3 # module to parse fio histogram log files, not using pandas
4 # runs in python v2 or v3
5 # to get help with the CLI: $ python fio-histo-log-pctiles.py -h
6 # this can be run standalone as a script but is callable
7 # assumes all threads run for same time duration
8 # assumes all threads are doing the same thing for the entire run
16 # separate read and write stats for randrw mixed workload
17 # report average latency if needed
18 # prove that it works (partially done with unit tests)
20 # to run unit tests, set UNITTEST environment variable to anything
21 # if you do this, don't pass normal CLI parameters to it
22 # otherwise it runs the CLI
24 import sys, os, math, copy
25 from copy import deepcopy
32 class FioHistoLogExc(Exception):
35 # if there is an error, print message, and exit with error status
38 print('ERROR: ' + msg)
41 # convert histogram log file into a list of
42 # (time_ms, direction, bsz, buckets) tuples where
43 # - time_ms is the time in msec at which the log record was written
44 # - direction is 0 (read) or 1 (write)
45 # - bsz is block size (not used)
46 # - buckets is a CSV list of counters that make up the histogram
47 # caller decides if the expected number of counters are present
50 def exception_suffix( record_num, pathname ):
51 return 'in histogram record %d file %s' % (record_num+1, pathname)
53 # log file parser raises FioHistoLogExc exceptions
54 # it returns histogram buckets in whatever unit fio uses
56 def parse_hist_file(logfn, buckets_per_interval):
57 max_timestamp_ms = 0.0
59 with open(logfn, 'r') as f:
60 records = [ l.strip() for l in f.readlines() ]
62 for k, r in enumerate(records):
67 int_tokens = [ int(t) for t in tokens ]
68 except ValueError as e:
69 raise FioHistoLogExc('non-integer value %s' % exception_suffix(k+1, logfn))
71 neg_ints = list(filter( lambda tk : tk < 0, int_tokens ))
73 raise FioHistoLogExc('negative integer value %s' % exception_suffix(k+1, logfn))
75 if len(int_tokens) < 3:
76 raise FioHistoLogExc('too few numbers %s' % exception_suffix(k+1, logfn))
78 time_ms = int_tokens[0]
79 if time_ms > max_timestamp_ms:
80 max_timestamp_ms = time_ms
82 direction = int_tokens[1]
83 if direction != 0 and direction != 1:
84 raise FioHistoLogExc('invalid I/O direction %s' % exception_suffix(k+1, logfn))
88 raise FioHistoLogExc('block size too large %s' % exception_suffix(k+1, logfn))
90 buckets = int_tokens[3:]
91 if len(buckets) != buckets_per_interval:
92 raise FioHistoLogExc('%d buckets per interval but %d expected in %s' %
93 (len(buckets), buckets_per_interval, exception_suffix(k+1, logfn)))
94 intervals.append((time_ms, direction, bsz, buckets))
95 if len(intervals) == 0:
96 raise FioHistoLogExc('no records in %s' % logfn)
97 return (intervals, max_timestamp_ms)
100 # compute time range for each bucket index in histogram record
101 # see comments in https://github.com/axboe/fio/blob/master/stat.h
102 # for description of bucket groups and buckets
103 # fio v3 bucket ranges are in nanosec (since response times are measured in nanosec)
104 # but we convert fio v3 nanosecs to floating-point microseconds
106 def time_ranges(groups, counters_per_group, fio_version=3):
109 bucket_intervals = []
110 for g in range(0, groups):
111 for b in range(0, counters_per_group):
112 rmin = float(bucket_base)
113 rmax = rmin + bucket_width
115 rmin /= nsec_per_usec
116 rmax /= nsec_per_usec
117 bucket_intervals.append( [rmin, rmax] )
118 bucket_base += bucket_width
121 return bucket_intervals
124 # compute number of time quantum intervals in the test
126 def get_time_intervals(time_quantum, max_timestamp_ms):
127 # round down to nearest second
128 max_timestamp = max_timestamp_ms // msec_per_sec
129 # round up to nearest whole multiple of time_quantum
130 time_interval_count = (max_timestamp + time_quantum) // time_quantum
131 end_time = time_interval_count * time_quantum
132 return (end_time, time_interval_count)
134 # align raw histogram log data to time quantum so
135 # we can then combine histograms from different threads with addition
136 # for randrw workload we count both reads and writes in same output bucket
137 # but we separate reads and writes for purposes of calculating
138 # end time for histogram record.
