blob: b61b7df81809f67896efd4e7d5fd1f4d70068382 [file] [log] [blame]
#!/usr/bin/env vpython
# Copyright 2016 The Chromium Authors. All rights reserved.
# Use of this source code is governed by a BSD-style license that can be
# found in the LICENSE file.
""" Generates legacy perf dashboard json from non-telemetry based perf tests.
Taken from chromium/build/scripts/slave/
import collections
import json
import math
import logging
import re
class LegacyResultsProcessor(object):
"""Class for any log processor expecting standard data to be graphed.
The log will be parsed looking for any lines of the forms:
<*>RESULT <graph_name>: <trace_name>= <value> <units>
<*>RESULT <graph_name>: <trace_name>= [<value>,value,value,...] <units>
<*>RESULT <graph_name>: <trace_name>= {<mean>, <std deviation>} <units>
For example,
*RESULT vm_final_browser: OneTab= 8488 kb
RESULT startup: ref= [167.00,148.00,146.00,142.00] ms
RESULT TabCapturePerformance_foo: Capture= {30.7, 1.45} ms
The leading * is optional; it indicates that the data from that line should
be considered "important", which may mean for example that it's graphed by
If multiple values are given in [], their mean and (sample) standard
deviation will be written; if only one value is given, that will be written.
A trailing comma is permitted in the list of values.
NOTE: All lines except for RESULT lines are ignored, including the Avg and
Stddev lines output by Telemetry!
Any of the <fields> except <value> may be empty, in which case the
not-terribly-useful defaults will be used. The <graph_name> and <trace_name>
should not contain any spaces, colons (:) nor equals-signs (=). Furthermore,
the <trace_name> will be used on the waterfall display, so it should be kept
short. If the trace_name ends with '_ref', it will be interpreted as a
reference value, and shown alongside the corresponding main value on the
Semantic note: The terms graph and chart are used interchangeably here.
r'(?P<GRAPH>[^:]*): (?P<TRACE>[^=]*)= '
r'(?P<VALUE>[\{\[]?[-\d\., ]+[\}\]]?)('
r' ?(?P<UNITS>.+))?')
# TODO(eyaich): Determine if this format is still used by any perf tests
r'(?P<GRAPH>[^:]*): (?P<TRACE>[^=]*)= '
def __init__(self):
# A dict of Graph objects, by name.
self._graphs = {}
# A dict mapping output file names to lists of lines in a file.
self._output = {}
self._percentiles = [.1, .25, .5, .75, .90, .95, .99]
class Trace(object):
"""Encapsulates data for one trace. Here, this means one point."""
def __init__(self):
self.important = False
self.values = []
self.mean = 0.0
self.stddev = 0.0
def __str__(self):
result = _FormatHumanReadable(self.mean)
if self.stddev:
result += '+/-%s' % _FormatHumanReadable(self.stddev)
return result
class Graph(object):
"""Encapsulates a set of points that should appear on the same graph."""
def __init__(self):
self.units = None
self.traces = {}
def IsImportant(self):
"""A graph is considered important if any of its traces is important."""
for trace in self.traces.itervalues():
if trace.important:
return True
return False
def BuildTracesDict(self):
"""Returns a dictionary mapping trace names to [value, stddev]."""
traces_dict = {}
for name, trace in self.traces.items():
traces_dict[name] = [str(trace.mean), str(trace.stddev)]
return traces_dict
def GenerateJsonResults(self, filename):
# Iterate through the file and process each output line
with open(filename) as f:
for line in f.readlines():
# After all results have been seen, generate the graph json data
return self.GenerateGraphJson()
def _PrependLog(self, filename, data):
"""Prepends some data to an output file."""
self._output[filename] = data + self._output.get(filename, [])
def ProcessLine(self, line):
"""Processes one result line, and updates the state accordingly."""
results_match =
histogram_match =
if results_match:
elif histogram_match:
raise Exception("Error: Histogram results parsing not supported yet")
def _ProcessResultLine(self, line_match):
"""Processes a line that matches the standard RESULT line format.
