ob-metaflow-extensions 1.1.170__py2.py3-none-any.whl → 1.4.35__py2.py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Potentially problematic release.


This version of ob-metaflow-extensions might be problematic. Click here for more details.

Files changed (65) hide show
  1. metaflow_extensions/outerbounds/plugins/__init__.py +6 -2
  2. metaflow_extensions/outerbounds/plugins/apps/app_cli.py +0 -0
  3. metaflow_extensions/outerbounds/plugins/apps/app_deploy_decorator.py +146 -0
  4. metaflow_extensions/outerbounds/plugins/apps/core/__init__.py +10 -0
  5. metaflow_extensions/outerbounds/plugins/apps/core/_state_machine.py +506 -0
  6. metaflow_extensions/outerbounds/plugins/apps/core/_vendor/__init__.py +0 -0
  7. metaflow_extensions/outerbounds/plugins/apps/core/_vendor/spinner/__init__.py +4 -0
  8. metaflow_extensions/outerbounds/plugins/apps/core/_vendor/spinner/spinners.py +478 -0
  9. metaflow_extensions/outerbounds/plugins/apps/core/app_cli.py +1200 -0
  10. metaflow_extensions/outerbounds/plugins/apps/core/app_config.py +146 -0
  11. metaflow_extensions/outerbounds/plugins/apps/core/artifacts.py +0 -0
  12. metaflow_extensions/outerbounds/plugins/apps/core/capsule.py +958 -0
  13. metaflow_extensions/outerbounds/plugins/apps/core/click_importer.py +24 -0
  14. metaflow_extensions/outerbounds/plugins/apps/core/code_package/__init__.py +3 -0
  15. metaflow_extensions/outerbounds/plugins/apps/core/code_package/code_packager.py +618 -0
  16. metaflow_extensions/outerbounds/plugins/apps/core/code_package/examples.py +125 -0
  17. metaflow_extensions/outerbounds/plugins/apps/core/config/__init__.py +12 -0
  18. metaflow_extensions/outerbounds/plugins/apps/core/config/cli_generator.py +161 -0
  19. metaflow_extensions/outerbounds/plugins/apps/core/config/config_utils.py +868 -0
  20. metaflow_extensions/outerbounds/plugins/apps/core/config/schema_export.py +288 -0
  21. metaflow_extensions/outerbounds/plugins/apps/core/config/typed_configs.py +139 -0
  22. metaflow_extensions/outerbounds/plugins/apps/core/config/typed_init_generator.py +398 -0
  23. metaflow_extensions/outerbounds/plugins/apps/core/config/unified_config.py +1088 -0
  24. metaflow_extensions/outerbounds/plugins/apps/core/config_schema.yaml +337 -0
  25. metaflow_extensions/outerbounds/plugins/apps/core/dependencies.py +115 -0
  26. metaflow_extensions/outerbounds/plugins/apps/core/deployer.py +303 -0
  27. metaflow_extensions/outerbounds/plugins/apps/core/experimental/__init__.py +89 -0
  28. metaflow_extensions/outerbounds/plugins/apps/core/perimeters.py +87 -0
  29. metaflow_extensions/outerbounds/plugins/apps/core/secrets.py +164 -0
  30. metaflow_extensions/outerbounds/plugins/apps/core/utils.py +233 -0
  31. metaflow_extensions/outerbounds/plugins/apps/core/validations.py +17 -0
  32. metaflow_extensions/outerbounds/plugins/aws/assume_role_decorator.py +25 -12
  33. metaflow_extensions/outerbounds/plugins/checkpoint_datastores/coreweave.py +9 -77
  34. metaflow_extensions/outerbounds/plugins/checkpoint_datastores/external_chckpt.py +85 -0
  35. metaflow_extensions/outerbounds/plugins/checkpoint_datastores/nebius.py +7 -78