139 # this requires us to weight a raw histogram bucket by the
140 # fraction of time quantum that the bucket overlaps the current
141 # time quantum interval
142 # for example, if we have a bucket with 515 samples for time interval
143 # [ 1010, 2014 ] msec since start of test, and time quantum is 1 sec, then
144 # for time quantum interval [ 1000, 2000 ] msec, the overlap is
145 # (2000 - 1010) / (2000 - 1000) = 0.99
146 # so the contribution of this bucket to this time quantum is
147 # 515 x 0.99 = 509.85
149 def align_histo_log(raw_histogram_log, time_quantum, bucket_count, max_timestamp_ms):
151 # slice up test time int intervals of time_quantum seconds
153 (end_time, time_interval_count) = get_time_intervals(time_quantum, max_timestamp_ms)
154 time_qtm_ms = time_quantum * msec_per_sec
155 end_time_ms = end_time * msec_per_sec
156 aligned_intervals = []
157 for j in range(0, time_interval_count):
158 aligned_intervals.append((
160 [ 0.0 for j in range(0, bucket_count) ] ))
162 log_record_count = len(raw_histogram_log)
163 for k, record in enumerate(raw_histogram_log):
165 # find next record with same direction to get end-time
166 # have to avoid going past end of array
167 # for fio randrw workload,
168 # we have read and write records on same time interval
169 # sometimes read and write records are in opposite order
170 # assertion checks that next read/write record
171 # can be separated by at most 2 other records
173 (time_msec, direction, sz, interval_buckets) = record
174 if k+1 < log_record_count:
175 (time_msec_end, direction2, _, _) = raw_histogram_log[k+1]
176 if direction2 != direction:
177 if k+2 < log_record_count:
178 (time_msec_end, direction2, _, _) = raw_histogram_log[k+2]
179 if direction2 != direction:
180 if k+3 < log_record_count:
181 (time_msec_end, direction2, _, _) = raw_histogram_log[k+3]
182 assert direction2 == direction
184 time_msec_end = end_time_ms
186 time_msec_end = end_time_ms
188 time_msec_end = end_time_ms
190 # calculate first quantum that overlaps this histogram record
192 qtm_start_ms = (time_msec // time_qtm_ms) * time_qtm_ms
193 qtm_end_ms = ((time_msec + time_qtm_ms) // time_qtm_ms) * time_qtm_ms
194 qtm_index = qtm_start_ms // time_qtm_ms
196 # for each quantum that overlaps this histogram record's time interval
198 while qtm_start_ms < time_msec_end: # while quantum overlaps record
200 # calculate fraction of time that this quantum
201 # overlaps histogram record's time interval
203 overlap_start = max(qtm_start_ms, time_msec)
204 overlap_end = min(qtm_end_ms, time_msec_end)
205 weight = float(overlap_end - overlap_start)
206 weight /= (time_msec_end - time_msec)
207 (_,aligned_histogram) = aligned_intervals[qtm_index]
208 for bx, b in enumerate(interval_buckets):
209 weighted_bucket = weight * b
210 aligned_histogram[bx] += weighted_bucket
212 # advance to the next time quantum
214 qtm_start_ms += time_qtm_ms
215 qtm_end_ms += time_qtm_ms
218 return aligned_intervals
220 # add histogram in "source" to histogram in "target"
221 # it is assumed that the 2 histograms are precisely time-aligned
223 def add_to_histo_from( target, source ):
224 for b in range(0, len(source)):
225 target[b] += source[b]
227 # compute percentiles
229 # buckets: histogram bucket array
230 # wanted: list of floating-pt percentiles to calculate
231 # time_ranges: [tmin,tmax) time interval for each bucket
232 # returns None if no I/O reported.
233 # otherwise we would be dividing by zero
234 # think of buckets as probability distribution function
235 # and this loop is integrating to get cumulative distribution function
237 def get_pctiles(buckets, wanted, time_ranges):
239 # get total of IO requests done
241 for io_count in buckets:
242 total_ios += io_count
244 # don't return percentiles if no I/O was done during interval
248 pctile_count = len(wanted)
250 # results returned as dictionary keyed by percentile
253 # index of next percentile in list
257 next_pctile = wanted[pctile_index]
259 # no one is interested in percentiles bigger than this but not 100.0
260 # this prevents floating-point error from preventing loop exit
263 # pct is the percentile corresponding to
264 # all I/O requests up through bucket b
267 for b, io_count in enumerate(buckets):
270 total_so_far += io_count
271 # last_pct_lt is the percentile corresponding to
272 # all I/O requests up to, but not including, bucket b
274 pct = 100.0 * float(total_so_far) / total_ios
275 # a single bucket could satisfy multiple pctiles
276 # so this must be a while loop
277 # for 100-percentile (max latency) case, no bucket exceeds it
278 # so we must stop there.
