1
|
1 |
|
from __future__ import print_function, absolute_import |
2
|
|
|
|
3
|
1 |
|
from os import getenv |
4
|
1 |
|
from time import sleep |
5
|
|
|
|
6
|
1 |
|
import click |
7
|
1 |
|
import getpass |
8
|
1 |
|
from datetime import datetime, timedelta |
9
|
|
|
|
10
|
1 |
|
from ecs_deploy.ecs import DeployAction, ScaleAction, RunAction, EcsClient, \ |
11
|
|
|
TaskPlacementError |
12
|
1 |
|
from ecs_deploy.newrelic import Deployment |
13
|
|
|
|
14
|
|
|
|
15
|
1 |
|
@click.group() |
16
|
|
|
def ecs(): # pragma: no cover |
17
|
|
|
pass |
18
|
|
|
|
19
|
|
|
|
20
|
1 |
|
def get_client(access_key_id, secret_access_key, region, profile): |
21
|
1 |
|
return EcsClient(access_key_id, secret_access_key, region, profile) |
22
|
|
|
|
23
|
|
|
|
24
|
1 |
|
@click.command() |
25
|
1 |
|
@click.argument('cluster') |
26
|
1 |
|
@click.argument('service') |
27
|
1 |
|
@click.option('-t', '--tag', |
28
|
|
|
help='Changes the tag for ALL container images') |
29
|
1 |
|
@click.option('-i', '--image', type=(str, str), multiple=True, |
30
|
|
|
help='Overwrites the image for a container: ' |
31
|
|
|
'<container> <image>') |
32
|
1 |
|
@click.option('-c', '--command', type=(str, str), multiple=True, |
33
|
|
|
help='Overwrites the command in a container: ' |
34
|
|
|
'<container> <command>') |
35
|
1 |
|
@click.option('-e', '--env', type=(str, str, str), multiple=True, |
36
|
|
|
help='Adds or changes an environment variable: ' |
37
|
|
|
'<container> <name> <value>') |
38
|
1 |
|
@click.option('-r', '--role', type=str, |
39
|
|
|
help='Sets the task\'s role ARN: <task role ARN>') |
40
|
1 |
|
@click.option('--task', type=str, |
41
|
|
|
help='Task definition to be deployed. Can be a task ARN ' |
42
|
|
|
'or a task family with optional revision') |
43
|
1 |
|
@click.option('--region', required=False, |
44
|
|
|
help='AWS region (e.g. eu-central-1)') |
45
|
1 |
|
@click.option('--access-key-id', required=False, |
46
|
|
|
help='AWS access key id') |
47
|
1 |
|
@click.option('--secret-access-key', required=False, |
48
|
|
|
help='AWS secret access key') |
49
|
1 |
|
@click.option('--profile', required=False, |
50
|
|
|
help='AWS configuration profile name') |
51
|
1 |
|
@click.option('--timeout', required=False, default=300, type=int, |
52
|
|
|
help='Amount of seconds to wait for deployment before ' |
53
|
|
|
'command fails (default: 300)') |
54
|
1 |
|
@click.option('--ignore-warnings', is_flag=True, |
55
|
|
|
help='Do not fail deployment on warnings (port already in use ' |
56
|
|
|
'or insufficient memory/CPU)') |
57
|
1 |
|
@click.option('--newrelic-apikey', required=False, |
58
|
|
|
help='New Relic API Key for recording the deployment') |
59
|
1 |
|
@click.option('--newrelic-appid', required=False, |
60
|
|
|
help='New Relic App ID for recording the deployment') |
61
|
1 |
|
@click.option('--comment', required=False, |
62
|
|
|
help='Description/comment for recording the deployment') |
63
|
1 |
|
@click.option('--user', required=False, |
64
|
|
|
help='User who executes the deployment (used for recording)') |
65
|
1 |
|
