
juga committed
429
430
431
432
433
434
435
436
437
        return rtt

    @staticmethod
    def result_types_from_results(results):
        rt_dict = dict([(result_type_to_key(rt.value),
                         num_results_of_type(results, rt.value))
                        for rt in _ResultType])
        return rt_dict

438
439
440
441
442
443
444
445
    @staticmethod
    def desc_bw_avg_from_results(results):
        """Obtain the last descriptor bandwidth average from the results."""
        for r in reversed(results):
            if r.relay_average_bandwidth is not None:
                return r.relay_average_bandwidth
        return None

446
447
448
449
450
451
452
453
    @staticmethod
    def desc_bw_bur_from_results(results):
        """Obtain the last descriptor bandwidth burst from the results."""
        for r in reversed(results):
            if r.relay_burst_bandwidth is not None:
                return r.relay_burst_bandwidth
        return None

454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
    @staticmethod
    def consensus_bandwidth_from_results(results):
        """Obtain the last consensus bandwidth from the results."""
        for r in reversed(results):
            if r.consensus_bandwidth is not None:
                return r.consensus_bandwidth
        return None

    @staticmethod
    def consensus_bandwidth_is_unmeasured_from_results(results):
        """Obtain the last consensus unmeasured flag from the results."""
        for r in reversed(results):
            if r.consensus_bandwidth_is_unmeasured is not None:
                return r.consensus_bandwidth_is_unmeasured
        return None

470
    @staticmethod
471
472
    def desc_bw_obs_mean_from_results(results):
        desc_bw_obs_ls = []
473
474
        for r in results:
            if r.relay_observed_bandwidth is not None:
475
476
477
                desc_bw_obs_ls.append(r.relay_observed_bandwidth)
        if desc_bw_obs_ls:
            return max(round(mean(desc_bw_obs_ls)), 1)
478
479
480
        return None

    @staticmethod
481
    def desc_bw_obs_last_from_results(results):
482
483
484
485
486
487
        # the last is at the end of the list
        for r in reversed(results):
            if r.relay_observed_bandwidth is not None:
                return r.relay_observed_bandwidth
        return None

juga  's avatar
juga committed
488
489
490
491
492
493
494
    @property
    def bw_keyvalue_tuple_ls(self):
        """Return list of KeyValue Bandwidth Line tuples."""
        # sort the list to generate determinist headers
        keyvalue_tuple_ls = sorted([(k, v) for k, v in self.__dict__.items()
                                    if k in BW_KEYVALUES])
        return keyvalue_tuple_ls
juga  's avatar
juga committed
495

juga  's avatar
juga committed
496
    @property
497
    def bw_keyvalue_v1str_ls(self):
juga  's avatar
juga committed
498
        """Return list of KeyValue Bandwidth Line strings following
499
        spec v1.X.X.
juga  's avatar
juga committed
500
        """
501
        bw_keyvalue_str = [KEYVALUE_SEP_V1 .join([k, str(v)])
juga  's avatar
juga committed
502
503
                           for k, v in self.bw_keyvalue_tuple_ls]
        return bw_keyvalue_str
juga  's avatar
juga committed
504

juga  's avatar
juga committed
505
    @property
506
507
508
509
    def bw_strv1(self):
        """Return Bandwidth Line string following spec v1.X.X."""
        bw_line_str = BW_KEYVALUE_SEP_V1.join(
                        self.bw_keyvalue_v1str_ls) + LINE_SEP
juga  's avatar
juga committed
510
511
512
513
514
515
        if len(bw_line_str) > BW_LINE_SIZE:
            # if this is the case, probably there are too many KeyValues,
            # or the limit needs to be changed in Tor
            log.warn("The bandwidth line %s is longer than %s",
                     len(bw_line_str), BW_LINE_SIZE)
        return bw_line_str
juga  's avatar
juga committed
516

juga  's avatar
juga committed
517

518
class V3BWFile(object):
juga  's avatar
juga committed
519
    """
520
    Create a Bandwidth List file following spec version 1.X.X
juga  's avatar
juga committed
521
522
523
524

