-
Notifications
You must be signed in to change notification settings - Fork 77
/
aggregators.py
234 lines (199 loc) · 9.15 KB
/
aggregators.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
## -*- coding: utf-8 -*-
##
## aggregators.py
##
## Author: Toke Høiland-Jørgensen (toke@toke.dk)
## Date: 16 oktober 2012
## Copyright (c) 2012, Toke Høiland-Jørgensen
##
## This program is free software: you can redistribute it and/or modify
## it under the terms of the GNU General Public License as published by
## the Free Software Foundation, either version 3 of the License, or
## (at your option) any later version.
##
## This program is distributed in the hope that it will be useful,
## but WITHOUT ANY WARRANTY; without even the implied warranty of
## MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
## GNU General Public License for more details.
##
## You should have received a copy of the GNU General Public License
## along with this program. If not, see <http://www.gnu.org/licenses/>.
import math, pprint, signal
from datetime import datetime
from . import runners, transformers
from .util import classname
from .settings import settings
import collections
class Aggregator(object):
"""Basic aggregator. Runs all jobs and returns their result."""
def __init__(self):
self.instances = {}
self.threads = {}
if settings.LOG_FILE is None:
self.logfile = None
else:
self.logfile = open(settings.LOG_FILE, "a")
self.postprocessors = []
def add_instance(self, name, config):
instance = dict(config)
if not 'delay' in instance:
instance['delay'] = 0
instance['runner'] = getattr(runners, classname(instance['runner'], 'Runner'))
if 'data_transform' in config:
instance['transformers'] = []
for t in [i.strip() for i in config['data_transform'].split(',')]:
if hasattr(transformers, t):
instance['transformers'].append(getattr(transformers, t))
self.instances[name] = instance
duplicates = config.get('duplicates', None)
if duplicates is not None:
for i in range(int(duplicates)-1):
self.instances["%s - %d" % (name, i+2)] = instance
def aggregate(self):
raise NotImplementedError()
def collect(self):
"""Create a ProcessRunner thread for each instance and start them. Wait
for the threads to exit, then collect the results."""
if self.logfile:
self.logfile.write("Start run at %s\n" % datetime.now())
result = {}
try:
for n,i in list(self.instances.items()):
self.threads[n] = i['runner'](n, **i)
self.threads[n].start()
for n,t in list(self.threads.items()):
while t.isAlive():
t.join(1)
self._log(n,t)
if t.result is None:
continue
elif isinstance(t.result, collections.Callable):
# If the result is callable, the runner is really a
# post-processor (Avg etc), and should be run as such (by the
# postprocess() method)
self.postprocessors.append(t.result)
else:
result[n] = t.result
if 'transformers' in self.instances[n]:
for tr in self.instances[n]['transformers']:
result[n] = tr(result[n])
except KeyboardInterrupt:
self.kill_runners()
raise
if self.logfile is not None:
self.logfile.write("Raw aggregated data:\n")
pprint.pprint(result, self.logfile)
return result
def kill_runners(self):
for t in list(self.threads.values()):
t.killed = True
if hasattr(t, 'prog'):
try:
t.prog.send_signal(signal.SIGINT)
except OSError:
pass
def postprocess(self, result):
for p in self.postprocessors:
result = p(result)
return result
def _log(self, name, runner):
if self.logfile is None:
return
self.logfile.write("Runner: %s - %s\n" % (name, runner.__class__.__name__))
self.logfile.write("Command: %s\nReturncode: %d\n" % (runner.command, runner.returncode))
self.logfile.write("Program stdout:\n")
self.logfile.write(" " + "\n ".join(runner.out.splitlines()) + "\n")
self.logfile.write("Program stderr:\n")
self.logfile.write(" " + "\n ".join(runner.err.splitlines()) + "\n")
class IterationAggregator(Aggregator):
"""Iteration aggregator. Runs the jobs multiple times and aggregates the
results. Assumes each job outputs one value."""
def __init__(self, *args, **kwargs):
self.iterations = settings.ITERATIONS
Aggregator.__init__(self, *args, **kwargs)
def aggregate(self, results):
results.x_values = list(range(1, self.iterations+1))
for i in range(self.iterations):
results.add_result(i+1, self.collect())
return results
class TimeseriesAggregator(Aggregator):
"""Time series aggregator. Runs the jobs (which are all assumed to output a
series of timed entries) and combines the times onto a single timeline,
aligning values to the same time steps (interpolating values as necessary).
Assumes each job outputs a list of pairs (time, value) where the times and
values are floating point values."""
def __init__(self, *args, **kwargs):
self.step = settings.STEP_SIZE
self.max_distance = self.step * 5.0
Aggregator.__init__(self, *args, **kwargs)
def aggregate(self, results):
measurements = self.collect()
if not measurements:
raise RuntimeError("No data to aggregate. Run with -l and check log file to investigate.")
results.create_series(list(measurements.keys()))
# We start steps at the minimum time value, and do as many steps as are
# necessary to get past the maximum time value with the selected step
# size
first_times = [i[0][0] for i in list(measurements.values()) if i and i[0]]
last_times = [i[-1][0] for i in list(measurements.values()) if i and i[-1]]
if not (first_times and last_times):
raise RuntimeError("No data to aggregate. Run with -l and check log file to investigate.")
t_0 = min(first_times)
t_max = max(last_times)
steps = int(math.ceil((t_max-t_0)/self.step))
time_labels = []
for s in range(steps):
time_label = self.step*s
t = t_0 + self.step*s
# for each step we need to find the interpolated measurement value
# at time t by interpolating between the nearest measurements before
# and after t
result = {}
# n is the name of this measurement (from the config), r is the list
# of measurement pairs (time,value)
for n,r in list(measurements.items()):
max_dist = self.max_distance
last = False
if not r:
continue
t_prev = v_prev = None
t_next = v_next = None
# Some measurements (notably UDP pings) give a spurious value
# for the last measurement, so cut off the very last data point
# from each series. This should hopefully not lose any valuable
# data.
for i in range(len(r)-1):
if r[i][0] > t:
if i > 0:
t_prev,v_prev = r[i-1]
else:
# minimum interpolation distance on first entry to
# avoid multiple interpolations to the same value
max_dist = 0.1
t_next,v_next = r[i]
break
if t_next is None:
t_next,v_next = r[-1]
last = True
if abs(t-t_next) <= max_dist:
if t_prev is None:
# The first/last data point for this measurement is after the
# current t. Don't interpolate, just use the value.
if last and results.last_datapoint(n) == v_next:
# Avoid repeating last interpolation
result[n] = None
else:
result[n] = v_next
else:
# We found the previous and next values; interpolate between
# them. We assume that the rate of change dv/dt is constant
# in the interval, and so can be calculated as
# (v_next-v_prev)/(t_next-t_prev). Then the value v_t at t
# can be calculated as v_t=v_prev + dv/dt*(t-t_prev)
dv_dt = (v_next-v_prev)/(t_next-t_prev)
result[n] = v_prev + dv_dt*(t-t_prev)
else:
# Interpolation distance is too long; don't use the value.
result[n] = None
results.append_datapoint(time_label, result)
return results