  36. metaflow_extensions/outerbounds/plugins/fast_bakery/baker.py +110 -0
  37. metaflow_extensions/outerbounds/plugins/fast_bakery/docker_environment.py +6 -2
  38. metaflow_extensions/outerbounds/plugins/fast_bakery/fast_bakery.py +1 -0
  39. metaflow_extensions/outerbounds/plugins/nvct/nvct_decorator.py +8 -8
  40. metaflow_extensions/outerbounds/plugins/optuna/__init__.py +48 -0
  41. metaflow_extensions/outerbounds/plugins/profilers/simple_card_decorator.py +96 -0
  42. metaflow_extensions/outerbounds/plugins/s3_proxy/__init__.py +7 -0
  43. metaflow_extensions/outerbounds/plugins/s3_proxy/binary_caller.py +132 -0
  44. metaflow_extensions/outerbounds/plugins/s3_proxy/constants.py +11 -0
  45. metaflow_extensions/outerbounds/plugins/s3_proxy/exceptions.py +13 -0
  46. metaflow_extensions/outerbounds/plugins/s3_proxy/proxy_bootstrap.py +59 -0
  47. metaflow_extensions/outerbounds/plugins/s3_proxy/s3_proxy_api.py +93 -0
  48. metaflow_extensions/outerbounds/plugins/s3_proxy/s3_proxy_decorator.py +250 -0
  49. metaflow_extensions/outerbounds/plugins/s3_proxy/s3_proxy_manager.py +225 -0
  50. metaflow_extensions/outerbounds/plugins/snowpark/snowpark_client.py +6 -3
  51. metaflow_extensions/outerbounds/plugins/snowpark/snowpark_decorator.py +13 -7
  52. metaflow_extensions/outerbounds/plugins/snowpark/snowpark_job.py +8 -2
  53. metaflow_extensions/outerbounds/plugins/torchtune/__init__.py +4 -0
  54. metaflow_extensions/outerbounds/plugins/vllm/__init__.py +173 -95
  55. metaflow_extensions/outerbounds/plugins/vllm/status_card.py +9 -9
  56. metaflow_extensions/outerbounds/plugins/vllm/vllm_manager.py +159 -9
  57. metaflow_extensions/outerbounds/remote_config.py +8 -3
  58. metaflow_extensions/outerbounds/toplevel/global_aliases_for_metaflow_package.py +63 -1
  59. metaflow_extensions/outerbounds/toplevel/ob_internal.py +3 -0
  60. metaflow_extensions/outerbounds/toplevel/plugins/optuna/__init__.py +1 -0
  61. metaflow_extensions/outerbounds/toplevel/s3_proxy.py +88 -0
  62. {ob_metaflow_extensions-1.1.170.dist-info → ob_metaflow_extensions-1.4.35.dist-info}/METADATA +2 -2
  63. {ob_metaflow_extensions-1.1.170.dist-info → ob_metaflow_extensions-1.4.35.dist-info}/RECORD +65 -21
  64. {ob_metaflow_extensions-1.1.170.dist-info → ob_metaflow_extensions-1.4.35.dist-info}/WHEEL +0 -0
  65. {ob_metaflow_extensions-1.1.170.dist-info → ob_metaflow_extensions-1.4.35.dist-info}/top_level.txt +0 -0
@@ -0,0 +1,303 @@
1
+ from .config import TypedCoreConfig, TypedDict
2
+ from .perimeters import PerimeterExtractor
3
+ from .capsule import CapsuleApi
4
+ import json
5
+ from ._state_machine import DEPLOYMENT_READY_CONDITIONS, LogLine
6
+ from .app_config import AppConfig, AppConfigError
7
+ from .capsule import CapsuleDeployer, list_and_filter_capsules
8
+ from functools import partial
9
+ import sys
10
+ import uuid
11
+ from typing import Type, Dict, List
12
+ from datetime import datetime
13
+
14
+
15
+ class AppDeployer(TypedCoreConfig):
16
+ """ """
17
+
18
+ __init__ = TypedCoreConfig.__init__
19
+
20
+ _app_config: AppConfig
21
+
22
+ _state = {}
23
+
24
+ __state_items = [
25
