/
wrapper.py
359 lines (296 loc) · 15.6 KB
/
wrapper.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
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
import logging
import os
from abc import ABC, abstractmethod
import boto3
import sagemaker
# noinspection PyProtectedMember
from sagemaker.estimator import _TrainingJob # need access to sagemaker internals to get last training job name
from sagemaker.multidatamodel import MultiDataModel
from sagemaker.processing import ProcessingInput, ScriptProcessor, FrameworkProcessor
from sagemaker.processing import ProcessingJob # Note: processing job is not marked as protected
from sagemaker.transformer import Transformer
# noinspection PyProtectedMember
from sagemaker.transformer import _TransformJob # need access to sagemaker internals to get last training job name
from sagemaker.sklearn import SKLearnProcessor
from sagemaker.spark import PySparkProcessor
from sagemaker_ssh_helper.log import SSHLog
from sagemaker_ssh_helper.manager import SSMManager
from sagemaker_ssh_helper.proxy import SSMProxy
class SSHEnvironmentWrapper(ABC):
logger = logging.getLogger('sagemaker-ssh-helper')
def __init__(self,
ssm_iam_role: str,
bootstrap_on_start: bool = True,
connection_wait_time_seconds: int = 600,
sagemaker_session: sagemaker.Session = None,
local_user_id: str = None,
log_to_stdout: bool = False):
f"""
:param ssm_iam_role: the SSM role without prefix, e.g. 'service-role/SageMakerRole'
See https://docs.aws.amazon.com/systems-manager/latest/userguide/sysman-managed-instance-activation.html .
:param bootstrap_on_start: Kick-off connection procedure upon sagemaker_ssh_helper.setup_and_start_ssh() .
:param connection_wait_time_seconds: How long to wait before a SageMaker entry point.
Can be 0 (don't wait).
"""
self.log_to_stdout = log_to_stdout
self.local_user_id = local_user_id
self.sagemaker_session = sagemaker_session or sagemaker.Session()
self.ssm_manager = SSMManager(region_name=self.sagemaker_session.boto_region_name)
self.ssh_log = SSHLog(region_name=self.sagemaker_session.boto_region_name)
if ssm_iam_role != '':
if self._is_arn(ssm_iam_role):
raise ValueError(f"ssm_iam_role should be only the part after role/, not a full ARN. "
f"Got: {ssm_iam_role}")
self.ssm_iam_role = ssm_iam_role
self.bootstrap_on_start = bootstrap_on_start
self.connection_wait_time_seconds = connection_wait_time_seconds
self.augmented = False
@classmethod
def dependency_dir(cls):
return os.path.dirname(__file__)
def _augment(self):
self.augmented = True
def _augment_env(self, env):
if self.local_user_id is None:
caller_id = boto3.client('sts').get_caller_identity()
user_id = caller_id.get('UserId')
else:
user_id = self.local_user_id
user_id_masked = list(user_id)
for i in range(3, len(user_id_masked) - 4):
user_id_masked[i] = '*'
user_id_masked = ''.join(user_id_masked)
self.logger.info(f"Passing '{user_id_masked}' as a value of the SSHOwner tag of an SSM managed instance")
env.update({'START_SSH': str(self.bootstrap_on_start).lower(),
'SSH_SSM_ROLE': self.ssm_iam_role,
'SSH_OWNER_TAG': user_id,
'SSH_LOG_TO_STDOUT': str(self.log_to_stdout).lower(),
'SSH_WAIT_TIME_SECONDS': f"{self.connection_wait_time_seconds}"})
@classmethod
def ssm_role_from_iam_arn(cls, iam_arn: str):
if not cls._is_arn(iam_arn):
raise ValueError(f"iam_arn should be a full ARN, got: '{iam_arn}'")
role_position = iam_arn.find(":role/")
if role_position == -1:
raise ValueError("':role/' not found in the iam_arn")
return iam_arn[role_position + 6:]
@abstractmethod
def get_instance_ids(self, retry=360):
f"""
:param retry: how many retries (each retry is 10 seconds), 360 is for 1 hour
"""
pass
def start_ssm_connection_and_continue(self, ssh_listen_port: int, retry: int = 360,
extra_args: str = ""):
proxy = self.start_ssm_connection(ssh_listen_port, retry, extra_args)
proxy.disconnect()
def start_ssm_connection(self, ssh_listen_port: int, retry: int = 360,
extra_args: str = "") -> SSMProxy:
instance_ids = self.get_instance_ids(retry)
if not instance_ids:
raise ValueError("instance_ids cannot be empty")
