├── .gitignore
├── 00-api_gateway_managed_endpoint.ipynb
├── LICENSE
├── README.md
├── app.py
├── cdk.json
├── config
    ├── configs.json
    ├── example-configs.json
    └── sagemaker-invoke.json
├── construct
    ├── sagemaker_async_endpoint_construct.py
    └── sagemaker_endpoint_construct.py
├── docs
    └── diagram.png
├── functions
    ├── auth
    │   └── auth.py
    ├── falcon
    │   └── app.py
    ├── flan
    │   └── app.py
    ├── start_stop_endpoint
    │   └── app.py
    └── update_expiry
    │   └── app.py
├── requirements-dev.txt
├── requirements.txt
├── source.bat
├── stack
    ├── __init__.py
    ├── api_stack.py
    ├── endpoint_manager_stack.py
    ├── foundation_model_stack.py
    ├── lambda_stack.py
    ├── sagemaker_endpoint_manager_stack.py
    ├── stepfunction_stack.py
    └── util.py
├── tests
    ├── __init__.py
    └── unit
    │   ├── __init__.py
    │   └── test_sagemaker_jumpstart_generative_ai_app_stack.py
└── utils
    ├── __init__.py
    └── sagemaker_helper.py


/.gitignore:
--------------------------------------------------------------------------------
 1 | *.swp
 2 | package-lock.json
 3 | __pycache__
 4 | .pytest_cache
 5 | .venv
 6 | *.egg-info
 7 | 
 8 | # CDK asset staging directory
 9 | .cdk.staging
10 | cdk.out
11 | 


--------------------------------------------------------------------------------
/00-api_gateway_managed_endpoint.ipynb:
--------------------------------------------------------------------------------
  1 | {
  2 |  "cells": [
  3 |   {
  4 |    "attachments": {},
  5 |    "cell_type": "markdown",
  6 |    "metadata": {},
  7 |    "source": [
  8 |     "# Deploying a auto start/stop Amazon SageMaker Foundation Model endpoint backed by a API Gateway/Lambda"
  9 |    ]
 10 |   },
 11 |   {
 12 |    "attachments": {},
 13 |    "cell_type": "markdown",
 14 |    "metadata": {},
 15 |    "source": [
 16 |     "In this notebook, we will run through examples of managing your SageMaker endpoint with the endpoint manager functionality.\n",
 17 |     "\n",
 18 |     "We will also walkthrough an example on how to interact with the API gateway endpoint secured by a Lambda authoerizer."
 19 |    ]
 20 |   },
 21 |   {
 22 |    "cell_type": "code",
 23 |    "execution_count": null,
 24 |    "metadata": {},
 25 |    "outputs": [],
 26 |    "source": [
 27 |     "import requests\n",
 28 |     "import json\n",
 29 |     "import time"
 30 |    ]
 31 |   },
 32 |   {
 33 |    "attachments": {},
 34 |    "cell_type": "markdown",
 35 |    "metadata": {},
 36 |    "source": [
 37 |     "Set your API gateway URL and auth token\n",
 38 |     "\n",
 39 |     "`https://<API_GATEWAY_ID>.execute-api.<REGION>.amazonaws.com/prod/`\n",
 40 |     "\n",
 41 |     "`'Authorization': '<YOUR_TOKEN>'`"
 42 |    ]
 43 |   },
 44 |   {
 45 |    "cell_type": "code",
 46 |    "execution_count": null,
 47 |    "metadata": {},
 48 |    "outputs": [],
 49 |    "source": [
 50 |     "url='https://<API_GATEWAY_ID>.execute-api.us-east-1.amazonaws.com/prod/'\n",
 51 |     "headers = {\n",
 52 |     "    'Content-Type': 'application/json',\n",
 53 |     "    'Accept': 'application/json',\n",
 54 |     "    'Authorization': '<YOUR_TOKEN>',\n",
 55 |     "    # 'X-Amzn-SageMaker-Custom-Attributes': 'accept_eula=true' # Uncomment this line if you are using a llama2\n",
 56 |     "}"
 57 |    ]
 58 |   },
 59 |   {
 60 |    "attachments": {},
 61 |    "cell_type": "markdown",
 62 |    "metadata": {},
 63 |    "source": [