+ "perimeter",
26
+ "api_url",
27
+ "code_package_url",
28
+ "code_package_key",
29
+ "image",
30
+ "project",
31
+ "branch",
32
+ ]
33
+
34
+ @property
35
+ def _deploy_config(self) -> AppConfig:
36
+ if not hasattr(self, "_app_config"):
37
+ self._app_config = AppConfig(self._config)
38
+ return self._app_config
39
+
40
+ # Things that need to be set before deploy
41
+ @classmethod
42
+ def _set_state(
43
+ cls,
44
+ perimeter: str,
45
+ api_url: str,
46
+ code_package_url: str = None,
47
+ code_package_key: str = None,
48
+ name_prefix: str = None,
49
+ image: str = None,
50
+ max_entropy: int = 4,
51
+ default_tags: List[Dict[str, str]] = None,
52
+ project: str = None,
53
+ branch: str = None,
54
+ ):
55
+ cls._state["perimeter"] = perimeter
56
+ cls._state["api_url"] = api_url
57
+ cls._state["code_package_url"] = code_package_url
58
+ cls._state["code_package_key"] = code_package_key
59
+ cls._state["name_prefix"] = name_prefix
60
+ cls._state["image"] = image
61
+ cls._state["max_entropy"] = max_entropy
62
+ cls._state["default_tags"] = default_tags
63
+ cls._state["project"] = project
64
+ cls._state["branch"] = branch
65
+
66
+ assert (
67
+ max_entropy > 0
68
+ ), "max_entropy must be greater than 0. Since AppDeployer's deploy fn can be called many time inside a step itself."
69
+
70
+ def deploy(
71
+ self,
72
+ readiness_condition=DEPLOYMENT_READY_CONDITIONS.ATLEAST_ONE_RUNNING,
73
+ max_wait_time=600,
74
+ readiness_wait_time=10,
75
+ logger_fn=partial(print, file=sys.stderr),
76
+ status_file=None,
77
+ no_loader=False,
78
+ **kwargs,
79
+ ) -> "DeployedApp":
80
+
81
+ # Name setting from top level if none is set in the code
82
+ if self._deploy_config._core_config.name is None:
83
+ name = self._state[
84
+ "name_prefix"
85
+ ] # for now the name-prefix cannot be very large.
86
+ entropy = uuid.uuid4().hex[: self._state["max_entropy"]]
87
+ self._deploy_config._core_config.name = f"{name}-{entropy}"
88
+
89
+ if len(self._state["default_tags"]) > 0:
90
+ self._deploy_config._core_config.tags = (
91
+ self._deploy_config._core_config.tags or []
92
+ ) + self._state["default_tags"]
93
+
94
+ self._deploy_config.commit()
95
+ # Set any state that might have been passed down from the top level
96
+ for k in self.__state_items:
97
+ if self._deploy_config.get_state(k) is None:
98
+ self._deploy_config.set_state(k, self._state[k])
99
+
100
+ capsule = CapsuleDeployer(
101
+ self._deploy_config,
102
+ self._state["api_url"],
103
+ create_timeout=max_wait_time,
104
+ debug_dir=None,
105
+ success_terminal_state_condition=readiness_condition,
106
+ readiness_wait_time=readiness_wait_time,
107
+ logger_fn=logger_fn,
108
+ )
109
+
110
+ currently_present_capsules = list_and_filter_capsules(
111
+ capsule.capsule_api,
112
+ None,
113
+ None,
114
+ capsule.name,
115
+ None,
116
+ None,
117
+ None,
118
+ )
119
+
120
+ force_upgrade = self._deploy_config.get_state("force_upgrade", False)
121
+
122
+ if len(currently_present_capsules) > 0:
123
+ # Only update the capsule if there is no upgrade in progress
124
+ # Only update a "already updating" capsule if the `--force-upgrade` flag is provided.