279 while ((next_pctile == 100.0 and pct >= almost_100) or
280 (next_pctile < 100.0 and pct > next_pctile)):
281 # interpolate between min and max time for bucket time interval
282 # we keep the time_ranges access inside this loop,
283 # even though it could be above the loop,
284 # because in many cases we will not be even entering
285 # the loop so we optimize out these accesses
286 range_max_time = time_ranges[b][1]
287 range_min_time = time_ranges[b][0]
288 offset_frac = (next_pctile - last_pct)/(pct - last_pct)
289 interpolation = range_min_time + (offset_frac*(range_max_time - range_min_time))
290 pctile_result[next_pctile] = interpolation
292 if pctile_index == pctile_count:
294 next_pctile = wanted[pctile_index]
295 if pctile_index == pctile_count:
297 assert pctile_index == pctile_count
301 # this is really the main program
303 def compute_percentiles_from_logs():
304 parser = argparse.ArgumentParser()
305 parser.add_argument("--fio-version", dest="fio_version",
306 default="3", choices=[2,3], type=int,
307 help="fio version (default=3)")
308 parser.add_argument("--bucket-groups", dest="bucket_groups", default="29", type=int,
309 help="fio histogram bucket groups (default=29)")
310 parser.add_argument("--bucket-bits", dest="bucket_bits",
311 default="6", type=int,
312 help="fio histogram buckets-per-group bits (default=6 means 64 buckets/group)")
313 parser.add_argument("--percentiles", dest="pctiles_wanted",
314 default="0 50 95 99 100", type=float, nargs='+',
315 help="fio histogram buckets-per-group bits (default=6 means 64 buckets/group)")
316 parser.add_argument("--time-quantum", dest="time_quantum",
317 default="1", type=int,
318 help="time quantum in seconds (default=1)")
319 parser.add_argument("--output-unit", dest="output_unit",
320 default="usec", type=str,
321 help="Latency percentile output unit: msec|usec|nsec (default usec)")
322 parser.add_argument("file_list", nargs='+')
323 args = parser.parse_args()
326 if not args.bucket_groups:
327 # default changes based on fio version
329 args.bucket_groups = 19
332 args.bucket_groups = 29
336 print('bucket groups = %d' % args.bucket_groups)
337 print('bucket bits = %d' % args.bucket_bits)
338 print('time quantum = %d sec' % args.time_quantum)
339 print('percentiles = %s' % ','.join([ str(p) for p in args.pctiles_wanted ]))
340 buckets_per_group = 1 << args.bucket_bits
341 print('buckets per group = %d' % buckets_per_group)
342 buckets_per_interval = buckets_per_group * args.bucket_groups
343 print('buckets per interval = %d ' % buckets_per_interval)
344 bucket_index_range = range(0, buckets_per_interval)
345 if args.time_quantum == 0:
346 print('ERROR: time-quantum must be a positive number of seconds')
347 print('output unit = ' + args.output_unit)
348 if args.output_unit == 'msec':
349 time_divisor = 1000.0
350 elif args.output_unit == 'usec':
353 # calculate response time interval associated with each histogram bucket
355 bucket_times = time_ranges(args.bucket_groups, buckets_per_group, fio_version=args.fio_version)
357 # construct template for each histogram bucket array with buckets all zeroes
358 # we just copy this for each new histogram
360 zeroed_buckets = [ 0.0 for r in bucket_index_range ]
362 # print CSV header just like fiologparser_hist does
365 for p in args.pctiles_wanted:
366 header += '%3.1f, ' % p
367 print('time (millisec), percentiles in increasing order with values in ' + args.output_unit)
370 # parse the histogram logs