    :param V3BWHeader v3bwheader: header
    :param list v3bwlines: V3BWLines
    """
juga  's avatar
juga committed
525
526
527
528
529
    def __init__(self, v3bwheader, v3bwlines):
        self.header = v3bwheader
        self.bw_lines = v3bwlines

    def __str__(self):
530
        return str(self.header) + ''.join([str(bw_line) or ''
juga  's avatar
juga committed
531
532
                                           for bw_line in self.bw_lines])

juga  's avatar
juga committed
533
    @classmethod
534
535
    def from_results(cls, results, scanner_country=None,
                     destinations_countries=None, state_fpath='',
536
                     scale_constant=SBWS_SCALE_CONSTANT,
537
538
                     scaling_method=TORFLOW_SCALING,
                     torflow_obs=TORFLOW_OBS_LAST,
juga  's avatar
juga committed
539
                     torflow_cap=TORFLOW_BW_MARGIN,
540
                     round_digs=PROP276_ROUND_DIG,
541
                     secs_recent=None, secs_away=None, min_num=0,
542
543
                     consensus_path=None, max_bw_diff_perc=MAX_BW_DIFF_PERC,
                     reverse=False):
544
545
546
547
548
549
        """Create V3BWFile class from sbws Results.

        :param dict results: see below
        :param str state_fpath: path to the state file
        :param int scaling_method:
            Scaling method to obtain the bandwidth
550
            Possible values: {None, SBWS_SCALING, TORFLOW_SCALING} = {0, 1, 2}
551
552
553
554
555
556
557
558
559
560
561
        :param int scale_constant: sbws scaling constant
        :param int torflow_obs: method to choose descriptor observed bandwidth
        :param bool reverse: whether to sort the bw lines descending or not

        Results are in the form::

            {'relay_fp1': [Result1, Result2, ...],
             'relay_fp2': [Result1, Result2, ...]}

        """
        log.info('Processing results to generate a bandwidth list file.')
562
563
        header = V3BWHeader.from_results(results, scanner_country,
                                         destinations_countries, state_fpath)
564
        bw_lines_raw = []
juga  's avatar
juga committed
565
566
        number_consensus_relays = cls.read_number_consensus_relays(
            consensus_path)
567
        state = State(state_fpath)
juga  's avatar
juga committed
568
        for fp, values in results.items():
569
            # log.debug("Relay fp %s", fp)
juga  's avatar
juga committed
570
571
            line = V3BWLine.from_results(values, secs_recent, secs_away,
                                         min_num)
juga  's avatar
juga committed
572
573
            if line is not None:
                bw_lines_raw.append(line)
574
        if not bw_lines_raw:
575
576
            log.info("After applying restrictions to the raw results, "
                     "there is not any. Scaling can not be applied.")
juga  's avatar
juga committed
577
578
            cls.update_progress(
                cls, bw_lines_raw, header, number_consensus_relays, state)
579
580
581
            return cls(header, [])
        if scaling_method == SBWS_SCALING:
            bw_lines = cls.bw_sbws_scale(bw_lines_raw, scale_constant)
582
            cls.warn_if_not_accurate_enough(bw_lines, scale_constant)
583
584
            # log.debug(bw_lines[-1])
        elif scaling_method == TORFLOW_SCALING:
juga  's avatar
juga committed
585
            bw_lines = cls.bw_torflow_scale(bw_lines_raw, torflow_obs,
586
                                            torflow_cap, round_digs)
587
            # log.debug(bw_lines[-1])
juga  's avatar
juga committed
588
589
            cls.update_progress(
                cls, bw_lines, header, number_consensus_relays, state)
juga  's avatar
juga committed
590
        else:
591
592
            bw_lines = cls.bw_kb(bw_lines_raw)
            # log.debug(bw_lines[-1])
593
594
        # Not using the result for now, just warning
        cls.is_max_bw_diff_perc_reached(bw_lines, max_bw_diff_perc)
juga  's avatar
juga committed
595
596
597
        f = cls(header, bw_lines)
        return f

598
    @classmethod
599
    def from_v1_fpath(cls, fpath):
600
601
602
603
        log.info('Parsing bandwidth file %s', fpath)
        with open(fpath) as fd:
            text = fd.read()
        all_lines = text.split(LINE_SEP)
604
605
        header, lines = V3BWHeader.from_lines_v1(all_lines)
        bw_lines = [V3BWLine.from_bw_line_v1(line) for line in lines]
606
607
        return cls(header, bw_lines)

