|
1
|
|
|
import os |
|
2
|
|
|
import unittest |
|
3
|
|
|
import tempfile |
|
4
|
|
|
import shutil |
|
5
|
|
|
from os.path import join as pjoin, abspath |
|
6
|
|
|
from mock import patch |
|
7
|
|
|
from subprocess import call |
|
8
|
|
|
import subprocess |
|
9
|
|
|
from nose.tools import assert_true, assert_equal |
|
10
|
|
|
from smartdispatch import smartdispatch_script |
|
11
|
|
|
import six |
|
12
|
|
|
import sys |
|
13
|
|
|
|
|
14
|
|
|
class TestSmartdispatcher(unittest.TestCase): |
|
15
|
|
|
|
|
16
|
|
|
def setUp(self): |
|
17
|
|
|
self.testing_dir = tempfile.mkdtemp() |
|
18
|
|
|
self.logs_dir = os.path.join(self.testing_dir, 'SMART_DISPATCH_LOGS') |
|
19
|
|
|
|
|
20
|
|
|
self.folded_commands = 'echo "[1 2 3 4]" "[6 7 8]" "[9 0]"' |
|
21
|
|
|
self.commands = ["echo 1 6 9", "echo 1 6 0", "echo 1 7 9", "echo 1 7 0", "echo 1 8 9", "echo 1 8 0", |
|
22
|
|
|
"echo 2 6 9", "echo 2 6 0", "echo 2 7 9", "echo 2 7 0", "echo 2 8 9", "echo 2 8 0", |
|
23
|
|
|
"echo 3 6 9", "echo 3 6 0", "echo 3 7 9", "echo 3 7 0", "echo 3 8 9", "echo 3 8 0", |
|
24
|
|
|
"echo 4 6 9", "echo 4 6 0", "echo 4 7 9", "echo 4 7 0", "echo 4 8 9", "echo 4 8 0"] |
|
25
|
|
|
self.nb_commands = len(self.commands) |
|
26
|
|
|
|
|
27
|
|
|
scripts_path = abspath(pjoin(os.path.dirname(__file__), os.pardir, "scripts")) |
|
28
|
|
|
self.smart_dispatch_command = '{} -C 1 -G 1 -q test -t 5:00 -x'.format(pjoin(scripts_path, 'smart-dispatch')) |
|
29
|
|
|
self.launch_command = "{0} launch {1}".format(self.smart_dispatch_command, self.folded_commands) |
|
30
|
|
|
self.resume_command = "{0} resume {{0}}".format(self.smart_dispatch_command) |
|
31
|
|
|
|
|
32
|
|
|
self.smart_dispatch_launcher_command = '{} -C 1 -G 1 -q test -t 5:00'.format(pjoin(scripts_path, 'smart-dispatch')) |
|
33
|
|
|
self.launcher_command = "{0} launch {1}".format(self.smart_dispatch_launcher_command, self.folded_commands) |
|
34
|
|
|
|
|
35
|
|
|
smart_dispatch_command_with_pool = '{} --pool 10 -C 1 -G 1 -q test -t 5:00 -x {{0}}'.format(pjoin(scripts_path, 'smart-dispatch')) |
|
36
|
|
|
self.launch_command_with_pool = smart_dispatch_command_with_pool.format('launch ' + self.folded_commands) |
|
37
|
|
|
self.nb_workers = 10 |
|
38
|
|
|
|
|
39
|
|
|
smart_dispatch_command_with_cores = '{} -C 1 -G 1 -c {{cores}} -q test -t 5:00 -x {{0}}'.format(pjoin(scripts_path, 'smart-dispatch')) |
|
40
|
|
|
self.launch_command_with_cores = smart_dispatch_command_with_cores.format('launch ' + self.folded_commands, cores='{cores}') |
|
41
|
|
|
|
|
42
|
|
|
smart_dispatch_command_with_gpus = '{} -C 1 -G 1 -g {{gpus}} -q test -t 5:00 -x {{0}}'.format(pjoin(scripts_path, 'smart-dispatch')) |
|
43
|
|
|
self.launch_command_with_gpus = smart_dispatch_command_with_gpus.format('launch ' + self.folded_commands, gpus='{gpus}') |
|
44
|
|
|
|
|
45
|
|
|
self._cwd = os.getcwd() |
|
46
|
|
|
os.chdir(self.testing_dir) |
|
47
|
|
|
|
|
48
|
|
|
def tearDown(self): |
|
49
|
|
|
os.chdir(self._cwd) |
|
50
|
|
|
shutil.rmtree(self.testing_dir) |
|
51
|
|
|
|
|
52
|
|
|
def test_main_launch(self): |
|
53
|
|
|
# Actual test |
|
54
|
|
|
exit_status = call(self.launch_command, shell=True) |
|
55
|
|
|
|
|
56
|
|
|
# Test validation |
|
57
|
|
|
assert_equal(exit_status, 0) |
|
58