 64 |     "### Real-time Endpoint Management Functions\n",
 65 |     "When you create your endpoint for the first time, it will initialize it with the default provision time in minutes. You can check the available time left on your endpoint by either querying a specific endpoint or get a list of endpoint.\n",
 66 |     "\n",
 67 |     "#### Querying time left for a specific endpoint\n",
 68 |     "\n",
 69 |     "You can check the time left for a specific endpoint by querying the `endpoint-expiry` api as part of the endpoint manager functionality and passing in the `EndpointName`. Below is an example on how to do this:"
 70 |    ]
 71 |   },
 72 |   {
 73 |    "cell_type": "code",
 74 |    "execution_count": null,
 75 |    "metadata": {},
 76 |    "outputs": [],
 77 |    "source": [
 78 |     "# Set endpoint you like to lookup\n",
 79 |     "endpoint_name = 'demo-Falcon40B-Endpoint'\n",
 80 |     "\n",
 81 |     "# Flan example\n",
 82 |     "# endpoint_name = 'demo-FlanT5-Endpoint'\n",
 83 |     "\n",
 84 |     "# endpoint_name = 'demo-LLama-Endpoint'\n",
 85 |     "\n",
 86 |     "# endpoint_name = 'demo-LLama2-Endpoint'"
 87 |    ]
 88 |   },
 89 |   {
 90 |    "cell_type": "code",
 91 |    "execution_count": null,
 92 |    "metadata": {},
 93 |    "outputs": [],
 94 |    "source": [
 95 |     "response = requests.get(url=f\"{url}/endpoint-expiry?EndpointName={endpoint_name}\", headers=headers)\n",
 96 |     "print(json.dumps(response.json(), indent=2))"
 97 |    ]
 98 |   },
 99 |   {
100 |    "attachments": {},
101 |    "cell_type": "markdown",
102 |    "metadata": {},
103 |    "source": [
104 |     "#### Querying time left for all managed endpoints\n",
105 |     "\n",
106 |     "You can also get a list of managed endpoints and their respective time left. This can be done by using the `endpoint-expiry` api as well."
107 |    ]
108 |   },
109 |   {
110 |    "cell_type": "code",
111 |    "execution_count": null,
112 |    "metadata": {},
113 |    "outputs": [],
114 |    "source": [
115 |     "response = requests.get(url=f\"{url}/endpoint-expiry\", headers=headers)\n",
116 |     "print(json.dumps(response.json(), indent=2))"
117 |    ]
118 |   },
119 |   {
120 |    "attachments": {},
121 |    "cell_type": "markdown",
122 |    "metadata": {},
123 |    "source": [
124 |     "#### Extending your real-time endpoint expiry time\n",
125 |     "The managed endpoint API also provides you with the ability to extend the expiry date. Below is an example on how you can extend an endpoint by 30 minutes."
126 |    ]
127 |   },
128 |   {
129 |    "cell_type": "code",
130 |    "execution_count": null,
131 |    "metadata": {},
132 |    "outputs": [],
133 |    "source": [
134 |     "payload = {\n",
135 |     "    \"EndpointName\": endpoint_name,\n",
136 |     "    \"minutes\": 30\n",
137 |     "}\n",
138 |     "response = requests.post(url=f\"{url}/endpoint-expiry\", headers=headers, json=payload)\n",
139 |     "print(json.dumps(response.json(), indent=2))"
140 |    ]
141 |   },
142 |   {
143 |    "attachments": {},
144 |    "cell_type": "markdown",
145 |    "metadata": {},
146 |    "source": [
147 |     "#### Adding a new real-time endpoint\n",
148 |     "With the endpoint management API, you can also add a new real-time endpoint with pre-existing Amazon SageMaker endpoint configurations.\n",
149 |     "\n",
150 |     "Note: You can use the endpoint manager for any model regardless if it is jumpstart or not as long as you have a define Amazon SageMaker endpoint configuration."
