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
|
import os
import time
import litmus_util as lu
from operator import methodcaller
from tracer import SchedTracer, LogTracer, PerfTracer, LinuxTracer, OverheadTracer
class ExperimentException(Exception):
'''Used to indicate when there are problems with an experiment.'''
def __init__(self, name):
self.name = name
class ExperimentDone(ExperimentException):
'''Raised when an experiment looks like it's been run already.'''
def __str__(self):
return "Experiment finished already: %d" % self.name
class ExperimentInterrupted(ExperimentException):
'''Raised when an experiment appears to be interrupted (partial results).'''
def __str__(self):
return "Experiment was interrupted in progress: %d" % self.name
class ExperimentFailed(ExperimentException):
def __str__(self):
return "Experiment failed during execution: %d" % self.name
class Experiment(object):
'''Execute one task-set and save the results. Experiments have unique IDs.'''
INTERRUPTED_DIR = ".interrupted"
def __init__(self, name, scheduler, working_dir, finished_dir, proc_entries, executables):
'''Run an experiment, optionally wrapped in tracing.'''
self.name = name
self.scheduler = scheduler
self.working_dir = working_dir
self.finished_dir = finished_dir
self.proc_entries = proc_entries
self.executables = executables
self.exec_out = None
self.exec_err = None
self.__make_dirs()
self.__assign_executable_cwds()
self.tracers = []
if SchedTracer.enabled():
self.log("Enabling sched_trace")
self.tracers.append( SchedTracer(working_dir) )
if LinuxTracer.enabled():
self.log("Enabling trace-cmd / ftrace / kernelshark")
self.tracers.append( LinuxTracer(working_dir) )
if LogTracer.enabled():
self.log("Enabling logging")
self.tracers.append( LogTracer(working_dir) )
if PerfTracer.enabled():
self.log("Tracking CPU performance counters")
self.tracers.append( PerfTracer(working_dir) )
# Overhead trace must be handled seperately, see __run_tasks
if OverheadTracer.enabled():
self.log("Enabling overhead tracing")
self.overhead_trace = OverheadTracer(working_dir)
else:
self.overhead_trace = None
def __make_dirs(self):
interrupted = None
if os.path.exists(self.finished_dir):
raise ExperimentDone(self.name)
if os.path.exists(self.working_dir):
self.log("Found interrupted experiment, saving in %s" %
Experiment.INTERRUPTED_DIR)
interrupted = "%s/%s" % (os.path.split(self.working_dir)[0],
Experiment.INTERRUPTED_DIR)
os.rename(self.working_dir, interrupted)
os.mkdir(self.working_dir)
if interrupted:
os.rename(interrupted, "%s/%s" % (self.working_dir,
os.path.split(interrupted)[1]))
def __assign_executable_cwds(self):
def assign_cwd(executable):
executable.cwd = self.working_dir
map(assign_cwd, self.executables)
def __run_tasks(self):
already_waiting = lu.waiting_tasks()
self.log("Starting the programs")
for e in self.executables:
try:
e.execute()
except:
raise Exception("Executable failed: %s" % e)
self.log("Sleeping until tasks are ready for release...")
start = time.clock()
while (lu.waiting_tasks() - already_waiting) < len(self.executables):
if time.clock() - start > 30.0:
raise Exception("Too much time has passed waiting for tasks!")
time.sleep(1)
# Overhead tracer must be started right after release or overhead
# measurements will be full of irrelevant records
if self.overhead_trace:
self.log("Starting overhead trace")
self.overhead_trace.start_tracing()
self.log("Releasing %d tasks" % len(self.executables))
released = lu.release_tasks()
ret = True
if released != len(self.executables):
# Some tasks failed to release, kill all tasks and fail
# Need to re-release non-released tasks before we can kill them though
self.log("Failed to release {} tasks! Re-releasing and killing".format(
len(self.executables) - released, len(self.executables)))
time.sleep(5)
released = lu.release_tasks()
self.log("Re-released %d tasks" % released)
time.sleep(5)
self.log("Killing all tasks")
map(methodcaller('kill'), self.executables)
ret = False
self.log("Waiting for program to finish...")
for e in self.executables:
if not e.wait():
ret = False
# And it must be stopped here for the same reason
if self.overhead_trace:
self.log("Stopping overhead trace")
self.overhead_trace.stop_tracing()
if not ret:
raise ExperimentFailed(self.name)
def __save_results(self):
os.rename(self.working_dir, self.finished_dir)
def log(self, msg):
print "[Exp %s]: %s" % (self.name, msg)
def run_exp(self):
succ = False
try:
self.setup()
try:
self.__run_tasks()
self.log("Saving results in %s" % self.finished_dir)
succ = True
finally:
self.teardown()
finally:
self.log("Switching to Linux scheduler")
lu.switch_scheduler("Linux")
if succ:
self.__save_results()
self.log("Experiment done!")
def setup(self):
self.log("Writing %d proc entries" % len(self.proc_entries))
map(methodcaller('write_proc'), self.proc_entries)
self.log("Switching to %s" % self.scheduler)
lu.switch_scheduler(self.scheduler)
self.log("Starting %d tracers" % len(self.tracers))
map(methodcaller('start_tracing'), self.tracers)
self.exec_out = open('%s/exec-out.txt' % self.working_dir, 'w')
self.exec_err = open('%s/exec-err.txt' % self.working_dir, 'w')
def set_out(executable):
executable.stdout_file = self.exec_out
executable.stderr_file = self.exec_err
map(set_out, self.executables)
def teardown(self):
self.exec_out and self.exec_out.close()
self.exec_err and self.exec_err.close()
self.log("Stopping tracers")
map(methodcaller('stop_tracing'), self.tracers)
|