@click.option('--diff/--no-diff', default=True, |
66
|
|
|
help='Print what values were changed in the task definition') |
67
|
1 |
|
@click.option('--deregister/--no-deregister', default=True, |
68
|
|
|
help='Deregister (default) or keep the old task definition') |
69
|
|
|
def deploy(cluster, service, tag, image, command, env, role, task, region, |
70
|
|
|
access_key_id, secret_access_key, profile, timeout, newrelic_apikey, |
71
|
|
|
newrelic_appid, comment, user, ignore_warnings, diff, deregister): |
72
|
|
|
""" |
73
|
|
|
Redeploy or modify a service. |
74
|
|
|
|
75
|
|
|
\b |
76
|
|
|
CLUSTER is the name of your cluster (e.g. 'my-custer') within ECS. |
77
|
|
|
SERVICE is the name of your service (e.g. 'my-app') within ECS. |
78
|
|
|
|
79
|
|
|
When not giving any other options, the task definition will not be changed. |
80
|
|
|
It will just be duplicated, so that all container images will be pulled |
81
|
|
|
and redeployed. |
82
|
|
|
""" |
83
|
|
|
|
84
|
1 |
|
try: |
85
|
1 |
|
client = get_client(access_key_id, secret_access_key, region, profile) |
86
|
1 |
|
deployment = DeployAction(client, cluster, service) |
87
|
|
|
|
88
|
1 |
|
td = get_task_definition(deployment, task) |
89
|
1 |
|
td.set_images(tag, **{key: value for (key, value) in image}) |
90
|
1 |
|
td.set_commands(**{key: value for (key, value) in command}) |
91
|
1 |
|
td.set_environment(env) |
92
|
1 |
|
td.set_role_arn(role) |
93
|
|
|
|
94
|
1 |
|
if diff: |
95
|
1 |
|
print_diff(td) |
96
|
|
|
|
97
|
1 |
|
new_task_definition = create_task_definition(deployment, td) |
98
|
1 |
|
record_deployment(tag, newrelic_apikey, newrelic_appid, comment, user) |
99
|
1 |
|
deploy_task_definition(deployment, new_task_definition) |
100
|
|
|
|
101
|
1 |
|
wait_for_finish( |
102
|
|
|
action=deployment, |
103
|
|
|
timeout=timeout, |
104
|
|
|
title='Deploying task definition', |
105
|
|
|
success_message='Deployment successful', |
106
|
|
|
failure_message='Deployment failed', |
107
|
|
|
ignore_warnings=ignore_warnings |
108
|
|
|
) |
109
|
|
|
|
110
|
1 |
|
if deregister: |
111
|
1 |
|
deregister_task_definition(deployment, td) |
112
|
|
|
|
113
|
1 |
|
except Exception as e: |
114
|
1 |
|
click.secho('%s\n' % str(e), fg='red', err=True) |
115
|
1 |
|
exit(1) |
116
|
|
|
|
117
|
|
|
|
118
|
1 |
|
@click.command() |
119
|
1 |
|
@click.argument('cluster') |
120
|
1 |
|
@click.argument('service') |
121
|
1 |
|
@click.argument('desired_count', type=int) |
122
|
1 |
|
@click.option('--region', |
123
|
|
|
help='AWS region (e.g. eu-central-1)') |
124
|
1 |
|
@click.option('--access-key-id', |
125
|
|
|
help='AWS access key id') |
126
|
1 |
|
@click.option('--secret-access-key', |
127
|
|
|
help='AWS secret access key') |
128
|
1 |
|
@click.option('--profile', |
129
|
|
|
help='AWS configuration profile name') |
130
|
1 |
|
@click.option('--timeout', default=300, type=int, |
131
|
|
|
help='AWS configuration profile') |
132
|
1 |
|
@click.option('--ignore-warnings', is_flag=True, |
133
|
|
|
help='Do not fail deployment on warnings (port already in use ' |
134
|
|
|
'or insufficient memory/CPU)') |