151 |    ]
152 |   },
153 |   {
154 |    "cell_type": "code",
155 |    "execution_count": null,
156 |    "metadata": {},
157 |    "outputs": [],
158 |    "source": [
159 |     "new_endpoint_name=\"\"\n",
160 |     "new_endpoint_config_name=\"\""
161 |    ]
162 |   },
163 |   {
164 |    "cell_type": "code",
165 |    "execution_count": null,
166 |    "metadata": {},
167 |    "outputs": [],
168 |    "source": [
169 |     "payload = {\n",
170 |     "    \"EndpointName\": new_endpoint_name,\n",
171 |     "    \"EndpointConfigName\": new_endpoint_config_name,\n",
172 |     "    \"minutes\": 30\n",
173 |     "}\n",
174 |     "response = requests.post(url=f\"{url}/endpoint-expiry\", headers=headers, json=payload)\n",
175 |     "print(json.dumps(response.json(), indent=2))"
176 |    ]
177 |   },
178 |   {
179 |    "attachments": {},
180 |    "cell_type": "markdown",
181 |    "metadata": {},
182 |    "source": [
183 |     "### Interacting with your endpoint via API gateway\n",
184 |     "\n",
185 |     "With the deploy API Gateway and model lambda, you can interact with your Amazon SageMaker endpoint through the internet via API Gateway. Below is an example on how to send your payload request to the falcon model."
186 |    ]
187 |   },
188 |   {
189 |    "cell_type": "code",
190 |    "execution_count": null,
191 |    "metadata": {},
192 |    "outputs": [],
193 |    "source": [
194 |     "# Falcon example\n",
195 |     "payload = {\n",
196 |     "    \"inputs\": \"Write a program to compute factorial in python:\",\n",
197 |     "    \"parameters\": {\"max_new_tokens\": 200}\n",
198 |     "}\n",
199 |     "\n",
200 |     "# FLAN example\n",
201 |     "# payload = {\n",
202 |     "#     'text_inputs':'Write a program to compute factorial in python:', \n",
203 |     "#     'max_length': 100, \n",
204 |     "#     'temperature': 0.0, \n",
205 |     "#     'seed': 321\n",
206 |     "# }\n",
207 |     "\n",
208 |     "# open-llama example\n",
209 |     "# payload = {\n",
210 |     "#     \"text_inputs\": \"Building a website can be done in 10 simple steps:\",\n",
211 |     "#     \"max_length\": 110,\n",
212 |     "#     \"no_repeat_ngram_size\": 3,\n",
213 |     "# }\n",
214 |     "\n",
215 |     "# llama-2 example\n",
216 |     "# payload = {\n",
217 |     "#         \"inputs\": \"Building a website can be done in 10 simple steps:\",\n",
218 |     "#         \"parameters\" :\n",
219 |     "#             {\n",
220 |     "#                 \"max_new_tokens\": 110, \n",
221 |     "#                 \"top_p\": 0.9, \n",
222 |     "#                 \"temperature\" : 0.1\n",
223 |     "#             }\n",
224 |     "# }\n",
225 |     "\n",
226 |     "response = requests.post(url=f\"{url}/falcon\", headers=headers, json=payload)\n",
227 |     "print(json.dumps(response.json(), indent=2))"
228 |    ]
229 |   },
230 |   {
231 |    "cell_type": "markdown",
232 |    "metadata": {},
233 |    "source": [
234 |     "## Asynchronously interacting with your real-time endpoint via API Gateway\n",
235 |     "\n",
236 |     "In some instances, the real-time endpoint may take longer than 30 seconds to return a responses. Due to API gateway's 30 second timeout hard limit, an asynchronous approach will be required. The following code show you an example on how to asynchronously interactive with your real-time endpoint via two API calls."