125
+ _curr_cap = currently_present_capsules[0]
126
+ this_capsule_is_being_updated = _curr_cap.get("status", {}).get(
127
+ "updateInProgress", False
128
+ )
129
+
130
+ if this_capsule_is_being_updated and not force_upgrade:
131
+ _upgrader = _curr_cap.get("metadata", {}).get("lastModifiedBy", None)
132
+ message = f"{capsule.capsule_type} is currently being upgraded"
133
+ if _upgrader:
134
+ message = (
135
+ f"{capsule.capsule_type} is currently being upgraded. Upgrade was launched by {_upgrader}. "
136
+ "If you wish to force upgrade, you can do so by providing the `--force-upgrade` flag."
137
+ )
138
+ raise AppConfigError(message)
139
+
140
+ logger_fn(
141
+ f"🚀 {'' if not force_upgrade else 'Force'} Upgrading {capsule.capsule_type.lower()} `{capsule.name}`....",
142
+ )
143
+ else:
144
+ logger_fn(
145
+ f"🚀 Deploying {capsule.capsule_type.lower()} `{capsule.name}`....",
146
+ )
147
+
148
+ capsule.create()
149
+ final_status = capsule.wait_for_terminal_state()
150
+ return DeployedApp(
151
+ final_status["id"],
152
+ final_status["auth_type"],
153
+ final_status["public_url"],
154
+ final_status["name"],
155
+ final_status["deployed_version"],
156
+ final_status["deployed_at"],
157
+ )
158
+
159
+
160
+ class DeployedApp:
161
+ def __init__(
162
+ self,
163
+ _id: str,
164
+ capsule_type: str,
165
+ public_url: str,
166
+ name: str,
167
+ deployed_version: str,
168
+ deployed_at: str,
169
+ ):
170
+ self._id = _id
171
+ self._capsule_type = capsule_type
172
+ self._public_url = public_url
173
+ self._name = name
174
+ self._deployed_version = deployed_version
175
+ self._deployed_at = deployed_at
176
+
177
+ def _get_capsule_api(self) -> CapsuleApi:
178
+ perimeter, api_server = PerimeterExtractor.during_metaflow_execution()
179
+ return CapsuleApi(api_server, perimeter)
180
+
181
+ def logs(self, previous=False) -> Dict[str, List[LogLine]]:
182
+ """
183
+ Returns a dictionary of worker_id to logs.
184
+ If `previous` is True, it will return the logs from the previous execution of the workers. Useful when debugging a crashlooping worker.
185
+ """
186
+ capsule_api = self._get_capsule_api()
187
+ # extract workers from capsule
188
+ workers = capsule_api.get_workers(self._id)
189
+ # get logs from workers
190
+ logs = {
191
+ # worker_id: logs
192
+ }
193
+ for worker in workers:
194
+ # TODO: Handle exceptions better over here.
195
+ logs[worker["workerId"]] = capsule_api.logs(
196
+ self._id, worker["workerId"], previous=previous
197
+ )
198
+ return logs
199
+
200
+ def info(self) -> dict:
201
+ """
202
+ Returns a dictionary representing the deployed app.
203
+ """
204
+ capsule_api = self._get_capsule_api()
205
+ capsule = capsule_api.get(self._id)
206
+ return capsule
207
+
208
+ def replicas(self):
209
+ capsule_api = self._get_capsule_api()
210
+ return capsule_api.get_workers(self._id)
211
+
212
+ def scale_to_zero(self):
213
+ """
214
+ Scales the DeployedApp to 0 replicas.
215
+ """
216
+ capsule_api = self._get_capsule_api()
217
+ return capsule_api.patch(
218
+ self._id,
219
+ {
220
+ "autoscalingConfig": {
221
+ "minReplicas": 0,
222
+ "maxReplicas": 0,
223
+ }
224
+ },
225
+ )
226
+
227
+ def delete(self):
228
+ """
229
+ Deletes the DeployedApp.