135
|
|
|
def scale(cluster, service, desired_count, access_key_id, secret_access_key, |
136
|
|
|
region, profile, timeout, ignore_warnings): |
137
|
|
|
""" |
138
|
|
|
Scale a service up or down. |
139
|
|
|
|
140
|
|
|
\b |
141
|
|
|
CLUSTER is the name of your cluster (e.g. 'my-custer') within ECS. |
142
|
|
|
SERVICE is the name of your service (e.g. 'my-app') within ECS. |
143
|
|
|
DESIRED_COUNT is the number of tasks your service should run. |
144
|
|
|
""" |
145
|
1 |
|
try: |
146
|
1 |
|
client = get_client(access_key_id, secret_access_key, region, profile) |
147
|
1 |
|
scaling = ScaleAction(client, cluster, service) |
148
|
1 |
|
click.secho('Updating service') |
149
|
1 |
|
scaling.scale(desired_count) |
150
|
1 |
|
click.secho( |
151
|
|
|
'Successfully changed desired count to: %s\n' % desired_count, |
152
|
|
|
fg='green' |
153
|
|
|
) |
154
|
1 |
|
wait_for_finish( |
155
|
|
|
action=scaling, |
156
|
|
|
timeout=timeout, |
157
|
|
|
title='Scaling service', |
158
|
|
|
success_message='Scaling successful', |
159
|
|
|
failure_message='Scaling failed', |
160
|
|
|
ignore_warnings=ignore_warnings |
161
|
|
|
) |
162
|
|
|
|
163
|
1 |
|
except Exception as e: |
164
|
1 |
|
click.secho('%s\n' % str(e), fg='red', err=True) |
165
|
1 |
|
exit(1) |
166
|
|
|
|
167
|
|
|
|
168
|
1 |
|
@click.command() |
169
|
1 |
|
@click.argument('cluster') |
170
|
1 |
|
@click.argument('task') |
171
|
1 |
|
@click.argument('count', required=False, default=1) |
172
|
1 |
|
@click.option('-c', '--command', type=(str, str), multiple=True, |
173
|
|
|
help='Overwrites the command in a container: ' |
174
|
|
|
'<container> <command>') |
175
|
1 |
|
@click.option('-e', '--env', type=(str, str, str), multiple=True, |
176
|
|
|
help='Adds or changes an environment variable: ' |
177
|
|
|
'<container> <name> <value>') |
178
|
1 |
|
@click.option('--region', |
179
|
|
|
help='AWS region (e.g. eu-central-1)') |
180
|
1 |
|
@click.option('--access-key-id', |
181
|
|
|
help='AWS access key id') |
182
|
1 |
|
@click.option('--secret-access-key', |
183
|
|
|
help='AWS secret access key') |
184
|
1 |
|
@click.option('--profile', |
185
|
|
|
help='AWS configuration profile name') |
186
|
1 |
|
@click.option('--diff/--no-diff', default=True, |
187
|
|
|
help='Print what values were changed in the task definition') |
188
|
|
|
def run(cluster, task, count, command, env, region, access_key_id, |
189
|
|
|
secret_access_key, profile, diff): |
190
|
|
|
""" |
191
|
|
|
Run a one-off task. |
192
|
|
|
|
193
|
|
|
\b |
194
|
|
|
CLUSTER is the name of your cluster (e.g. 'my-custer') within ECS. |
195
|
|
|
TASK is the name of your task definition (e.g. 'my-task') within ECS. |
196
|
|
|
COMMAND is the number of tasks your service should run. |
197
|
|
|
""" |
198
|
1 |
|
try: |
199
|
1 |
|
client = get_client(access_key_id, secret_access_key, region, profile) |
200
|
1 |
|
action = RunAction(client, cluster) |
201
|
|
|
|
202
|
1 |
|
td = action.get_task_definition(task) |
203
|
1 |
|
td.set_commands(**{key: value for (key, value) in command}) |
204
|
1 |
|