608
609
610
611
612
613
614
    @classmethod
    def from_v100_fpath(cls, fpath):
        log.info('Parsing bandwidth file %s', fpath)
        with open(fpath) as fd:
            text = fd.read()
        all_lines = text.split(LINE_SEP)
        header, lines = V3BWHeader.from_lines_v100(all_lines)
615
        bw_lines = sorted([V3BWLine.from_bw_line_v1(l) for l in lines],
616
617
618
                          key=lambda l: l.bw)
        return cls(header, bw_lines)

juga  's avatar
juga committed
619
620
621
622
623
624
625
    @staticmethod
    def bw_kb(bw_lines, reverse=False):
        bw_lines_scaled = copy.deepcopy(bw_lines)
        for l in bw_lines_scaled:
            l.bw = max(round(l.bw / 1000), 1)
        return sorted(bw_lines_scaled, key=lambda x: x.bw, reverse=reverse)

juga  's avatar
juga committed
626
    @staticmethod
627
    def bw_sbws_scale(bw_lines, scale_constant=SBWS_SCALE_CONSTANT,
juga  's avatar
juga committed
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
                      reverse=False):
        """Return a new V3BwLine list scaled using sbws method.

        :param list bw_lines:
            bw lines to scale, not self.bw_lines,
            since this method will be before self.bw_lines have been
            initialized.
        :param int scale_constant:
            the constant to multiply by the ratio and
            the bandwidth to obtain the new bandwidth
        :returns list: V3BwLine list
        """
        log.debug('Scaling bandwidth using sbws method.')
        m = median([l.bw for l in bw_lines])
        bw_lines_scaled = copy.deepcopy(bw_lines)
        for l in bw_lines_scaled:
            # min is to limit the bw to descriptor average-bandwidth
            # max to avoid bandwidth with 0 value
646
            l.bw = max(round(min(l.desc_bw_avg,
647
                                 l.bw * scale_constant / m)
juga  's avatar
juga committed
648
649
650
651
652
                             / 1000), 1)
        return sorted(bw_lines_scaled, key=lambda x: x.bw, reverse=reverse)

    @staticmethod
    def warn_if_not_accurate_enough(bw_lines,
653
                                    scale_constant=SBWS_SCALE_CONSTANT):
juga  's avatar
juga committed
654
655
656
657
658
659
660
661
        margin = 0.001
        accuracy_ratio = median([l.bw for l in bw_lines]) / scale_constant
        log.info('The generated lines are within {:.5}% of what they should '
                 'be'.format((1 - accuracy_ratio) * 100))
        if accuracy_ratio < 1 - margin or accuracy_ratio > 1 + margin:
            log.warning('There was %f%% error and only +/- %f%% is '
                        'allowed', (1 - accuracy_ratio) * 100, margin * 100)

662
    @staticmethod
663
664
665
    def is_max_bw_diff_perc_reached(bw_lines,
                                    max_bw_diff_perc=MAX_BW_DIFF_PERC):
        sum_consensus_bw = sum([l.desc_bw_obs_last for l in bw_lines])
666
667
668
669
670
671
672
673
674
675
676
        sum_bw = sum([l.bw for l in bw_lines])
        diff = min(sum_consensus_bw, sum_bw) / max(sum_consensus_bw, sum_bw)
        diff_perc = diff * 100
        log.info("The difference between the total consensus bandwidth "
                 "and the total measured bandwidth is %s%% percent",
                 diff_perc)
        if diff_perc > MAX_BW_DIFF_PERC:
            log.warning("It is more than %s%%", max_bw_diff_perc)
            return True
        return False

677
    @staticmethod
678
    def bw_torflow_scale(bw_lines, desc_bw_obs_type=TORFLOW_OBS_MEAN,
679
                         cap=TORFLOW_BW_MARGIN,
680
                         num_round_dig=PROP276_ROUND_DIG, reverse=False):
juga  's avatar
juga committed
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
        """
        Obtain final bandwidth measurements applying Torflow's scaling
        method.