|
|
|
assert_true(os.path.isdir(self.logs_dir)) |
|
59
|
|
|
assert_equal(len(os.listdir(self.logs_dir)), 1) |
|
60
|
|
|
|
|
61
|
|
|
batch_uid = os.listdir(self.logs_dir)[0] |
|
62
|
|
|
path_job_commands = os.path.join(self.logs_dir, batch_uid, "commands") |
|
63
|
|
|
assert_equal(len(os.listdir(path_job_commands)), self.nb_commands + 1) |
|
64
|
|
|
|
|
65
|
|
|
def test_launch_using_commands_file(self): |
|
66
|
|
|
# Actual test |
|
67
|
|
|
commands_filename = "commands_to_run.txt" |
|
68
|
|
|
open(commands_filename, 'w').write("\n".join(self.commands)) |
|
69
|
|
|
|
|
70
|
|
|
launch_command = self.smart_dispatch_command + " -f {0} launch".format(commands_filename) |
|
71
|
|
|
exit_status = call(launch_command, shell=True) |
|
72
|
|
|
|
|
73
|
|
|
# Test validation |
|
74
|
|
|
assert_equal(exit_status, 0) |
|
75
|
|
|
assert_true(os.path.isdir(self.logs_dir)) |
|
76
|
|
|
assert_equal(len(os.listdir(self.logs_dir)), 1) |
|
77
|
|
|
|
|
78
|
|
|
batch_uid = os.listdir(self.logs_dir)[0] |
|
79
|
|
|
path_job_commands = os.path.join(self.logs_dir, batch_uid, "commands") |
|
80
|
|
|
assert_equal(len(os.listdir(path_job_commands)), self.nb_commands + 1) |
|
81
|
|
|
assert_equal(open(pjoin(path_job_commands, 'commands.txt')).read(), "\n".join(self.commands) + "\n") |
|
82
|
|
|
|
|
83
|
|
|
def test_main_launch_with_pool_of_workers(self): |
|
84
|
|
|
# Actual test |
|
85
|
|
|
exit_status = call(self.launch_command_with_pool, shell=True) |
|
86
|
|
|
|
|
87
|
|
|
# Test validation |
|
88
|
|
|
assert_equal(exit_status, 0) |
|
89
|
|
|
assert_true(os.path.isdir(self.logs_dir)) |
|
90
|
|
|
assert_equal(len(os.listdir(self.logs_dir)), 1) |
|
91
|
|
|
|
|
92
|
|
|
batch_uid = os.listdir(self.logs_dir)[0] |
|
93
|
|
|
path_job_commands = os.path.join(self.logs_dir, batch_uid, "commands") |
|
94
|
|
|
assert_equal(len(os.listdir(path_job_commands)), self.nb_workers + 1) |
|
95
|
|
|
|
|
96
|
|
|
def test_main_launch_with_cores_command(self): |
|
97
|
|
|
# Actual test |
|
98
|
|
|
exit_status_0 = call(self.launch_command_with_cores.format(cores=0), shell=True) |
|
99
|
|
|
exit_status_100 = call(self.launch_command_with_cores.format(cores=100), shell=True) |
|
100
|
|
|
|
|
101
|
|
|
# Test validation |
|
102
|
|
|
assert_equal(exit_status_0, 2) |
|
103
|
|
|
assert_equal(exit_status_100, 2) |
|
104
|
|
|
assert_true(os.path.isdir(self.logs_dir)) |
|
105
|
|
|
|
|
106
|
|
|
def test_main_launch_with_gpus_command(self): |
|
107
|
|
|
# Actual test |
|
108
|
|
|
exit_status_0 = call(self.launch_command_with_gpus.format(gpus=0), shell=True) |
|
109
|
|
|
exit_status_100 = call(self.launch_command_with_gpus.format(gpus=100), shell=True) |
|
110
|
|
|
|
|
111
|
|
|
# Test validation |
|
112
|
|
|
assert_equal(exit_status_0, 0) |
|
113
|
|
|
assert_equal(exit_status_100, 2) |
|
114
|
|
|
assert_true(os.path.isdir(self.logs_dir)) |
|
115
|
|
|
|
|
116
|
|
View Code Duplication |
|
|
|
|
|
|
|
117
|
|
|
def test_gpu_check(self): |
|
118
|
|
|
|
|
119
|
|
|
argv = ['-x', '-g', '2', '-G', '1', '-C', '1', '-q', 'random', '-t', '00:00:10' ,'launch', 'echo', 'testing123'] |
|
120
|
|
|
|
|
121
|
|
|
# Test if the check fail |
|
122
|
|
|
with self.assertRaises(SystemExit) as context: |
|
123
|
|
|