instance_id = instance_ids[0]
if "mi-" not in instance_id:
raise ValueError(f"instance_id doesn't start with 'mi-': {instance_id}")
ssm_proxy = SSMProxy(ssh_listen_port, extra_args, self.sagemaker_session.boto_region_name)
ssm_proxy.connect_to_ssm_instance(instance_id)
if self.connection_wait_time_seconds > 0:
ssm_proxy.terminate_waiting_loop()
return ssm_proxy
@staticmethod
def _is_arn(arn):
import re
return re.match(r'^arn:(aws|aws-cn|aws-us-gov):iam::([0-9]+):role/(\S+)$', arn)
class SSHEstimatorWrapper(SSHEnvironmentWrapper):
def __init__(self, estimator: sagemaker.estimator.EstimatorBase, ssm_iam_role: str = '',
bootstrap_on_start: bool = True, connection_wait_time_seconds: int = 600,
ssh_instance_count: int = 2, local_user_id: str = None,
log_to_stdout: bool = False):
super().__init__(ssm_iam_role, bootstrap_on_start, connection_wait_time_seconds,
estimator.sagemaker_session, local_user_id, log_to_stdout)
if estimator.instance_groups is not None:
# TODO: add support for heterogeneous clusters
self.logger.warning("Heterogeneous clusters are not yet supported, SSH Helper will start only on one node")
self.ssh_instance_count = 1
elif ssh_instance_count <= estimator.instance_count:
self.ssh_instance_count = ssh_instance_count
else:
self.ssh_instance_count = estimator.instance_count
if self.ssm_iam_role == '':
self.ssm_iam_role = SSHEnvironmentWrapper.ssm_role_from_iam_arn(estimator.role)
self.estimator = estimator
def _augment(self):
super()._augment()
self.logger.info(f'Turning on SSH to training job for estimator {self.estimator.__class__}')
env = self.estimator.environment
if env is None:
env = {}
self._augment_env(env)
# TODO: promote ssh_instance_count to processing/inference wrappers
env.update({'SSH_INSTANCE_COUNT': str(self.ssh_instance_count)})
self.estimator.environment = env
def get_instance_ids(self, retry=360):
training_job = self._latest_training_job()
return self.ssm_manager.get_training_instance_ids(training_job.name, retry * 10, self.ssh_instance_count)
def _latest_training_job(self):
training_job: _TrainingJob = self.estimator.latest_training_job
if training_job is None:
raise AssertionError("No training jobs found for estimator. Did you call estimator.fit() first?")
return training_job
def wait_training_job(self):
training_job = self._latest_training_job()
training_job.wait()
def stop_training_job(self):
training_job = self._latest_training_job()
training_job.stop()
training_job.wait()
@classmethod
def create(cls, estimator: sagemaker.estimator.EstimatorBase, connection_wait_time_seconds: int = 600,
ssh_instance_count: int = 2, local_user_id: str = None, log_to_stdout: bool = False):
# noinspection PyProtectedMember
if estimator._current_job_name:
raise AssertionError("You should call wrapper.create() before estimator.fit().")
result = SSHEstimatorWrapper(estimator, connection_wait_time_seconds=connection_wait_time_seconds,
ssh_instance_count=ssh_instance_count, local_user_id=local_user_id,
log_to_stdout=log_to_stdout)
result._augment()
return result
class SSHModelWrapper(SSHEnvironmentWrapper):
def __init__(self, model: sagemaker.model.Model,
ssm_iam_role: str = '',
bootstrap_on_start: bool = True, connection_wait_time_seconds: int = 600):
super().__init__(ssm_iam_role,
bootstrap_on_start, connection_wait_time_seconds, model.sagemaker_session)
if self.ssm_iam_role == '':
self.ssm_iam_role = SSHEnvironmentWrapper.ssm_role_from_iam_arn(model.role)
self.model = model
def _augment(self):
super()._augment()
self.logger.info(f'Turning on SSH to endpoint for model {self.model.__class__}')
env = self.model.env
if env is None:
env = {}
self._augment_env(env)
self.model.env = env
def get_instance_ids(self, retry=360):
return self.ssh_log.get_endpoint_ssm_instance_ids(self.model.endpoint_name, retry * 10)
def wait_for_endpoint(self):
self.sagemaker_session.wait_for_endpoint(self.model.endpoint_name)
@classmethod
def create(cls, model: sagemaker.model.Model, connection_wait_time_seconds: int = 600):
if model.endpoint_name:
raise AssertionError("You should call wrapper.create() before model.deploy().")