230
+ """
231
+ capsule_api = self._get_capsule_api()
232
+ return capsule_api.delete(self._id)
233
+
234
+ @property
235
+ def id(self) -> str:
236
+ return self._id
237
+
238
+ @property
239
+ def auth_style(self) -> str:
240
+ # TODO : Fix naming here.
241
+ return self._capsule_type
242
+
243
+ @property
244
+ def public_url(self) -> str:
245
+ return self._public_url
246
+
247
+ @property
248
+ def name(self) -> str:
249
+ return self._name
250
+
251
+ @property
252
+ def deployed_version(self) -> str:
253
+ return self._deployed_version
254
+
255
+ def to_dict(self) -> dict:
256
+ return {
257
+ "id": self._id,
258
+ "auth_style": self.auth_style, # TODO : Fix naming here.
259
+ "public_url": self._public_url,
260
+ "name": self._name,
261
+ "deployed_version": self._deployed_version,
262
+ "deployed_at": self._deployed_at,
263
+ }
264
+
265
+ @classmethod
266
+ def from_dict(cls, data: dict):
267
+ return cls(
268
+ _id=data["id"],
269
+ capsule_type=data["capsule_type"],
270
+ public_url=data["public_url"],
271
+ name=data["name"],
272
+ deployed_version=data["deployed_version"],
273
+ deployed_at=data["deployed_at"],
274
+ )
275
+
276
+ @property
277
+ def deployed_at(self) -> datetime:
278
+ return datetime.fromisoformat(self._deployed_at)
279
+
280
+ def __repr__(self) -> str:
281
+ return (
282
+ f"DeployedApp(id='{self._id}', "
283
+ f"name='{self._name}', "
284
+ f"public_url='{self._public_url}', "
285
+ f"deployed_version='{self._deployed_version}')"
286
+ )
287
+
288
+
289
+ class apps:
290
+
291
+ _name_prefix = None
292
+
293
+ @classmethod
294
+ def set_name_prefix(cls, name_prefix: str):
295
+ cls._name_prefix = name_prefix
296
+
297
+ @property
298
+ def name_prefix(self) -> str:
299
+ return self._name_prefix
300
+
301
+ @property
302
+ def Deployer(self) -> Type[AppDeployer]:
303
+ return AppDeployer
@@ -0,0 +1,89 @@
1
+ import os
2
+ from typing import TYPE_CHECKING
3
+
4
+ if TYPE_CHECKING:
5
+ from ..app_config import AppConfig
6
+
7
+ DEFAULT_BRANCH = "test"
8
+
9
+
10
+ # Account for project / branch and the capsule input.
11
+ def capsule_input_overrides(app_config: "AppConfig", capsule_input: dict):
12
+ project = app_config.get_state("project", None)
13
+ # Update the project/branch related configurations.
14
+ if project is not None:
15
+ branch = app_config.get_state("branch", DEFAULT_BRANCH)
16
+ capsule_input["tags"].extend(
17
+ [dict(key="project", value=project), dict(key="branch", value=branch)]
18
+ )
19
+
20
+ persistence = app_config.get_state("persistence", None)
21
+ if persistence is not None and persistence != "none":
22
+ capsule_input["persistence"] = persistence
23
+
24
+ capsule_input["generateStaticUrl"] = app_config.get_state(
25
+ "generate_static_url", False
26
+ )
27
+
28
+ model_asset_conf = app_config.get_state("models", None)
29
+ data_asset_conf = app_config.get_state("data", None)
30
+ code_info = _code_info(app_config)
31
+ # todo:fix me
32
+ _objects_key = "associatedObjects"
33
+ if model_asset_conf or data_asset_conf or code_info:
34
+ capsule_input[_objects_key] = {}
35
+
36
+ if model_asset_conf:
37
+ capsule_input[_objects_key]["models"] = [
38
+ {"assetId": x["asset_id"], "assetInstanceId": x["asset_instance_id"]}
39
+ for x in model_asset_conf
40
+ ]
41
+ if data_asset_conf:
42
+ capsule_input[_objects_key]["data"] = [
43