237 |    ]
238 |   },
239 |   {
240 |    "cell_type": "code",
241 |    "execution_count": null,
242 |    "metadata": {},
243 |    "outputs": [],
244 |    "source": [
245 |     "# Falcon example\n",
246 |     "payload = {\n",
247 |     "    \"endpointname\": endpoint_name,\n",
248 |     "    \"body\": {\"inputs\": \"Write a program to compute factorial in python:\", \"parameters\": {\"max_new_tokens\": 200}}\n",
249 |     "}\n",
250 |     "\n",
251 |     "response = requests.post(url=f\"{url}/startexecution\", headers=headers, json=payload)\n",
252 |     "print(json.dumps(response.json(), indent=2))\n",
253 |     "\n",
254 |     "# Get the execution arn\n",
255 |     "executionArn = response.json().get('executionArn')"
256 |    ]
257 |   },
258 |   {
259 |    "cell_type": "code",
260 |    "execution_count": null,
261 |    "metadata": {},
262 |    "outputs": [],
263 |    "source": [
264 |     "payload ={\n",
265 |     "    \"executionArn\": executionArn\n",
266 |     "}\n",
267 |     "response = requests.post(url=f\"{url}/describeexecution\", headers=headers, json=payload)\n",
268 |     "\n",
269 |     "while 'RUNNING' in response.json().get('status'):\n",
270 |     "    time.sleep(1)\n",
271 |     "    response = requests.post(url=f\"{url}/describeexecution\", headers=headers, json=payload)\n",
272 |     "    print(\"Step is still running...\")\n",
273 |     "\n",
274 |     "# Return results\n",
275 |     "output = response.json().get('output')\n",
276 |     "output_json = json.loads(output)\n",
277 |     "body = json.loads(output_json['Body'])\n",
278 |     "body[0]['generated_text']"
279 |    ]
280 |   },
281 |   {
282 |    "attachments": {},
283 |    "cell_type": "markdown",
284 |    "metadata": {},
285 |    "source": [
286 |     "#### Interactive with your endpoint via Langchain/APIGateway\n",
287 |     "\n",
288 |     "The deploy API can be used with the [Amazon API Gateway/Langchain](https://python.langchain.com/docs/ecosystem/integrations/amazon_api_gateway) integration.\n",
289 |     "\n",
290 |     "The following example will walk you through on how to interact with the API Gateway backed by a Lambda authorizer."
291 |    ]
292 |   },
293 |   {
294 |    "cell_type": "code",
295 |    "execution_count": null,
296 |    "metadata": {},
297 |    "outputs": [],
298 |    "source": [
299 |     "!pip install langchain==0.0.238"
300 |    ]
301 |   },
302 |   {
303 |    "cell_type": "code",
304 |    "execution_count": null,
305 |    "metadata": {},
306 |    "outputs": [],
307 |    "source": [
308 |     "from langchain.llms import AmazonAPIGateway\n",
309 |     "from langchain.agents import load_tools\n",
310 |     "from langchain.agents import initialize_agent\n",
311 |     "from langchain.agents import AgentType"
312 |    ]
313 |   },
314 |   {
315 |    "cell_type": "code",
316 |    "execution_count": null,
317 |    "metadata": {},
318 |    "outputs": [],
319 |    "source": [
320 |     "llm = AmazonAPIGateway(api_url=f\"{url}/falcon\", headers=headers)"
321 |    ]
322 |   },
323 |   {
324 |    "attachments": {},
325 |    "cell_type": "markdown",
326 |    "metadata": {},
327 |    "source": [
328 |     "### Langchain LLM example"
329 |    ]
330 |   },
331 |   {
332 |    "cell_type": "code",
333 |    "execution_count": null,
334 |    "metadata": {},
335 |    "outputs": [],
336 |    "source": [
337 |     "parameters = {\n",
338 |     "    \"max_new_tokens\": 100,\n",
339 |     "    \"num_return_sequences\": 1,\n",
340 |     "    \"top_k\": 50,\n",
341 |     "    \"top_p\": 0.95,\n",
342 |     "    \"do_sample\": False,\n",
343 |     "    \"return_full_text\": True,\n",
344 |     "    \"temperature\": 0.2,\n",
345 |     "}\n",
346 |     "\n",
347 |     "prompt = \"what day comes after Friday?\"\n",
348 |     "llm.model_kwargs = parameters\n",
349 |     "llm(prompt)"
350 |    ]
351 |   },
352 |   {
353 |    "attachments": {},
354 |    "cell_type": "markdown",
355 |    "metadata": {},
356 |    "source": [
357 |     "### Langchain/APIGateway Agent Example"
358 |    ]
359 |   },
360 |   {
361 |    "cell_type": "code",
362 |    "execution_count": null,
363 |    "metadata": {},
364 |    "outputs": [],
365 |    "source": [
366 |     "parameters = {\n",
367 |     "    \"max_new_tokens\": 50,\n",
368 |     "    \"num_return_sequences\": 1,\n",
369 |     "    \"top_k\": 250,\n",
370 |     "    \"top_p\": 0.25,\n",
371 |     "    \"do_sample\": False,\n",