smartdispatch_script.main(argv=argv) |
|
124
|
|
|
|
|
125
|
|
|
self.assertTrue(context.exception.code, 2) |
|
126
|
|
|
|
|
127
|
|
|
# Test if the test pass |
|
128
|
|
|
argv[2] = '0' |
|
129
|
|
|
|
|
130
|
|
|
try: |
|
131
|
|
|
smartdispatch_script.main(argv=argv) |
|
132
|
|
|
except SystemExit as e: |
|
133
|
|
|
self.fail("The command failed the check, but it was supposed to pass.") |
|
134
|
|
|
|
|
135
|
|
|
|
|
136
|
|
|
def test_cpu_check(self): |
|
137
|
|
|
|
|
138
|
|
|
argv = ['-x', '-c', '2', '-C', '1', '-G', '1', '-t', '00:00:10', '-q', 'random', 'launch', 'echo', 'testing123'] |
|
139
|
|
|
|
|
140
|
|
|
# Test if the check fail |
|
141
|
|
|
with self.assertRaises(SystemExit) as context: |
|
142
|
|
|
smartdispatch_script.main(argv=argv) |
|
143
|
|
|
|
|
144
|
|
|
self.assertTrue(context.exception.code, 2) |
|
145
|
|
|
|
|
146
|
|
|
# Test if the test pass |
|
147
|
|
|
argv[2] = '1' |
|
148
|
|
|
|
|
149
|
|
|
try: |
|
150
|
|
|
smartdispatch_script.main(argv=argv) |
|
151
|
|
|
except SystemExit as e: |
|
152
|
|
|
self.fail("The command failed the check, but it was supposed to pass.") |
|
153
|
|
|
|
|
154
|
|
|
|
|
155
|
|
|
@patch('subprocess.check_output') |
|
156
|
|
|
def test_launch_job_check(self, mock_check_output): |
|
157
|
|
|
|
|
158
|
|
|
#For this test, we won't call the script directly, since we want to mock subprocess.check_output |
|
159
|
|
|
argv = ['-t', '0:0:1', '-G', '1', '-C', '1', '-q', 'random', 'launch', 'echo', 'testing123'] |
|
160
|
|
|
|
|
161
|
|
|
# Test if the test pass (i.e the script run normaly) |
|
162
|
|
|
mock_check_output.side_effect = None |
|
163
|
|
|
mock_check_output.return_value = "" |
|
164
|
|
|
|
|
165
|
|
|
try: |
|
166
|
|
|
smartdispatch_script.main(argv=argv) |
|
167
|
|
|
except SystemExit as e: |
|
168
|
|
|
self.fail("The launcher had no problem, but the script failed nonetheless.") |
|
169
|
|
|
|
|
170
|
|
|
# Test if the check fail |
|
171
|
|
|
mock_check_output.side_effect = subprocess.CalledProcessError(1, 1, "A wild error appeared!") |
|
172
|
|
|
|
|
173
|
|
|
try: |
|
174
|
|
|
with self.assertRaises(SystemExit) as context: |
|
175
|
|
|
smartdispatch_script.main(argv=argv) |
|
176
|
|
|
|
|
177
|
|
|
self.assertTrue(context.exception.code, 2) |
|
178
|
|
|
|
|
179
|
|
|
except subprocess.CalledProcessError: |
|
180
|
|
|
# Rerasing the exception |
|
181
|
|
|
orig_exc_type, orig_exc_value, orig_exc_traceback = sys.exc_info() |
|
182
|
|
|
|
|
183
|
|
|
new_exc = Exception("smartdispatch_script.main() raised subprocess.CalledProcessError unexpectedly") |
|
184
|
|
|
new_exc.reraised = True |
|
185
|
|
|
six.reraise(type(new_exc), new_exc, orig_exc_traceback) |
|
186
|
|
|
|
|
187
|
|
|
def test_main_resume(self): |
|
188
|
|
|
# Setup |
|
189
|
|
|
call(self.launch_command, shell=True) |
|
190
|
|
|
batch_uid = os.listdir(self.logs_dir)[0] |
|
191
|
|
|
|
|
192
|
|
|
# Simulate that some commands are in the running state. |
|
193
|
|
|
path_job_commands = os.path.join(self.logs_dir, batch_uid, "commands") |
|
194
|
|
|
pending_commands_file = pjoin(path_job_commands, "commands.txt") |
|
195
|
|
|
running_commands_file = pjoin(path_job_commands, "running_commands.txt") |
|
196
|
|
|
commands = open(pending_commands_file).read().strip().split("\n") |
