-
Notifications
You must be signed in to change notification settings - Fork 0
/
launch_instances.py
196 lines (147 loc) · 7.15 KB
/
launch_instances.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
import boto.ec2
import boto.ec2.cloudwatch
import os
import subprocess
import numpy as np
from time import sleep
from timeit import default_timer as timer
from boto.manage.cmdshell import sshclient_from_instance
from settings import AWS_WORKER_AMI, AWS_BUCKET_NAME, AWS_REDUCER_INSTANCE_TYPE, AWS_WORKER_INSTANCE_TYPE, AWS_KEY_NAME, AWS_SECURITY_GROUP, REDUCER_SCRIPT, STARTUP_SCRIPT, WORKER_SCRIPT, SSH_LOCAL_KEY
MAX_RETRIES = 100
def launch_workers(num_workers):
if num_workers == 0:
return []
connection = boto.ec2.connect_to_region('us-west-2')
new_reservation = connection.run_instances(
AWS_WORKER_AMI,
key_name=AWS_KEY_NAME,
instance_type=AWS_WORKER_INSTANCE_TYPE,
security_group_ids=[AWS_SECURITY_GROUP],
min_count=num_workers,
max_count=num_workers)
for instance in new_reservation.instances:
connection.create_tags([instance.id],
{"Name" : "Worker"})
for instance in new_reservation.instances:
while instance.state == u'pending':
sleep(2)
instance.update()
upload_and_run_script(STARTUP_SCRIPT, instance)
return new_reservation.instances
def add_alarm_to_instances(instances):
connection = boto.ec2.cloudwatch.connect_to_region('us-west-2')
topic = 'arn:aws:sns:us-west-2:197873462522:scarp-cpu'
for i in instances:
sleep(1)
alarm = boto.ec2.cloudwatch.MetricAlarm(
name='low-cpu-' + i.id,
namespace='AWS/EC2',
metric='CPUUtilization',
statistic='Average',
period=300,
comparison='<=',
threshold=10,
evaluation_periods=5,
dimensions={'InstanceId' : [i.id]},
alarm_actions=[topic])
alarm.add_alarm_action('arn:aws:automate:us-west-2:ec2:terminate')
connection.put_metric_alarm(alarm)
def launch_reducer():
connection = boto.ec2.connect_to_region('us-west-2')
new_reservation = connection.run_instances(
AWS_WORKER_AMI,
key_name=AWS_KEY_NAME,
instance_type=AWS_REDUCER_INSTANCE_TYPE,
security_group_ids=[AWS_SECURITY_GROUP])
instance = new_reservation.instances[0]
connection.create_tags([instance.id],
{"Name" : "Reducer"})
while instance.state == u'pending':
sleep(5)
instance.update()
return instance
def upload_and_run_script(script, instance):
instance.update()
dns = instance.public_dns_name
with open('runme.sh', 'w') as f:
f.write(script)
commands = []
commands.append(['ssh', '-o IdentityFile=/home/ubuntu/aws-scarp.pem', '-o StrictHostKeyChecking=no','ubuntu@' + dns, 'sudo killall ipython screen'])
commands.append(['ssh', '-o IdentityFile=/home/ubuntu/aws-scarp.pem', '-o StrictHostKeyChecking=no','ubuntu@' + dns, 'screen -wipe'])
commands.append(['ssh', '-o IdentityFile=/home/ubuntu/aws-scarp.pem', '-o StrictHostKeyChecking=no','ubuntu@' + dns, 'sudo sysctl -w vm.drop_caches=3'])
commands.append(['scp', '-o IdentityFile=/home/ubuntu/aws-scarp.pem', '-o StrictHostKeyChecking=no', 'runme.sh', 'ubuntu@' + dns + ':/home/ubuntu/'])
commands.append(['ssh', '-o IdentityFile=/home/ubuntu/aws-scarp.pem', '-o StrictHostKeyChecking=no','ubuntu@' + dns, 'sudo /etc/init.d/screen-cleanup start'])
commands.append(['ssh', '-o IdentityFile=/home/ubuntu/aws-scarp.pem', '-o StrictHostKeyChecking=no','ubuntu@' + dns, 'screen -d -m chmod +x runme.sh'])
commands.append(['ssh', '-o IdentityFile=/home/ubuntu/aws-scarp.pem', '-o StrictHostKeyChecking=no','ubuntu@' + dns, 'screen -d -m ./runme.sh'])
commands.append(['ssh', '-o IdentityFile=/home/ubuntu/aws-scarp.pem', '-o StrictHostKeyChecking=no','ubuntu@' + dns, 'cd /home/ubuntu/scarp-reduce && screen -d -m ipython monitor.py'])
DEVNULL = open(os.devnull, 'w')
for command in commands:
start = timer()
subprocess.call(command, stdout=DEVNULL, stderr=subprocess.STDOUT)
stop = timer()
#print('Execute command: {:.2f} s'.format(stop - start))
DEVNULL.close()
def get_worker_instances():
connection = boto.ec2.connect_to_region('us-west-2')
reservations = connection.get_all_instances(filters={'tag:Name' : 'Worker', 'instance-state-name' : 'running'})
if len(reservations) == 0:
return []
else:
instance_list = [x.instances for x in reservations]
workers = set().union(*instance_list)
return workers
def run_job(instance, param):
script = WORKER_SCRIPT.format(param[0], param[1])
upload_and_run_script(script, instance)
instance.update()
connection = boto.ec2.connect_to_region('us-west-2')
connection.create_tags([instance.id], {"Param" : "{}".format(param)})
print("START: {} Started processing {:d} {:.2f}".format(instance.public_dns_name, param[0], param[1]))
def launch_jobs(d, num_workers):
min_age = 0
max_age = 3.5
ages = np.linspace(min_age, max_age, num_workers)
old_workers = get_worker_instances()
if len(old_workers) < num_workers:
workers = launch_workers(num_workers - len(old_workers))
workers.extend(old_workers)
else:
workers = old_workers
add_alarm_to_instances(workers)
for age, instance in zip(ages, workers):
run_job(instance, [d, age])
def terminate_instances(name="Worker"):
connection = boto.ec2.connect_to_region('us-west-2')
reservations = connection.get_all_instances(filters={'tag:Name' : name, 'instance-state-name' : 'running'})
for r in reservations:
ids = [i.id for i in r.instances]
connection.terminate_instances(ids=ids)
if __name__ == "__main__":
#d_age = 0.1
#num_ages = int((max_age - min_age) / d_age)
d = int(sys.argv[1])
num_workers = int(sys.argv[2])
remote_dir = sys.argv[3]
min_age = 0
max_age = 3.5
ages = np.linspace(min_age, max_age, num_workers)
start = timer()
old_workers = get_worker_instances()
num_workers -= len(old_workers)
print("Using {} active nodes...".format(len(old_workers)))
print("Starting {} new nodes...".format(num_workers))
workers = launch_workers(num_workers)
#print("Adding CPU usage alarms to instances...")
#add_alarm_to_instances(workers)
workers.extend(old_workers)
stop = timer()
#print("Spin up: {:.2f} s".format((stop - start) / num_ages))
start = timer()
for age, instance in zip(ages, workers):
upload_and_run_script(STARTUP_SCRIPT, instance)
run_job(instance, [d, age])
stop = timer()
#reducer_instance = launch_reducer()
#reducer_script = REDUCER_SCRIPT.format(num_workers, remote_dir, d, num_workers, remote_dir)
#upload_and_run_script(reducer_script, reducer_instance)
#print("{}: Started reducer".format(reducer_instance.public_dns_name))