result = SSHModelWrapper(model, connection_wait_time_seconds=connection_wait_time_seconds)
result._augment()
return result
class SSHMultiModelWrapper(SSHEnvironmentWrapper):
def __init__(self, mdm: sagemaker.multidatamodel.MultiDataModel,
ssm_iam_role: str = '',
bootstrap_on_start: bool = True, connection_wait_time_seconds: int = 600):
super().__init__(ssm_iam_role,
bootstrap_on_start, connection_wait_time_seconds, mdm.sagemaker_session)
self.mdm = mdm
if mdm.model:
self.model = mdm.model
if self.ssm_iam_role == '':
self.ssm_iam_role = SSHEnvironmentWrapper.ssm_role_from_iam_arn(mdm.model.role)
self.model_wrapper = SSHModelWrapper(mdm.model, self.ssm_iam_role,
bootstrap_on_start,
connection_wait_time_seconds)
else:
self.model = None
if self.ssm_iam_role == '':
self.ssm_iam_role = SSHEnvironmentWrapper.ssm_role_from_iam_arn(mdm.role)
def _augment(self):
super()._augment()
if self.model:
# noinspection PyProtectedMember
self.model_wrapper._augment()
else:
self.logger.info(f'Turning on SSH to endpoint for multi data model {self.mdm.__class__}')
env = self.mdm.env
if env is None:
env = {}
self._augment_env(env)
self.mdm.env = env
def get_instance_ids(self, retry=360):
return self.ssh_log.get_endpoint_ssm_instance_ids(self.mdm.endpoint_name, retry * 10)
def wait_for_endpoint(self):
self.sagemaker_session.wait_for_endpoint(self.mdm.endpoint_name)
@classmethod
def create(cls, mdm: sagemaker.multidatamodel.MultiDataModel, connection_wait_time_seconds: int = 600):
if hasattr(mdm, 'endpoint_name') and mdm.endpoint_name:
raise AssertionError("You should call wrapper.create() before mdm.deploy().")
result = SSHMultiModelWrapper(mdm, connection_wait_time_seconds=connection_wait_time_seconds)
result._augment()
return result
class SSHProcessorWrapper(SSHEnvironmentWrapper):
def __init__(self, processor: sagemaker.processing.Processor,
ssm_iam_role: str = '',
bootstrap_on_start: bool = True,
connection_wait_time_seconds: int = 600):
super().__init__(ssm_iam_role, bootstrap_on_start, connection_wait_time_seconds,
processor.sagemaker_session)
if self.ssm_iam_role == '':
self.ssm_iam_role = SSHEnvironmentWrapper.ssm_role_from_iam_arn(processor.role)
self.processor = processor
def _augment(self):
super()._augment()
self.logger.info(f'Turning on SSH to processor {self.processor.__class__}')
env = self.processor.env
if env is None:
env = {}
self._augment_env(env)
self.processor.env = env
def get_instance_ids(self, retry=360):
job: ProcessingJob = self.processor.latest_job
return self.ssm_manager.get_processing_instance_ids(job.job_name, retry * 10)
def wait_processing_job(self):
job: ProcessingJob = self.processor.latest_job
job.wait()
def augmented_input(self):
f"""
Attaches the helper as the processing input. Required for processing jobs until the package is in PyPI.
Useful for processing jobs that don't support source_dir in run() method, e. g. {PySparkProcessor} and
{ScriptProcessor} / {SKLearnProcessor}
:return: a ProcessingInput to pass into processor#run(..., inputs=[...])
"""
if isinstance(self.processor, FrameworkProcessor):
self.logger.info("The processor {self.processor.__class__} is a subclass of FrameworkProcessor. "
"It's recommended to pass SageMaker SSH Helper as a dependency to the run() method "
"with dependencies=[SSHProcessorWrapper.dependency_dir()].")
return ProcessingInput(source=SSHProcessorWrapper.dependency_dir(),
destination='/opt/ml/processing/input/sagemaker_ssh_helper',
input_name='sagemaker_ssh_helper')
@classmethod
def create(cls, processor: sagemaker.processing.Processor, connection_wait_time_seconds: int = 600):
if processor.latest_job:
raise AssertionError("You should call wrapper.create() before processor.run()")
result = SSHProcessorWrapper(processor, connection_wait_time_seconds=connection_wait_time_seconds)
result._augment()
return result
class SSHTransformerWrapper(SSHEnvironmentWrapper):
def __init__(self, transformer: sagemaker.transformer.Transformer, model_wrapper: SSHModelWrapper):
super().__init__('', True, model_wrapper.connection_wait_time_seconds, transformer.sagemaker_session)
self.transformer = transformer
self.model_wrapper = model_wrapper
def _augment(self):
super()._augment()
def get_instance_ids(self, retry=360):
job: _TransformJob = self.transformer.latest_transform_job
return self.ssm_manager.get_transformer_instance_ids(job.job_name, retry * 10)
def wait_transform_job(self):
job: _TransformJob = self.transformer.latest_transform_job
job.wait()
@classmethod
def create(cls, transformer: sagemaker.transformer.Transformer, model_wrapper: SSHModelWrapper):
if not model_wrapper.augmented:
raise ValueError(f"Model Wrapper is not yet augmented. Consider constructing object with create().")
if model_wrapper.model.name != transformer.model_name:
raise ValueError(f"Transformer and model should have the same name, "
f"got: {transformer.model_name} and {transformer.model_name}")
if transformer.latest_transform_job:
raise AssertionError("You should call wrapper.create() before transformer.transform()")
result = SSHTransformerWrapper(transformer, model_wrapper)
result._augment()
return result