|
197
|
|
View Code Duplication |
with open(running_commands_file, 'w') as running_commands: |
|
|
|
|
|
|
198
|
|
|
running_commands.write("\n".join(commands[::2]) + "\n") |
|
199
|
|
|
with open(pending_commands_file, 'w') as pending_commands: |
|
200
|
|
|
pending_commands.write("\n".join(commands[1::2]) + "\n") |
|
201
|
|
|
|
|
202
|
|
|
# Actual test (should move running commands back to pending). |
|
203
|
|
|
exit_status = call(self.resume_command.format(batch_uid), shell=True) |
|
204
|
|
|
|
|
205
|
|
|
# Test validation |
|
206
|
|
|
assert_equal(exit_status, 0) |
|
207
|
|
|
assert_true(os.path.isdir(self.logs_dir)) |
|
208
|
|
|
assert_equal(len(os.listdir(self.logs_dir)), 1) |
|
209
|
|
|
assert_equal(len(open(running_commands_file).readlines()), 0) |
|
210
|
|
|
assert_equal(len(open(pending_commands_file).readlines()), len(commands)) |
|
211
|
|
|
|
|
212
|
|
|
# Test when batch_uid is a path instead of a jobname. |
|
213
|
|
|
# Setup |
|
214
|
|
|
batch_uid = os.path.join(self.logs_dir, os.listdir(self.logs_dir)[0]) |
|
215
|
|
|
|
|
216
|
|
|
# Simulate that some commands are in the running state. |
|
217
|
|
|
path_job_commands = os.path.join(self.logs_dir, batch_uid, "commands") |
|
218
|
|
|
pending_commands_file = pjoin(path_job_commands, "commands.txt") |
|
219
|
|
|
running_commands_file = pjoin(path_job_commands, "running_commands.txt") |
|
220
|
|
|
commands = open(pending_commands_file).read().strip().split("\n") |
|
221
|
|
|
with open(running_commands_file, 'w') as running_commands: |
|
222
|
|
|
running_commands.write("\n".join(commands[::2]) + "\n") |
|
223
|
|
|
with open(pending_commands_file, 'w') as pending_commands: |
|
224
|
|
|
pending_commands.write("\n".join(commands[1::2]) + "\n") |
|
225
|
|
|
|
|
226
|
|
|
# Actual test (should move running commands back to pending). |
|
227
|
|
|
exit_status = call(self.resume_command.format(batch_uid), shell=True) |
|
228
|
|
|
|
|
229
|
|
|
# Test validation |
|
230
|
|
|
assert_equal(exit_status, 0) |
|
231
|
|
View Code Duplication |
assert_true(os.path.isdir(self.logs_dir)) |
|
|
|
|
|
|
232
|
|
|
assert_equal(len(os.listdir(self.logs_dir)), 1) |
|
233
|
|
|
assert_equal(len(open(running_commands_file).readlines()), 0) |
|
234
|
|
|
assert_equal(len(open(pending_commands_file).readlines()), len(commands)) |
|
235
|
|
|
|
|
236
|
|
|
def test_main_resume_by_expanding_pool_default(self): |
|
237
|
|
|
# Create SMART_DISPATCH_LOGS structure. |
|
238
|
|
|
call(self.launch_command, shell=True) |
|
239
|
|
|
batch_uid = os.listdir(self.logs_dir)[0] |
|
240
|
|
|
|
|
241
|
|
|
# Simulate that some commands are in the running state. |
|
242
|
|
|
nb_commands_files = 2 # 'commands.txt' and 'running_commands.txt' |
|
243
|
|
|
path_job_commands = os.path.join(self.logs_dir, batch_uid, "commands") |
|
244
|
|
|
pending_commands_file = pjoin(path_job_commands, "commands.txt") |
|
245
|
|
|
running_commands_file = pjoin(path_job_commands, "running_commands.txt") |
|
246
|
|
|
commands = open(pending_commands_file).read().strip().split("\n") |
|
247
|
|
|
with open(running_commands_file, 'w') as running_commands: |
|
248
|
|
|
running_commands.write("\n".join(commands[::2]) + "\n") |
|
249
|
|
|