371 # assumption: each bucket has a monotonically increasing time
372 # assumption: time ranges do not overlap for a single thread's records
373 # (exception: if randrw workload, then there is a read and a write
374 # record for the same time interval)
376 max_timestamp_all_logs = 0
378 for fn in args.file_list:
380 (hist_files[fn], max_timestamp_ms) = parse_hist_file(fn, buckets_per_interval)
381 except FioHistoLogExc as e:
383 max_timestamp_all_logs = max(max_timestamp_all_logs, max_timestamp_ms)
385 (end_time, time_interval_count) = get_time_intervals(args.time_quantum, max_timestamp_all_logs)
386 all_threads_histograms = [ ((j*args.time_quantum*msec_per_sec), deepcopy(zeroed_buckets))
387 for j in range(0, time_interval_count) ]
389 for logfn in hist_files.keys():
390 aligned_per_thread = align_histo_log(hist_files[logfn],
392 buckets_per_interval,
393 max_timestamp_all_logs)
394 for t in range(0, time_interval_count):
395 (_, all_threads_histo_t) = all_threads_histograms[t]
396 (_, log_histo_t) = aligned_per_thread[t]
397 add_to_histo_from( all_threads_histo_t, log_histo_t )
399 # calculate percentiles across aggregate histogram for all threads
401 for (t_msec, all_threads_histo_t) in all_threads_histograms:
402 record = '%d, ' % t_msec
403 pct = get_pctiles(all_threads_histo_t, args.pctiles_wanted, bucket_times)
405 for w in args.pctiles_wanted:
408 pct_keys = [ k for k in pct.keys() ]
409 pct_values = [ str(pct[wanted]/time_divisor) for wanted in sorted(pct_keys) ]
410 record += ', '.join(pct_values)
419 ##### below are unit tests ##############
421 import tempfile, shutil
422 from os.path import join
423 should_not_get_here = False
425 class Test(unittest2.TestCase):
428 # a little less typing please
429 def A(self, boolean_val):
430 self.assertTrue(boolean_val)
432 # initialize unit test environment
436 d = tempfile.mkdtemp()
439 # remove anything left by unit test environment
440 # unless user sets UNITTEST_LEAVE_FILES environment variable
443 def tearDownClass(cls):
444 if not os.getenv("UNITTEST_LEAVE_FILES"):
445 shutil.rmtree(cls.tempdir)
448 self.fn = join(Test.tempdir, self.id())
450 def test_a_add_histos(self):
453 add_to_histo_from( a, b )
454 self.A(a == [2.5, 4.5])
455 self.A(b == [1.5, 2.5])
457 def test_b1_parse_log(self):
458 with open(self.fn, 'w') as f:
459 f.write('1234, 0, 4096, 1, 2, 3, 4\n')
460 f.write('5678,1,16384,5,6,7,8 \n')
461 (raw_histo_log, max_timestamp) = parse_hist_file(self.fn, 4) # 4 buckets per interval
462 self.A(len(raw_histo_log) == 2 and max_timestamp == 5678)
463 (time_ms, direction, bsz, histo) = raw_histo_log[0]
464 self.A(time_ms == 1234 and direction == 0 and bsz == 4096 and histo == [ 1, 2, 3, 4 ])
465 (time_ms, direction, bsz, histo) = raw_histo_log[1]
466 self.A(time_ms == 5678 and direction == 1 and bsz == 16384 and histo == [ 5, 6, 7, 8 ])
468 def test_b2_parse_empty_log(self):
469 with open(self.fn, 'w') as f:
472 (raw_histo_log, max_timestamp_ms) = parse_hist_file(self.fn, 4)
473 self.A(should_not_get_here)
474 except FioHistoLogExc as e:
475 self.A(str(e).startswith('no records'))
477 def test_b3_parse_empty_records(self):
478 with open(self.fn, 'w') as f:
480 f.write('1234, 0, 4096, 1, 2, 3, 4\n')
481 f.write('5678,1,16384,5,6,7,8 \n')
483 (raw_histo_log, max_timestamp_ms) = parse_hist_file(self.fn, 4)
484 self.A(len(raw_histo_log) == 2 and max_timestamp_ms == 5678)
485 (time_ms, direction, bsz, histo) = raw_histo_log[0]
486 self.A(time_ms == 1234 and direction == 0 and bsz == 4096 and histo == [ 1, 2, 3, 4 ])