        From Torflow's README.spec.txt (section 2.2)::

            In this way, the resulting network status consensus bandwidth values  # NOQA
            are effectively re-weighted proportional to how much faster the node  # NOQA
            was as compared to the rest of the network.

        The variables and steps used in Torflow:

        **strm_bw**::

            The strm_bw field is the average (mean) of all the streams for the relay  # NOQA
            identified by the fingerprint field.
            strm_bw = sum(bw stream x)/|n stream|

        **filt_bw**::

            The filt_bw field is computed similarly, but only the streams equal to  # NOQA
            or greater than the strm_bw are counted in order to filter very slow  # NOQA
            streams due to slow node pairings.

        **filt_sbw and strm_sbw**::

            for rs in RouterStats.query.filter(stats_clause).\
                  options(eagerload_all('router.streams.circuit.routers')).all():  # NOQA
              tot_sbw = 0
              sbw_cnt = 0
              for s in rs.router.streams:
                if isinstance(s, ClosedStream):
                  skip = False
                  #for br in badrouters:
                  #  if br != rs:
                  #    if br.router in s.circuit.routers:
                  #      skip = True
                  if not skip:
                    # Throw out outliers < mean
                    # (too much variance for stddev to filter much)
                    if rs.strm_closed == 1 or s.bandwidth() >= rs.sbw:
                      tot_sbw += s.bandwidth()
                      sbw_cnt += 1

            if sbw_cnt: rs.filt_sbw = tot_sbw/sbw_cnt
            else: rs.filt_sbw = None

        **filt_avg, and strm_avg**::

            Once we have determined the most recent measurements for each node, we  # NOQA
            compute an average of the filt_bw fields over all nodes we have measured.  # NOQA

        ::

            filt_avg = sum(map(lambda n: n.filt_bw, nodes.itervalues()))/float(len(nodes))  # NOQA
            strm_avg = sum(map(lambda n: n.strm_bw, nodes.itervalues()))/float(len(nodes))  # NOQA

        **true_filt_avg and true_strm_avg**::

            for cl in ["Guard+Exit", "Guard", "Exit", "Middle"]:
                true_filt_avg[cl] = filt_avg
                true_strm_avg[cl] = strm_avg

        In the non-pid case, all types of nodes get the same avg

        **n.fbw_ratio and n.fsw_ratio**::

            for n in nodes.itervalues():
                n.fbw_ratio = n.filt_bw/true_filt_avg[n.node_class()]
                n.sbw_ratio = n.strm_bw/true_strm_avg[n.node_class()]

        **n.ratio**::

            These averages are used to produce ratios for each node by dividing the  # NOQA
            measured value for that node by the network average.

        ::

            # Choose the larger between sbw and fbw
              if n.sbw_ratio > n.fbw_ratio:
                n.ratio = n.sbw_ratio
              else:
                n.ratio = n.fbw_ratio

        **desc_bw**:

juga  's avatar
juga committed
767
768
769
770
        It is the minimum of all the descriptor bandwidth values::

            bws = map(int, g)
            bw_observed = min(bws)
juga  's avatar
juga committed
771
772
773
774

            return Router(ns.idhex, ns.nickname, bw_observed, dead, exitpolicy,
            ns.flags, ip, version, os, uptime, published, contact, rate_limited,  # NOQA
            ns.orhash, ns.bandwidth, extra_info_digest, ns.unmeasured)
juga  's avatar
juga committed
775

juga  's avatar
juga committed
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
            self.desc_bw = max(bw,1) # Avoid div by 0

        **new_bw**::

            These ratios are then multiplied by the most recent observed descriptor  # NOQA
            bandwidth we have available for each node, to produce a new value for  # NOQA
            the network status consensus process.

        ::

            n.new_bw = n.desc_bw*n.ratio

        The descriptor observed bandwidth is multiplied by the ratio.