with open(pending_commands_file, 'w') as pending_commands: |
|
250
|
|
|
pending_commands.write("\n".join(commands[1::2]) + "\n") |
|
251
|
|
|
|
|
252
|
|
|
# Remove PBS files so we can check that new ones are going to be created. |
|
253
|
|
|
for f in os.listdir(path_job_commands): |
|
254
|
|
|
if f.startswith('job_commands_') and f.endswith('.sh'): |
|
255
|
|
|
os.remove(pjoin(path_job_commands, f)) |
|
256
|
|
|
|
|
257
|
|
|
# Should NOT move running commands back to pending but should add new workers. |
|
258
|
|
|
command_line = self.resume_command.format(batch_uid) |
|
259
|
|
|
command_line += " --expandPool" |
|
260
|
|
|
exit_status = call(command_line, shell=True) |
|
261
|
|
|
|
|
262
|
|
|
# Test validation |
|
263
|
|
|
assert_equal(exit_status, 0) |
|
264
|
|
|
assert_equal(len(open(running_commands_file).readlines()), len(commands[::2])) |
|
265
|
|
|
assert_equal(len(open(pending_commands_file).readlines()), len(commands[1::2])) |
|
266
|
|
|
|
|
267
|
|
|
nb_job_commands_files = len(os.listdir(path_job_commands)) |
|
268
|
|
|
assert_equal(nb_job_commands_files-nb_commands_files, len(commands[1::2])) |
|
269
|
|
|
|
|
270
|
|
|
def test_main_resume_by_expanding_pool(self): |
|
271
|
|
|
# Create SMART_DISPATCH_LOGS structure. |
|
272
|
|
|
call(self.launch_command, shell=True) |
|
273
|
|
|
batch_uid = os.listdir(self.logs_dir)[0] |
|
274
|
|
|
|
|
275
|
|
|
# Simulate that some commands are in the running state. |
|
276
|
|
|
nb_commands_files = 2 # 'commands.txt' and 'running_commands.txt' |
|
277
|
|
|
path_job_commands = os.path.join(self.logs_dir, batch_uid, "commands") |
|
278
|
|
|
pending_commands_file = pjoin(path_job_commands, "commands.txt") |
|
279
|
|
|
running_commands_file = pjoin(path_job_commands, "running_commands.txt") |
|
280
|
|
|
commands = open(pending_commands_file).read().strip().split("\n") |
|
281
|
|
|
with open(running_commands_file, 'w') as running_commands: |
|
282
|
|
|
running_commands.write("\n".join(commands[::2]) + "\n") |
|
283
|
|
|
with open(pending_commands_file, 'w') as pending_commands: |
|
284
|
|
|
pending_commands.write("\n".join(commands[1::2]) + "\n") |
|
285
|
|
|
|
|
286
|
|
|
# Remove PBS files so we can check that new ones are going to be created. |
|
287
|
|
|
for f in os.listdir(path_job_commands): |
|
288
|
|
|
if f.startswith('job_commands_') and f.endswith('.sh'): |
|
289
|
|
|
os.remove(pjoin(path_job_commands, f)) |
|
290
|
|
|
|
|
291
|
|
|
# Should NOT move running commands back to pending but should add new workers. |
|
292
|
|
|
nb_workers_to_add = 3 |
|
293
|
|
|
command_line = self.resume_command.format(batch_uid) |
|
294
|
|
|
command_line += " --expandPool {}".format(nb_workers_to_add) |
|
295
|
|
|
exit_status = call(command_line, shell=True) |
|
296
|
|
|
|
|
297
|
|
|
# Test validation |
|
298
|
|
|
assert_equal(exit_status, 0) |
|
299
|
|
|
assert_equal(len(open(running_commands_file).readlines()), len(commands[::2])) |
|
300
|
|
|
assert_equal(len(open(pending_commands_file).readlines()), len(commands[1::2])) |
|
301
|
|
|
|
|
302
|
|
|
nb_job_commands_files = len(os.listdir(path_job_commands)) |
|
303
|
|
|
assert_equal(nb_job_commands_files-nb_commands_files, nb_workers_to_add) |
|
304
|
|
|
|
SMART-Lab /
smartdispatch
Loading history...