line_match: A MatchObject as returned by
match_dict = line_match.groupdict()
graph_name = match_dict['GRAPH'].strip()
trace_name = match_dict['TRACE'].strip()
graph = self._graphs.get(graph_name, self.Graph())
graph.units = (match_dict['UNITS'] or '').strip()
trace = graph.traces.get(trace_name, self.Trace())
value = match_dict['VALUE']
trace.important = match_dict['IMPORTANT'] or False
# Compute the mean and standard deviation for a list or a histogram,
# or the numerical value of a scalar value.
if value.startswith('['):
value_list = [float(x) for x in value.strip('[],').split(',')]
except ValueError:
# Report, but ignore, corrupted data lines. (Lines that are so badly
# broken that they don't even match the RESULTS_REGEX won't be
# detected.)
logging.warning("Bad test output: '%s'" % value.strip())
trace.values += value_list
trace.mean, trace.stddev, filedata = self._CalculateStatistics(
trace.values, trace_name)
assert filedata is not None
for filename in filedata:
self._PrependLog(filename, filedata[filename])
elif value.startswith('{'):
stripped = value.strip('{},')
trace.mean, trace.stddev = [float(x) for x in stripped.split(',')]
except ValueError:
logging.warning("Bad test output: '%s'" % value.strip())
trace.mean, trace.stddev, filedata = self._CalculateStatistics(
trace.values, trace_name)
assert filedata is not None
for filename in filedata:
self._PrependLog(filename, filedata[filename])
except ValueError:
logging.warning("Bad test output: '%s'" % value.strip())
graph.traces[trace_name] = trace
self._graphs[graph_name] = graph
def GenerateGraphJson(self):
"""Writes graph json for each graph seen.
charts = {}
for graph_name, graph in self._graphs.iteritems():
traces = graph.BuildTracesDict()
# Traces should contain exactly two elements: [mean, stddev].
for _, trace in traces.iteritems():
assert len(trace) == 2
graph_dict = collections.OrderedDict([
('traces', traces),
('units', str(graph.units)),
# Include a sorted list of important trace names if there are any.
important = [t for t in graph.traces.keys() if graph.traces[t].important]
if important:
graph_dict['important'] = sorted(important)
charts[graph_name] = graph_dict
return json.dumps(charts)
# _CalculateStatistics needs to be a member function.
# pylint: disable=R0201
# Unused argument value_list.
# pylint: disable=W0613
def _CalculateStatistics(self, value_list, trace_name):
"""Returns a tuple with some statistics based on the given value list.
This method may be overridden by subclasses wanting a different standard
deviation calcuation (or some other sort of error value entirely).
value_list: the list of values to use in the calculation
trace_name: the trace that produced the data (not used in the base
implementation, but subclasses may use it)
A 3-tuple - mean, standard deviation, and a dict which is either
empty or contains information about some file contents.
n = len(value_list)
if n == 0:
return 0.0, 0.0, {}
mean = float(sum(value_list)) / n
variance = sum([(element - mean)**2 for element in value_list]) / n
stddev = math.sqrt(variance)
return mean, stddev, {}
def _FormatHumanReadable(number):
"""Formats a float into three significant figures, using metric suffixes.
Only m, k, and M prefixes (for 1/1000, 1000, and 1,000,000) are used.
0.0387 => 38.7m
1.1234 => 1.12
10866 => 10.8k
682851200 => 683M
metric_prefixes = {-3: 'm', 0: '', 3: 'k', 6: 'M'}
scientific = '%.2e' % float(number) # 6.83e+005
e_idx = scientific.find('e') # 4, or 5 if negative
digits = float(scientific[:e_idx]) # 6.83
exponent = int(scientific[e_idx + 1:]) # int('+005') = 5
while exponent % 3:
digits *= 10
exponent -= 1
while exponent > 6:
digits *= 10
exponent -= 1
while exponent < -3:
digits /= 10
exponent += 1
if digits >= 100:
# Don't append a meaningless '.0' to an integer number.
digits = int(digits) # pylint: disable=redefined-variable-type
# Exponent is now divisible by 3, between -3 and 6 inclusive: (-3, 0, 3, 6).
return '%s%s' % (digits, metric_prefixes[exponent])