        **Limit the bandwidth to a maximum**::

            NODE_CAP = 0.05

        ::

            if n.new_bw > tot_net_bw*NODE_CAP:
              plog("INFO", "Clipping extremely fast "+n.node_class()+" node "+n.idhex+"="+n.nick+  # NOQA
                   " at "+str(100*NODE_CAP)+"% of network capacity ("+
                   str(n.new_bw)+"->"+str(int(tot_net_bw*NODE_CAP))+") "+
                   " pid_error="+str(n.pid_error)+
                   " pid_error_sum="+str(n.pid_error_sum))
              n.new_bw = int(tot_net_bw*NODE_CAP)

        However, tot_net_bw does not seems to be updated when not using pid.
        This clipping would make faster relays to all have the same value.

        All of that can be expressed as:

        .. math::
810

811
812
813
814
           bwn_i =& min\\left(bwnew_i,
                      \\sum_{i=1}^{n}bwnew_i \\times 0.05\\right) \\

                 &= min\\left(
juga  's avatar
juga committed
815
816
                      \\left(min\\left(bwobs_i, bwavg_i, bwbur_i \\right) \\times r_i\\right),
                        \\sum_{i=1}^{n}\\left(min\\left(bwobs_i, bwavg_i, bwbur_i \\right) \\times r_i\\right)
817
818
819
                        \\times 0.05\\right)\\

                 &= min\\left(
juga  's avatar
juga committed
820
821
                      \\left(min\\left(bwobs_i, bwavg_i, bwbur_i \\right) \\times max\\left(rf_i, rs_i\\right)\\right),
                        \\sum_{i=1}^{n}\\left(min\\left(bwobs_i, bwavg_i, bwbur_i \\right) \\times
822
823
824
                          max\\left(rf_i, rs_i\\right)\\right) \\times 0.05\\right)\\

                 &= min\\left(
juga  's avatar
juga committed
825
                      \\left(min\\left(bwobs_i, bwavg_i, bwbur_i \\right) \\times max\\left(\\frac{bwfilt_i}{bwfilt},
826
                          \\frac{bw_i}{bwstrm}\\right)\\right),
juga  's avatar
juga committed
827
                        \\sum_{i=1}^{n}\\left(min\\left(bwobs_i, bwavg_i, bwbur_i \\right) \\times
828
829
                          max\\left(\\frac{bwfilt_i}{bwfilt},
                            \\frac{bw_i}{bwstrm}\\right)\\right) \\times 0.05\\right)
juga  's avatar
juga committed
830
831
832
833
834

        """
        log.info("Calculating relays' bandwidth using Torflow method.")
        bw_lines_tf = copy.deepcopy(bw_lines)
        # mean (Torflow's strm_avg)
835
        mu = mean([l.bw_mean for l in bw_lines])
juga  's avatar
juga committed
836
        # filtered mean (Torflow's filt_avg)
837
        muf = mean([max(l.bw_mean, mu) for l in bw_lines])
juga  's avatar
juga committed
838
        # bw sum (Torflow's tot_net_bw or tot_sbw)
839
        sum_bw = sum([l.bw_mean for l in bw_lines])
juga  's avatar
juga committed
840
841
        # Torflow's clipping
        hlimit = sum_bw * TORFLOW_BW_MARGIN
juga  's avatar
juga committed
842
843
844
        log.debug('sum %s', sum_bw)
        log.debug('mu %s', mu)
        log.debug('muf %s', muf)
juga  's avatar
juga committed
845
        log.debug('hlimit %s', hlimit)
juga  's avatar
juga committed
846
        for l in bw_lines_tf:
847
848
849
850
            if desc_bw_obs_type == TORFLOW_OBS_LAST:
                desc_bw_obs = l.desc_bw_obs_last
            elif desc_bw_obs_type == TORFLOW_OBS_MEAN:
                desc_bw_obs = l.desc_bw_obs_mean
851
852
853
854
855
856
857
858
859
            # Excerpt from bandwidth-file-spec.txt section 2.3
            # A relay's MaxAdvertisedBandwidth limits the bandwidth-avg in its
            # descriptor.
            # Therefore generators MUST limit a relay's measured bandwidth to
            # its descriptor's bandwidth-avg.
            # Generators SHOULD NOT limit measured bandwidths based on
            # descriptors' bandwidth-observed, because that penalises new
            # relays.
            # See https://trac.torproject.org/projects/tor/ticket/8494
860
861
862
863
864
865
866
867
868
            if l.desc_bw_bur is not None:
                # Because in previous versions results were not storing
                # desc_bw_bur
                desc_bw = min(desc_bw_obs, l.desc_bw_bur, l.desc_bw_avg)
            else:
                desc_bw = min(desc_bw_obs, l.desc_bw_avg)
            # In previous versions results were not storing consensus_bandwidth
            if l.consensus_bandwidth_is_unmeasured \
                    or l.consensus_bandwidth is None:
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
                min_bandwidth = desc_bw
            # If the relay is measured, use the minimum between the descriptors
            # bandwidth and the consensus bandwidth, so that
            # MaxAdvertisedBandwidth limits the consensus weight
            # The consensus bandwidth in a measured relay has been obtained
            # doing the same calculation as here
            else:
                min_bandwidth = min(desc_bw, l.consensus_bandwidth)
            # Torflow's scaling
            ratio_stream = l.bw_mean / mu
            ratio_stream_filtered = max(l.bw_mean, mu) / muf
            ratio = max(ratio_stream, ratio_stream_filtered)
            bw_scaled = ratio * min_bandwidth
            # round and convert to KB
            bw_new = kb_round_x_sig_dig(bw_scaled, digits=num_round_dig)
juga  's avatar
juga committed
884
885
886
            # Cap maximum bw
            if cap is not None:
                bw_new = min(hlimit, bw_new)
887
888
            # avoid 0
            l.bw = max(bw_new, 1)
juga  's avatar
juga committed
889
        return sorted(bw_lines_tf, key=lambda x: x.bw, reverse=reverse)
890