+ {"assetId": x["asset_id"], "assetInstanceId": x["asset_instance_id"]}
44
+ for x in data_asset_conf
45
+ ]
46
+ if code_info:
47
+ capsule_input[_objects_key]["code"] = code_info
48
+
49
+ return capsule_input
50
+
51
+
52
+ def _code_info(app_config: "AppConfig"):
53
+ from metaflow.metaflow_git import get_repository_info
54
+
55
+ try:
56
+ from metaflow.metaflow_git import _call_git # type: ignore
57
+ except ImportError:
58
+ # Fallback if _call_git is not available
59
+ def _call_git(args, path=None):
60
+ return "", 1, True
61
+
62
+ package_dirs = app_config.get_state("packaging_directories")
63
+ if package_dirs is None:
64
+ return None
65
+ for directory in package_dirs:
66
+ repo_info = get_repository_info(directory)
67
+ if len(repo_info) == 0:
68
+ continue
69
+ git_log_info, returncode, failed = _call_git(
70
+ ["log", "-1", "--pretty=%B"],
71
+ path=directory,
72
+ )
73
+ repo_url = repo_info["repo_url"]
74
+ if isinstance(repo_url, str):
75
+ _url = (
76
+ repo_url if not repo_url.endswith(".git") else repo_url.rstrip(".git")
77
+ )
78
+ else:
79
+ _url = str(repo_url)
80
+ _code_info = {
81
+ "commitId": repo_info["commit_sha"],
82
+ "commitLink": os.path.join(_url, "commit", str(repo_info["commit_sha"])),
83
+ }
84
+ if not failed and returncode == 0 and isinstance(git_log_info, str):
85
+ _code_info["commitMessage"] = git_log_info.strip()
86
+
87
+ return _code_info
88
+
89
+ return None
@@ -0,0 +1,87 @@
1
+ import os
2
+ import json
3
+ from typing import Tuple, Union
4
+
5
+
6
+ class PerimeterExtractor:
7
+ @classmethod
8
+ def for_ob_cli(
9
+ cls, config_dir: str, profile: str
10
+ ) -> Union[Tuple[str, str], Tuple[None, None]]:
11
+ """
12
+ This function will be called when we are trying to extract the perimeter
13
+ via the ob cli's execution. We will rely on the following logic:
14
+ 1. check environment variables like OB_CURRENT_PERIMETER / OBP_PERIMETER
15
+ 2. run init config to extract the perimeter related configurations.
16
+
17
+ Returns
18
+ -------
19
+ Tuple[str, str] : Tuple containing perimeter name , API server url.
20
+ """
21
+ from outerbounds.utils import metaflowconfig
22
+
23
+ perimeter = None
24
+ api_server = None
25
+ if os.environ.get("OB_CURRENT_PERIMETER") or os.environ.get("OBP_PERIMETER"):
26
+ perimeter = os.environ.get("OB_CURRENT_PERIMETER") or os.environ.get(
27
+ "OBP_PERIMETER"
28
+ )
29
+
30
+ if os.environ.get("OBP_API_SERVER"):
31
+ api_server = os.environ.get("OBP_API_SERVER")
32
+
33
+ if perimeter is None or api_server is None:
34
+ metaflow_config = metaflowconfig.init_config(config_dir, profile)
35
+ perimeter = metaflow_config.get("OBP_PERIMETER")
36
+ api_server = metaflowconfig.get_sanitized_url_from_config(
37
+ config_dir, profile, "OBP_API_SERVER"
38
+ )
39
+
40
+ return perimeter, api_server # type: ignore
41
+
42
+ @classmethod
43
+ def during_metaflow_execution(cls) -> Union[Tuple[str, str], Tuple[None, None]]:
44
+ from metaflow.metaflow_config_funcs import init_config
45
+
46
+ clean_url = (
47
+ lambda url: f"https://{url}".rstrip("/")
48
+ if not url.startswith("https://")
49
+ else url
50
+ )
51
+
52
+ config = init_config()
53
+ api_server, perimeter, integrations_url = None, None, None
54
+ perimeter = config.get(
55
+ "OBP_PERIMETER", os.environ.get("OBP_PERIMETER", perimeter)
56
+ )
57
+ if perimeter is None:
58
+ raise RuntimeError(
59