td.set_environment(env) |
205
|
|
|
|
206
|
1 |
|
if diff: |
207
|
1 |
|
print_diff(td, 'Using task definition: %s' % task) |
208
|
|
|
|
209
|
1 |
|
action.run(td, count, 'ECS Deploy') |
210
|
|
|
|
211
|
1 |
|
click.secho( |
212
|
|
|
'Successfully started %d instances of task: %s' % ( |
213
|
|
|
len(action.started_tasks), |
214
|
|
|
td.family_revision |
215
|
|
|
), |
216
|
|
|
fg='green' |
217
|
|
|
) |
218
|
|
|
|
219
|
1 |
|
for started_task in action.started_tasks: |
220
|
1 |
|
click.secho('- %s' % started_task['taskArn'], fg='green') |
221
|
1 |
|
click.secho(' ') |
222
|
|
|
|
223
|
1 |
|
except Exception as e: |
224
|
1 |
|
click.secho('%s\n' % str(e), fg='red', err=True) |
225
|
1 |
|
exit(1) |
226
|
|
|
|
227
|
|
|
|
228
|
1 |
|
def wait_for_finish(action, timeout, title, success_message, failure_message, |
229
|
|
|
ignore_warnings): |
230
|
1 |
|
click.secho(title, nl=False) |
231
|
1 |
|
waiting = True |
232
|
1 |
|
waiting_timeout = datetime.now() + timedelta(seconds=timeout) |
233
|
1 |
|
service = action.get_service() |
234
|
1 |
|
inspected_until = None |
235
|
1 |
|
while waiting and datetime.now() < waiting_timeout: |
236
|
1 |
|
click.secho('.', nl=False) |
237
|
1 |
|
service = action.get_service() |
238
|
1 |
|
inspected_until = inspect_errors( |
239
|
|
|
service=service, |
240
|
|
|
failure_message=failure_message, |
241
|
|
|
ignore_warnings=ignore_warnings, |
242
|
|
|
since=inspected_until, |
243
|
|
|
timeout=False |
244
|
|
|
) |
245
|
1 |
|
waiting = not action.is_deployed(service) |
246
|
|
|
|
247
|
1 |
|
if waiting: |
248
|
1 |
|
sleep(1) |
249
|
|
|
|
250
|
1 |
|
inspect_errors( |
251
|
|
|
service=service, |
252
|
|
|
failure_message=failure_message, |
253
|
|
|
ignore_warnings=ignore_warnings, |
254
|
|
|
since=inspected_until, |
255
|
|
|
timeout=waiting |
256
|
|
|
) |
257
|
|
|
|
258
|
1 |
|
click.secho('\n%s\n' % success_message, fg='green') |
259
|
|
|
|
260
|
|
|
|
261
|
1 |
|
def deploy_task_definition(deployment, task_definition): |
262
|
1 |
|
click.secho('Updating service') |
263
|
1 |
|
deployment.deploy(task_definition) |
264
|
|
|
|
265
|
1 |
|
message = 'Successfully changed task definition to: %s:%s\n' % ( |
266
|
|
|
task_definition.family, |
267
|
|
|
task_definition.revision |
268
|
|
|
) |
269
|
|
|
|
270
|
1 |
|
click.secho(message, fg='green') |
271
|
|
|
|
272
|
|
|
|
273
|
1 |
|
def get_task_definition(action, task): |
274
|
1 |
|
if task: |
275
|
1 |
|
task_definition = action.get_task_definition(task) |
276
|
1 |
|
click.secho('Deploying based on task definition: %s' % task) |
277
|
|
|
else: |
278
|
1 |
|
task_definition = action.get_current_task_definition(action.service) |
279
|
|
|
|
280
|
1 |
|
return task_definition |
281
|
|
|
|
282
|
|
|
|
283
|
1 |
|
def create_task_definition(action, task_definition): |
284
|
1 |
|
click.secho('Creating new task definition revision') |
285
|
1 |
|
new_td = action.update_task_definition(task_definition) |
286
|
|
|
|
287
|
1 |
|
click.secho( |
288
|
|
|
'Successfully created revision: %d\n' % new_td.revision, |
289
|
|
|
fg='green' |
290
|
|
|
) |