891
    @staticmethod
juga  's avatar
juga committed
892
    def read_number_consensus_relays(consensus_path):
893
894
895
896
897
        """Read the number of relays in the Network from the cached consensus
        file."""
        num = None
        try:
            num = len(list(parse_file(consensus_path)))
898
        except (FileNotFoundError, AttributeError):
899
900
901
902
903
904
            log.info("It is not possible to obtain statistics about the "
                     "percentage of measured relays because the cached "
                     "consensus file is not found.")
        log.debug("Number of relays in the network %s", num)
        return num

905
    @staticmethod
juga  's avatar
juga committed
906
    def measured_progress_stats(bw_lines, number_consensus_relays,
907
                                min_perc_reached_before):
908
909
910
911
912
        """ Statistics about measurements progress,
        to be included in the header.

        :param list bw_lines: the bw_lines after scaling and applying filters.
        :param str consensus_path: the path to the cached consensus file.
913
        :param str state_fpath: the path to the state file
914
915
916
917
918
919
920
921
922
        :returns dict, bool: Statistics about the progress made with
            measurements and whether the percentage of measured relays has been
            reached.

        """
        # cached-consensus should be updated every time that scanner get the
        # network status or descriptors?
        # It will not be updated to the last consensus, but the list of
        # measured relays is not either.
juga  's avatar
juga committed
923
        assert isinstance(number_consensus_relays, int)
924
925
        assert isinstance(bw_lines, list)
        statsd = {}
juga  's avatar
juga committed
926
927
928
929
930
931
932
933
934
        statsd['number_eligible_relays'] = len(bw_lines)
        statsd['number_consensus_relays'] = number_consensus_relays
        statsd['minimum_number_eligible_relays'] = round(
            statsd['number_consensus_relays'] * MIN_REPORT / 100)
        statsd['percent_eligible_relays'] = round(
            len(bw_lines) * 100 / statsd['number_consensus_relays'])
        statsd['minimum_percent_eligible_relays'] = MIN_REPORT
        if statsd['number_eligible_relays'] < \
                statsd['minimum_number_eligible_relays']:
935
            # if min percent was was reached before, warn
936
937
938
939
            # otherwise, debug
            if min_perc_reached_before is not None:
                log.warning('The percentage of the measured relays is less '
                            'than the %s%% of the relays in the network (%s).',
juga  's avatar
juga committed
940
                            MIN_REPORT, statsd['number_consensus_relays'])
941
942
943
            else:
                log.info('The percentage of the measured relays is less '
                         'than the %s%% of the relays in the network (%s).',
juga  's avatar
juga committed
944
                         MIN_REPORT, statsd['number_consensus_relays'])
945
946
947
948
949
            return statsd, False
        return statsd, True