+ "Perimeter not found in metaflow config or environment variables"
60
+ )
61
+
62
+ api_server = config.get(
63
+ "OBP_API_SERVER", os.environ.get("OBP_API_SERVER", api_server)
64
+ )
65
+
66
+ if api_server is not None and not api_server.startswith("https://"):
67
+ api_server = clean_url(api_server)
68
+
69
+ if api_server is not None:
70
+ return perimeter, api_server
71
+
72
+ integrations_url = config.get(
73
+ "OBP_INTEGRATIONS_URL", os.environ.get("OBP_INTEGRATIONS_URL", None)
74
+ )
75
+
76
+ if integrations_url is not None and not integrations_url.startswith("https://"):
77
+ integrations_url = clean_url(integrations_url)
78
+
79
+ if integrations_url is not None:
80
+ api_server = integrations_url.rstrip("/integrations")
81
+
82
+ if api_server is None:
83
+ raise RuntimeError(
84
+ "API server not found in metaflow config or environment variables"
85
+ )
86
+
87
+ return perimeter, api_server
@@ -0,0 +1,164 @@
1
+ from typing import Dict
2
+
3
+ import base64
4
+ import json
5
+ import requests
6
+ import random
7
+ import time
8
+ import sys
9
+
10
+ from .utils import safe_requests_wrapper, TODOException
11
+
12
+
13
+ class OuterboundsSecretsException(Exception):
14
+ pass
15
+
16
+
17
+ class SecretNotFound(OuterboundsSecretsException):
18
+ pass
19
+
20
+
21
+ class OuterboundsSecretsApiResponse:
22
+ def __init__(self, response):
23
+ self.response = response
24
+
25
+ @property
26
+ def secret_resource_id(self):
27
+ return self.response["secret_resource_id"]
28
+
29
+ @property
30
+ def secret_backend_type(self):
31
+ return self.response["secret_backend_type"]
32
+
33
+
34
+ class SecretRetriever:
35
+ def get_secret_as_dict(self, secret_id, options={}, role=None):
36
+ """
37
+ Supports a special way of specifying secrets sources in outerbounds using the format:
38
+ @secrets(sources=["outerbounds.<integrations_name>"])
39
+
40
+ When invoked it makes a requests to the integrations secrets metadata endpoint on the
41
+ keywest server to get the cloud resource id for a secret. It then uses that to invoke
42
+ secrets manager on the core oss and returns the secrets.
43
+ """
44
+ headers = {"Content-Type": "application/json", "Connection": "keep-alive"}
45
+ perimeter, integrations_url = self._get_secret_configs()
46
+ integration_name = secret_id
47
+ request_payload = {
48
+ "perimeter_name": perimeter,
49
+ "integration_name": integration_name,
50
+ }
51
+ response = self._make_request(integrations_url, headers, request_payload)
52
+ secret_resource_id = response.secret_resource_id
53
+ secret_backend_type = response.secret_backend_type
54
+
55
+ from metaflow.plugins.secrets.secrets_decorator import (
56
+ get_secrets_backend_provider,
57
+ )
58
+
59
+ secrets_provider = get_secrets_backend_provider(secret_backend_type)
60
+ secret_dict = secrets_provider.get_secret_as_dict(
61
+ secret_resource_id, options={}, role=role
62
+ )
63
+
64
+ # Outerbounds stores secrets as binaries. Hence we expect the returned secret to be
65
+ # {<cloud-secret-name>: <base64 encoded full secret>}. We decode the secret here like:
66
+ # 1. decode the base64 encoded full secret
67
+ # 2. load the decoded secret as a json
68
+ # 3. decode the base64 encoded values in the dict
69
+ # 4. return the decoded dict
70
+ binary_secret = next(iter(secret_dict.values()))
71
+ return self._decode_secret(binary_secret)