372 |     "    \"temperature\": 0.1,\n",
373 |     "}\n",
374 |     "\n",
375 |     "llm.model_kwargs = parameters\n",
376 |     "\n",
377 |     "# Next, let's load some tools to use. Note that the `llm-math` tool uses an LLM, so we need to pass that in.\n",
378 |     "tools = load_tools([\"python_repl\", \"llm-math\"], llm=llm)\n",
379 |     "\n",
380 |     "# Finally, let's initialize an agent with the tools, the language model, and the type of agent we want to use.\n",
381 |     "agent = initialize_agent(\n",
382 |     "    tools,\n",
383 |     "    llm,\n",
384 |     "    agent=AgentType.ZERO_SHOT_REACT_DESCRIPTION,\n",
385 |     "    verbose=True,\n",
386 |     ")\n",
387 |     "\n",
388 |     "# Now let's test it out!\n",
389 |     "agent.run(\"\"\"\n",
390 |     "Write a Python script that prints \"Hello, world!\"\n",
391 |     "\"\"\")"
392 |    ]
393 |   },
394 |   {
395 |    "cell_type": "code",
396 |    "execution_count": null,
397 |    "metadata": {},
398 |    "outputs": [],
399 |    "source": []
400 |   }
401 |  ],
402 |  "metadata": {
403 |   "kernelspec": {
404 |    "display_name": "Python 3",
405 |    "language": "python",
406 |    "name": "python3"
407 |   },
408 |   "language_info": {
409 |    "codemirror_mode": {
410 |     "name": "ipython",
411 |     "version": 3
412 |    },
413 |    "file_extension": ".py",
414 |    "mimetype": "text/x-python",
415 |    "name": "python",
416 |    "nbconvert_exporter": "python",
417 |    "pygments_lexer": "ipython3",
418 |    "version": "3.11.3"
419 |   },
420 |   "orig_nbformat": 4
421 |  },
422 |  "nbformat": 4,
423 |  "nbformat_minor": 2
424 | }
425 | 


--------------------------------------------------------------------------------
/LICENSE:
--------------------------------------------------------------------------------
  1 |                                  Apache License
  2 |                            Version 2.0, January 2004
  3 |                         http://www.apache.org/licenses/
  4 | 
  5 |    TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
  6 | 
  7 |    1. Definitions.
  8 | 
  9 |       "License" shall mean the terms and conditions for use, reproduction,
 10 |       and distribution as defined by Sections 1 through 9 of this document.
 11 | 
 12 |       "Licensor" shall mean the copyright owner or entity authorized by
 13 |       the copyright owner that is granting the License.
 14 | 
 15 |       "Legal Entity" shall mean the union of the acting entity and all
 16 |       other entities that control, are controlled by, or are under common
 17 |       control with that entity. For the purposes of this definition,
 18 |       "control" means (i) the power, direct or indirect, to cause the
 19 |       direction or management of such entity, whether by contract or
 20 |       otherwise, or (ii) ownership of fifty percent (50%) or more of the
 21 |       outstanding shares, or (iii) beneficial ownership of such entity.
 22 | 
 23 |       "You" (or "Your") shall mean an individual or Legal Entity
 24 |       exercising permissions granted by this License.
 25 | 
 26 |       "Source" form shall mean the preferred form for making modifications,
 27 |       including but not limited to software source code, documentation
 28 |       source, and configuration files.
 29 | 
 30 |       "Object" form shall mean any form resulting from mechanical
 31 |       transformation or translation of a Source form, including but
 32 |       not limited to compiled object code, generated documentation,
 33 |       and conversions to other media types.
 34 | 
 35 |       "Work" shall mean the work of authorship, whether in Source or
 36 |       Object form, made available under the License, as indicated by a
 37 |       copyright notice that is included in or attached to the work
 38 |       (an example is provided in the Appendix below).
 39 | 
 40 |       "Derivative Works" shall mean any work, whether in Source or Object
 41 |       form, that is based on (or derived from) the Work and for which the
 42 |       editorial revisions, annotations, elaborations, or other modifications
 43 |       represent, as a whole, an original work of authorship. For the purposes
 44 |       of this License, Derivative Works shall not include works that remain
 45 |       separable from, or merely link (or bind by name) to the interfaces of,
 46 |       the Work and Derivative Works thereof.