    @property
    def is_min_perc(self):
juga  's avatar
juga committed
950
951
        if getattr(self.header, 'number_eligible_relays', 0) \
                < getattr(self.header, 'minimum_number_eligible_relays', 0):
952
953
954
            return False
        return True

juga  's avatar
juga committed
955
    @property
juga  's avatar
juga committed
956
957
    def sum_bw(self):
        return sum([l.bw for l in self.bw_lines])
juga  's avatar
juga committed
958
959

    @property
juga  's avatar
juga committed
960
    def num(self):
juga  's avatar
juga committed
961
962
963
        return len(self.bw_lines)

    @property
juga  's avatar
juga committed
964
965
966
967
968
969
    def mean_bw(self):
        return mean([l.bw for l in self.bw_lines])

    @property
    def median_bw(self):
        return median([l.bw for l in self.bw_lines])
juga  's avatar
juga committed
970

juga  's avatar
juga committed
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
    @property
    def max_bw(self):
        return max([l.bw for l in self.bw_lines])

    @property
    def min_bw(self):
        return min([l.bw for l in self.bw_lines])

    @property
    def info_stats(self):
        if not self.bw_lines:
            return
        [log.info(': '.join([attr, str(getattr(self, attr))])) for attr in
         ['sum_bw', 'mean_bw', 'median_bw', 'num',
          'max_bw', 'min_bw']]

juga  's avatar
juga committed
987
988
    def update_progress(self, bw_lines, header, number_consensus_relays,
                        state):
989
        min_perc_reached_before = state.get('min_perc_reached')
juga  's avatar
juga committed
990
        if number_consensus_relays is not None:
991
            statsd, success = self.measured_progress_stats(
juga  's avatar
juga committed
992
                bw_lines, number_consensus_relays, min_perc_reached_before)
993
994
            # add statistics about progress always
            header.add_stats(**statsd)
995
996
997
998
999
1000
1001
            if not success:
                bw_lines = []
                state['min_perc_reached'] = None
            else:
                state['min_perc_reached'] = now_isodt_str()
        return bw_lines

juga  's avatar
juga committed
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
    def bw_line_for_node_id(self, node_id):
        """Returns the bandwidth line for a given node fingerprint.

        Used to combine data when plotting.
        """
        bwl = [l for l in self.bw_lines if l.node_id == node_id]
        if bwl:
            return bwl[0]
        return None

1012
1013
1014
1015
1016
1017
1018
1019
1020
    def to_plt(self, attrs=['bw'], sorted_by=None):
        """Return bandwidth data in a format useful for matplotlib.

        Used from external tool to plot.
        """
        x = [i for i in range(0, self.num)]
        ys = [[getattr(l, k) for l in self.bw_lines] for k in attrs]
        return x, ys, attrs

juga  's avatar
juga committed
1021
1022
1023
1024
1025
    def write(self, output):
        if output == '/dev/stdout':
            log.info("Writing to stdout is not supported.")
            return
        log.info('Writing v3bw file to %s', output)
1026
1027
        # To avoid inconsistent reads, the bandwidth data is written to an
        # archive path, then atomically symlinked to 'latest.v3bw'
juga  's avatar
juga committed
1028
1029
        out_dir = os.path.dirname(output)
        out_link = os.path.join(out_dir, 'latest.v3bw')
1030
        out_link_tmp = out_link + '.tmp'
juga  's avatar
juga committed
1031
1032
1033
1034
1035
1036
        with DirectoryLock(out_dir):
            with open(output, 'wt') as fd:
                fd.write(str(self.header))
                for line in self.bw_lines:
                    fd.write(str(line))
            output_basename = os.path.basename(output)
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
            # To atomically symlink a file, we need to create a temporary link,
            # then rename it to the final link name. (POSIX guarantees that
            # rename is atomic.)
            log.debug('Creating symlink {} -> {}.'
                      .format(out_link_tmp, output_basename))
            os.symlink(output_basename, out_link_tmp)
            log.debug('Renaming symlink {} -> {} to {} -> {}.'
                      .format(out_link_tmp, output_basename,
                              out_link, output_basename))
            os.rename(out_link_tmp, out_link)