72
+
73
+ def _is_base64_encoded(self, data):
74
+ try:
75
+ if isinstance(data, str):
76
+ # Check if the string can be base64 decoded
77
+ base64.b64decode(data).decode("utf-8")
78
+ return True
79
+ return False
80
+ except Exception:
81
+ return False
82
+
83
+ def _decode_secret(self, secret):
84
+ try:
85
+ result = {}
86
+ secret_str = secret
87
+ if self._is_base64_encoded(secret):
88
+ # we check if the secret string is base64 encoded because the returned secret from
89
+ # AWS secret manager is base64 encoded while the secret from GCP is not
90
+ secret_str = base64.b64decode(secret).decode("utf-8")
91
+
92
+ secret_dict = json.loads(secret_str)
93
+ for key, value in secret_dict.items():
94
+ result[key] = base64.b64decode(value).decode("utf-8")
95
+
96
+ return result
97
+ except Exception as e:
98
+ raise OuterboundsSecretsException(f"Error decoding secret: {e}")
99
+
100
+ def _get_secret_configs(self):
101
+ from metaflow_extensions.outerbounds.remote_config import init_config # type: ignore
102
+ from os import environ
103
+
104
+ conf = init_config()
105
+ if "OBP_PERIMETER" in conf:
106
+ perimeter = conf["OBP_PERIMETER"]
107
+ else:
108
+ # if the perimeter is not in metaflow config, try to get it from the environment
109
+ perimeter = environ.get("OBP_PERIMETER", "")
110
+
111
+ if "OBP_INTEGRATIONS_URL" in conf:
112
+ integrations_url = conf["OBP_INTEGRATIONS_URL"]
113
+ else:
114
+ # if the integrations is not in metaflow config, try to get it from the environment
115
+ integrations_url = environ.get("OBP_INTEGRATIONS_URL", "")
116
+
117
+ if not perimeter:
118
+ raise OuterboundsSecretsException(
119
+ "No perimeter set. Please make sure to run `outerbounds configure <...>` command which can be found on the Outerbounds UI or reach out to your Outerbounds support team."
120
+ )
121
+
122
+ if not integrations_url:
123
+ raise OuterboundsSecretsException(
124
+ "No integrations url set. Please notify your Outerbounds support team about this issue."
125
+ )
126
+
127
+ integrations_secrets_metadata_url = f"{integrations_url}/secrets/metadata"
128
+ return perimeter, integrations_secrets_metadata_url
129
+
130
+ def _make_request(self, url, headers: Dict, payload: Dict):
131
+ try:
132
+ from metaflow.metaflow_config import SERVICE_HEADERS
133
+
134
+ request_headers = {**headers, **(SERVICE_HEADERS or {})}
135
+ except ImportError:
136
+ raise OuterboundsSecretsException(
137
+ "Failed to create app: No Metaflow service headers found"
138
+ )
139
+
140
+ response = safe_requests_wrapper(
141
+ requests.get,
142
+ url,
143
+ data=json.dumps(payload),
144
+ headers=request_headers,
145
+ conn_error_retries=5,
146
+ retryable_status_codes=[409],
147
+ )
148
+ self._handle_error_response(response)
149
+ return OuterboundsSecretsApiResponse(response.json())
150
+
151
+ @staticmethod
152
+ def _handle_error_response(response: requests.Response):
153
+ if response.status_code >= 500:
154
+ raise OuterboundsSecretsException(
155
+ f"Server error: {response.text}. Please reach out to your Outerbounds support team."
156
+ )
157
+ status_code = response.status_code
158
+ if status_code == 404:
159
+ raise SecretNotFound(f"Secret not found: {response.text}")
160
+
161
+ if status_code >= 400:
162
+ raise OuterboundsSecretsException(
163
+ f"status_code={status_code}\t\n\t\t{response.text}"
164
+ )