487 (time_ms, direction, bsz, histo) = raw_histo_log[1]
488 self.A(time_ms == 5678 and direction == 1 and bsz == 16384 and histo == [ 5, 6, 7, 8 ])
490 def test_b4_parse_non_int(self):
491 with open(self.fn, 'w') as f:
492 f.write('12, 0, 4096, 1a, 2, 3, 4\n')
494 (raw_histo_log, _) = parse_hist_file(self.fn, 4)
496 except FioHistoLogExc as e:
497 self.A(str(e).startswith('non-integer'))
499 def test_b5_parse_neg_int(self):
500 with open(self.fn, 'w') as f:
501 f.write('-12, 0, 4096, 1, 2, 3, 4\n')
503 (raw_histo_log, _) = parse_hist_file(self.fn, 4)
505 except FioHistoLogExc as e:
506 self.A(str(e).startswith('negative integer'))
508 def test_b6_parse_too_few_int(self):
509 with open(self.fn, 'w') as f:
512 (raw_histo_log, _) = parse_hist_file(self.fn, 4)
514 except FioHistoLogExc as e:
515 self.A(str(e).startswith('too few numbers'))
517 def test_b7_parse_invalid_direction(self):
518 with open(self.fn, 'w') as f:
519 f.write('100, 2, 4096, 1, 2, 3, 4\n')
521 (raw_histo_log, _) = parse_hist_file(self.fn, 4)
523 except FioHistoLogExc as e:
524 self.A(str(e).startswith('invalid I/O direction'))
526 def test_b8_parse_bsz_too_big(self):
527 with open(self.fn+'_good', 'w') as f:
528 f.write('100, 1, %d, 1, 2, 3, 4\n' % (1<<24))
529 (raw_histo_log, max_timestamp_ms) = parse_hist_file(self.fn+'_good', 4)
530 with open(self.fn+'_bad', 'w') as f:
531 f.write('100, 1, 20000000, 1, 2, 3, 4\n')
533 (raw_histo_log, _) = parse_hist_file(self.fn+'_bad', 4)
535 except FioHistoLogExc as e:
536 self.A(str(e).startswith('block size too large'))
538 def test_b9_parse_wrong_bucket_count(self):
539 with open(self.fn, 'w') as f:
540 f.write('100, 1, %d, 1, 2, 3, 4, 5\n' % (1<<24))
542 (raw_histo_log, _) = parse_hist_file(self.fn, 4)
544 except FioHistoLogExc as e:
545 self.A(str(e).__contains__('buckets per interval'))
547 def test_c1_time_ranges(self):
548 ranges = time_ranges(3, 2) # fio_version defaults to 3
549 expected_ranges = [ # fio_version 3 is in nanoseconds
550 [0.000, 0.001], [0.001, 0.002], # first group
551 [0.002, 0.003], [0.003, 0.004], # second group same width
552 [0.004, 0.006], [0.006, 0.008]] # subsequent groups double width
553 self.A(ranges == expected_ranges)
554 ranges = time_ranges(3, 2, fio_version=3)
555 self.A(ranges == expected_ranges)
556 ranges = time_ranges(3, 2, fio_version=2)
557 expected_ranges_v2 = [ [ 1000.0 * min_or_max for min_or_max in time_range ]
558 for time_range in expected_ranges ]
559 self.A(ranges == expected_ranges_v2)
560 # see fio V3 stat.h for why 29 groups and 2^6 buckets/group
561 normal_ranges_v3 = time_ranges(29, 64)
562 # for v3, bucket time intervals are measured in nanoseconds
563 self.A(len(normal_ranges_v3) == 29 * 64 and normal_ranges_v3[-1][1] == 64*(1<<(29-1))/1000.0)
564 normal_ranges_v2 = time_ranges(19, 64, fio_version=2)
565 # for v2, bucket time intervals are measured in microseconds so we have fewer buckets
566 self.A(len(normal_ranges_v2) == 19 * 64 and normal_ranges_v2[-1][1] == 64*(1<<(19-1)))
568 def test_d1_align_histo_log_1_quantum(self):
569 with open(self.fn, 'w') as f:
570 f.write('100, 1, 4096, 1, 2, 3, 4')
571 (raw_histo_log, max_timestamp_ms) = parse_hist_file(self.fn, 4)
572 self.A(max_timestamp_ms == 100)
573 aligned_log = align_histo_log(raw_histo_log, 5, 4, max_timestamp_ms)
574 self.A(len(aligned_log) == 1)
575 (time_ms0, h) = aligned_log[0]
576 self.A(time_ms0 == 0 and h == [1.0, 2.0, 3.0, 4.0])