 47 | 
 48 |       "Contribution" shall mean any work of authorship, including
 49 |       the original version of the Work and any modifications or additions
 50 |       to that Work or Derivative Works thereof, that is intentionally
 51 |       submitted to Licensor for inclusion in the Work by the copyright owner
 52 |       or by an individual or Legal Entity authorized to submit on behalf of
 53 |       the copyright owner. For the purposes of this definition, "submitted"
 54 |       means any form of electronic, verbal, or written communication sent
 55 |       to the Licensor or its representatives, including but not limited to
 56 |       communication on electronic mailing lists, source code control systems,
 57 |       and issue tracking systems that are managed by, or on behalf of, the
 58 |       Licensor for the purpose of discussing and improving the Work, but
 59 |       excluding communication that is conspicuously marked or otherwise
 60 |       designated in writing by the copyright owner as "Not a Contribution."
 61 | 
 62 |       "Contributor" shall mean Licensor and any individual or Legal Entity
 63 |       on behalf of whom a Contribution has been received by Licensor and
 64 |       subsequently incorporated within the Work.
 65 | 
 66 |    2. Grant of Copyright License. Subject to the terms and conditions of
 67 |       this License, each Contributor hereby grants to You a perpetual,
 68 |       worldwide, non-exclusive, no-charge, royalty-free, irrevocable
 69 |       copyright license to reproduce, prepare Derivative Works of,
 70 |       publicly display, publicly perform, sublicense, and distribute the
 71 |       Work and such Derivative Works in Source or Object form.
 72 | 
 73 |    3. Grant of Patent License. Subject to the terms and conditions of
 74 |       this License, each Contributor hereby grants to You a perpetual,
 75 |       worldwide, non-exclusive, no-charge, royalty-free, irrevocable
 76 |       (except as stated in this section) patent license to make, have made,
 77 |       use, offer to sell, sell, import, and otherwise transfer the Work,
 78 |       where such license applies only to those patent claims licensable
 79 |       by such Contributor that are necessarily infringed by their
 80 |       Contribution(s) alone or by combination of their Contribution(s)
 81 |       with the Work to which such Contribution(s) was submitted. If You
 82 |       institute patent litigation against any entity (including a
 83 |       cross-claim or counterclaim in a lawsuit) alleging that the Work
 84 |       or a Contribution incorporated within the Work constitutes direct
 85 |       or contributory patent infringement, then any patent licenses
 86 |       granted to You under this License for that Work shall terminate
 87 |       as of the date such litigation is filed.
 88 | 
 89 |    4. Redistribution. You may reproduce and distribute copies of the
 90 |       Work or Derivative Works thereof in any medium, with or without
 91 |       modifications, and in Source or Object form, provided that You
 92 |       meet the following conditions:
 93 | 
 94 |       (a) You must give any other recipients of the Work or
 95 |           Derivative Works a copy of this License; and
 96 | 
 97 |       (b) You must cause any modified files to carry prominent notices
 98 |           stating that You changed the files; and
 99 | 
100 |       (c) You must retain, in the Source form of any Derivative Works
101 |           that You distribute, all copyright, patent, trademark, and
102 |           attribution notices from the Source form of the Work,
103 |           excluding those notices that do not pertain to any part of
104 |           the Derivative Works; and
105 | 
106 |       (d) If the Work includes a "NOTICE" text file as part of its
107 |           distribution, then any Derivative Works that You distribute must
108 |           include a readable copy of the attribution notices contained
109 |           within such NOTICE file, excluding those notices that do not
110 |           pertain to any part of the Derivative Works, in at least one
111 |           of the following places: within a NOTICE text file distributed
112 |           as part of the Derivative Works; within the Source form or
113 |           documentation, if provided along with the Derivative Works; or,
114 |           within a display generated by the Derivative Works, if and
115 |           wherever such third-party notices normally appear. The contents
116 |           of the NOTICE file are for informational purposes only and
117 |           do not modify the License. You may add Your own attribution
118 |           notices within Derivative Works that You distribute, alongside
119 |           or as an addendum to the NOTICE text from the Work, provided
120 |           that such additional attribution notices cannot be construed
121 |           as modifying the License.
122 | 
123 |       You may add Your own copyright statement to Your modifications and
124 |       may provide additional or different license terms and conditions
125 |       for use, reproduction, or distribution of Your modifications, or
126 |       for any such Derivative Works as a whole, provided Your use,
127 |       reproduction, and distribution of the Work otherwise complies with
128 |       the conditions stated in this License.