291
|
|
|
|
292
|
1 |
|
return new_td |
293
|
|
|
|
294
|
|
|
|
295
|
1 |
|
def deregister_task_definition(action, task_definition): |
296
|
1 |
|
click.secho('Deregister old task definition revision') |
297
|
1 |
|
action.deregister_task_definition(task_definition) |
298
|
1 |
|
click.secho( |
299
|
|
|
'Successfully deregistered revision: %d\n' % task_definition.revision, |
300
|
|
|
fg='green' |
301
|
|
|
) |
302
|
|
|
|
303
|
|
|
|
304
|
1 |
|
def record_deployment(revision, api_key, app_id, comment, user): |
305
|
1 |
|
api_key = getenv('NEW_RELIC_API_KEY', api_key) |
306
|
1 |
|
app_id = getenv('NEW_RELIC_APP_ID', app_id) |
307
|
|
|
|
308
|
1 |
|
if not revision or not api_key or not app_id: |
309
|
1 |
|
return False |
310
|
|
|
|
311
|
1 |
|
user = user or getpass.getuser() |
312
|
|
|
|
313
|
1 |
|
click.secho('Recording deployment in New Relic', nl=False) |
314
|
|
|
|
315
|
1 |
|
deployment = Deployment(api_key, app_id, user) |
316
|
1 |
|
deployment.deploy(revision, '', comment) |
317
|
|
|
|
318
|
1 |
|
click.secho('\nDone\n', fg='green') |
319
|
|
|
|
320
|
1 |
|
return True |
321
|
|
|
|
322
|
|
|
|
323
|
1 |
|
def print_diff(task_definition, title='Updating task definition'): |
324
|
1 |
|
if task_definition.diff: |
325
|
1 |
|
click.secho(title) |
326
|
1 |
|
for diff in task_definition.diff: |
327
|
1 |
|
click.secho(str(diff), fg='blue') |
328
|
1 |
|
click.secho('') |
329
|
|
|
|
330
|
|
|
|
331
|
1 |
|
def inspect_errors(service, failure_message, ignore_warnings, since, timeout): |
332
|
1 |
|
error = False |
333
|
1 |
|
last_error_timestamp = since |
334
|
|
|
|
335
|
1 |
|
warnings = service.get_warnings(since) |
336
|
1 |
|
for timestamp in warnings: |
337
|
1 |
|
message = warnings[timestamp] |
338
|
1 |
|
click.secho('') |
339
|
1 |
|
if ignore_warnings: |
340
|
1 |
|
last_error_timestamp = timestamp |
341
|
1 |
|
click.secho( |
342
|
|
|
text='%s\nWARNING: %s' % (timestamp, message), |
343
|
|
|
fg='yellow', |
344
|
|
|
err=False |
345
|
|
|
) |
346
|
1 |
|
click.secho('Continuing.', nl=False) |
347
|
|
|
else: |
348
|
1 |
|
click.secho( |
349
|
|
|
text='%s\nERROR: %s\n' % (timestamp, message), |
350
|
|
|
fg='red', |
351
|
|
|
err=True |
352
|
|
|
) |
353
|
1 |
|
error = True |
354
|
|
|
|
355
|
1 |
|
if service.older_errors: |
356
|
1 |
|
click.secho('') |
357
|
1 |
|
click.secho('Older errors', fg='yellow', err=True) |
358
|
1 |
|
for timestamp in service.older_errors: |
359
|
1 |
|
click.secho( |
360
|
|
|
text='%s\n%s\n' % (timestamp, service.older_errors[timestamp]), |
361
|
|
|
fg='yellow', |
362
|
|
|
err=True |
363
|
|
|
) |
364
|
|
|
|
365
|
1 |
|
if timeout: |
366
|
1 |
|
error = True |
367
|
1 |
|
failure_message += ' (timeout)' |
368
|
|
|
|
369
|
1 |
|
if error: |
370
|
1 |
|
raise TaskPlacementError(failure_message) |
371
|
|
|
|
372
|
1 |
|
return last_error_timestamp |
373
|
|
|
|
374
|
|
|
|
375
|
1 |
|
ecs.add_command(deploy) |
376
|
1 |
|
ecs.add_command(scale) |
377
|
1 |
|
ecs.add_command(run) |
378
|
|
|
|
379
|
|
|
if __name__ == '__main__': # pragma: no cover |
380
|
|
|
ecs() |
381
|
|
|
|