578 # we need this to compare 2 lists of floating point numbers for equality
579 # because of floating-point imprecision
581 def compare_2_floats(self, x, y):
582 if x == 0.0 or y == 0.0:
583 return (x+y) < 0.0000001
585 return (math.fabs(x-y)/x) < 0.00001
587 def is_close(self, buckets, buckets_expected):
588 if len(buckets) != len(buckets_expected):
590 compare_buckets = lambda k: self.compare_2_floats(buckets[k], buckets_expected[k])
591 indices_close = list(filter(compare_buckets, range(0, len(buckets))))
592 return len(indices_close) == len(buckets)
594 def test_d2_align_histo_log_2_quantum(self):
595 with open(self.fn, 'w') as f:
596 f.write('2000, 1, 4096, 1, 2, 3, 4\n')
597 f.write('7000, 1, 4096, 1, 2, 3, 4\n')
598 (raw_histo_log, max_timestamp_ms) = parse_hist_file(self.fn, 4)
599 self.A(max_timestamp_ms == 7000)
600 (_, _, _, raw_buckets1) = raw_histo_log[0]
601 (_, _, _, raw_buckets2) = raw_histo_log[1]
602 aligned_log = align_histo_log(raw_histo_log, 5, 4, max_timestamp_ms)
603 self.A(len(aligned_log) == 2)
604 (time_ms1, h1) = aligned_log[0]
605 (time_ms2, h2) = aligned_log[1]
606 # because first record is from time interval [2000, 7000]
607 # we weight it according
608 expect1 = [float(b) * 0.6 for b in raw_buckets1]
609 expect2 = [float(b) * 0.4 for b in raw_buckets1]
610 for e in range(0, len(expect2)):
611 expect2[e] += raw_buckets2[e]
612 self.A(time_ms1 == 0 and self.is_close(h1, expect1))
613 self.A(time_ms2 == 5000 and self.is_close(h2, expect2))
615 # what to expect if histogram buckets are all equal
616 def test_e1_get_pctiles_flat_histo(self):
617 with open(self.fn, 'w') as f:
618 buckets = [ 100 for j in range(0, 128) ]
619 f.write('9000, 1, 4096, %s\n' % ', '.join([str(b) for b in buckets]))
620 (raw_histo_log, max_timestamp_ms) = parse_hist_file(self.fn, 128)
621 self.A(max_timestamp_ms == 9000)
622 aligned_log = align_histo_log(raw_histo_log, 5, 128, max_timestamp_ms)
623 time_intervals = time_ranges(4, 32)
624 # since buckets are all equal, then median is halfway through time_intervals
625 # and max latency interval is at end of time_intervals
626 self.A(time_intervals[64][1] == 0.066 and time_intervals[127][1] == 0.256)
627 pctiles_wanted = [ 0, 50, 100 ]
629 for (time_ms, histo) in aligned_log:
630 pct_vs_time.append(get_pctiles(histo, pctiles_wanted, time_intervals))
631 self.A(pct_vs_time[0] == None) # no I/O in this time interval
632 expected_pctiles = { 0:0.000, 50:0.064, 100:0.256 }
633 self.A(pct_vs_time[1] == expected_pctiles)
635 # what to expect if just the highest histogram bucket is used
636 def test_e2_get_pctiles_highest_pct(self):
637 fio_v3_bucket_count = 29 * 64
638 with open(self.fn, 'w') as f:
639 # make a empty fio v3 histogram
640 buckets = [ 0 for j in range(0, fio_v3_bucket_count) ]
641 # add one I/O request to last bucket
643 f.write('9000, 1, 4096, %s\n' % ', '.join([str(b) for b in buckets]))
644 (raw_histo_log, max_timestamp_ms) = parse_hist_file(self.fn, fio_v3_bucket_count)
645 self.A(max_timestamp_ms == 9000)
646 aligned_log = align_histo_log(raw_histo_log, 5, fio_v3_bucket_count, max_timestamp_ms)
647 (time_ms, histo) = aligned_log[1]
648 time_intervals = time_ranges(29, 64)
649 expected_pctiles = { 100.0:(64*(1<<28))/1000.0 }
650 pct = get_pctiles( histo, [ 100.0 ], time_intervals )
651 self.A(pct == expected_pctiles)
653 # we are using this module as a standalone program
655 if __name__ == '__main__':
656 if os.getenv('UNITTEST'):
657 sys.exit(unittest2.main())
659 compute_percentiles_from_logs()