129 | 
130 |    5. Submission of Contributions. Unless You explicitly state otherwise,
131 |       any Contribution intentionally submitted for inclusion in the Work
132 |       by You to the Licensor shall be under the terms and conditions of
133 |       this License, without any additional terms or conditions.
134 |       Notwithstanding the above, nothing herein shall supersede or modify
135 |       the terms of any separate license agreement you may have executed
136 |       with Licensor regarding such Contributions.
137 | 
138 |    6. Trademarks. This License does not grant permission to use the trade
139 |       names, trademarks, service marks, or product names of the Licensor,
140 |       except as required for reasonable and customary use in describing the
141 |       origin of the Work and reproducing the content of the NOTICE file.
142 | 
143 |    7. Disclaimer of Warranty. Unless required by applicable law or
144 |       agreed to in writing, Licensor provides the Work (and each
145 |       Contributor provides its Contributions) on an "AS IS" BASIS,
146 |       WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
147 |       implied, including, without limitation, any warranties or conditions
148 |       of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
149 |       PARTICULAR PURPOSE. You are solely responsible for determining the
150 |       appropriateness of using or redistributing the Work and assume any
151 |       risks associated with Your exercise of permissions under this License.
152 | 
153 |    8. Limitation of Liability. In no event and under no legal theory,
154 |       whether in tort (including negligence), contract, or otherwise,
155 |       unless required by applicable law (such as deliberate and grossly
156 |       negligent acts) or agreed to in writing, shall any Contributor be
157 |       liable to You for damages, including any direct, indirect, special,
158 |       incidental, or consequential damages of any character arising as a
159 |       result of this License or out of the use or inability to use the
160 |       Work (including but not limited to damages for loss of goodwill,
161 |       work stoppage, computer failure or malfunction, or any and all
162 |       other commercial damages or losses), even if such Contributor
163 |       has been advised of the possibility of such damages.
164 | 
165 |    9. Accepting Warranty or Additional Liability. While redistributing
166 |       the Work or Derivative Works thereof, You may choose to offer,
167 |       and charge a fee for, acceptance of support, warranty, indemnity,
168 |       or other liability obligations and/or rights consistent with this
169 |       License. However, in accepting such obligations, You may act only
170 |       on Your own behalf and on Your sole responsibility, not on behalf
171 |       of any other Contributor, and only if You agree to indemnify,
172 |       defend, and hold each Contributor harmless for any liability
173 |       incurred by, or claims asserted against, such Contributor by reason
174 |       of your accepting any such warranty or additional liability.
175 | 
176 |    END OF TERMS AND CONDITIONS
177 | 
178 |    APPENDIX: How to apply the Apache License to your work.
179 | 
180 |       To apply the Apache License to your work, attach the following
181 |       boilerplate notice, with the fields enclosed by brackets "[]"
182 |       replaced with your own identifying information. (Don't include
183 |       the brackets!)  The text should be enclosed in the appropriate
184 |       comment syntax for the file format. We also recommend that a
185 |       file or class name and description of purpose be included on the
186 |       same "printed page" as the copyright notice for easier
187 |       identification within third-party archives.
188 | 
189 |    Copyright 2023 🤖 ANZ ML
190 | 
191 |    Licensed under the Apache License, Version 2.0 (the "License");
192 |    you may not use this file except in compliance with the License.
193 |    You may obtain a copy of the License at
194 | 
195 |        http://www.apache.org/licenses/LICENSE-2.0
196 | 
197 |    Unless required by applicable law or agreed to in writing, software
198 |    distributed under the License is distributed on an "AS IS" BASIS,
199 |    WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
200 |    See the License for the specific language governing permissions and
201 |    limitations under the License.
202 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
  1 | 
  2 | # Amazon SageMaker Endpoint Manager
  3 | 
  4 | A solution that will allow you to deploy Amazon SageMaker Foundation Model endpoint backed by a API Gateway/Lambda with automation to start and stop your endpoints.
  5 | 
  6 | The Amazon SageMaker Endpoint manager solution will allow you to expose your Amazon SageMaker foundation models through the Amazon API Gateway with a dynamodb authorizer using CDK. Amazon SageMaker Endpoint manager also features a mechanism to manage your real-time endpoint which will automatically start and stop your Amazon SageMaker endpoint through an expiry datetime configuration for each endpoint. Similar to a parking meter whereby you top up credits to ensure that your parking does not expire, in this case, you keep renewing the endpoint expiry date time to keep it running.
  7 | 
  8 | This solution was designed to solve a recurring problem with users leaving their Amazon SageMaker endpoint on and forgetting to delete them after usage. The approach taken to solve this in the first iteration is to enforce users to renew their endpoint expiry times based on how much time need it up for. By doing so, it raises the awareness of the cost of the endpoint (particularly for LLM endpoint) and also ensure that the user is intentional as to how time they actually need to use the endpoint for testing and development.
  9 | 
 10 | ---
 11 | ## What's New
 12 | 
 13 | 17/08/2023
 14 | - Added the ability to asynchronously invoke the Amazon SageMaker real-time endpoint via API gateway
 15 | - Collapsed stacks into a single root stack for a simplified deployment 
 16 | 
 17 | 01/08/2023
 18 | - Added support for loading jumpstart models which uses a Model Package (i.e. Marketplace subscription) instead of a Docker image
 19 | - Added the ability to send custom attributes via API Gateway/AWS integration when invoking the Amazon SageMaker endpoint by specifying the `X-Amzn-SageMaker-Custom-Attributes` in the request header.
 20 | 
 21 | 27/07/2023
 22 | - Added lambda exception handling
 23 | - Added AWS Gateway direct AWS integration with Amazon SageMaker. To use AWS integration, set the `type` in the `integration` property to `api`.
 24 | 
 25 | 20/07/2023
 26 | - Added support for multiple endpoints and apis in a configuration file - `config/configs.json` file.
 27 | - Consolidated realtime and async stack to simplify deployment.
 28 | 
 29 | ---
 30 | 
 31 | ## Table of contents
 32 | - [Amazon SageMaker Endpoint Manager](#amazon-sagemaker-endpoint-manager)
 33 |   - [What's New](#whats-new)
 34 |   - [Table of contents](#table-of-contents)
 35 |   - [Demo Overview](#demo-overview)
 36 |   - [How to deploy the stack](#how-to-deploy-the-stack)
 37 |   - [Real-time Endpoint Management Functions - Querying your real-time endpoint expiry time](#real-time-endpoint-management-functions---querying-your-real-time-endpoint-expiry-time)
 38 |   - [Real-time Endpoint Management Functions - Extending your real-time endpoint expiry time](#real-time-endpoint-management-functions---extending-your-real-time-endpoint-expiry-time)
 39 |   - [Real-time Endpoint Management Functions - Adding a new real-time endpoint](#real-time-endpoint-management-functions---adding-a-new-real-time-endpoint)
 40 |   - [Interacting with your real-time endpoint via API](#interacting-with-your-real-time-endpoint-via-api)
 41 |   - [Asynchronously interacting with your real-time endpoint via API](#asynchronously-interacting-with-your-real-time-endpoint-via-api)
 42 |   - [Example Notebook](#example-notebook)
 43 |   - [Endpoint Manager Configurations](#endpoint-manager-configurations)
 44 |     - [**Jumpstart model**](#jumpstart-model)
 45 |     - [**Schedule Configuration**](#schedule-configuration)
 46 |     - [**Integration Configuration**](#integration-configuration)
 47 |     - [**Integration Properties**](#integration-properties)
 48 |   - [How does the endpoint manager work?](#how-does-the-endpoint-manager-work)
 49 |   - [To Do](#to-do)
 50 |   - [References](#references)
 51 | ## Demo Overview
 52 | 
 53 | ![Architecture Diagram](docs/diagram.png)
 54 | 
 55 | ---
 56 | 
 57 | ## How to deploy the stack
 58 | 
 59 | 1. Create a python virtualenv
 60 | 
 61 |     ```
 62 |     $ python3 -m venv .venv
 63 |     ```
 64 | 
 65 | 2. Activate your virtual environment
 66 | 
 67 |     For Mac/Linux platform:
 68 |     ```
 69 |     $ source .venv/bin/activate
 70 |     ```
 71 | 
 72 |     If you are a Windows platform, you would activate the virtualenv like this:
 73 | 
 74 |     ```
 75 |     % .venv\Scripts\activate.bat
 76 |     ```
 77 | 3. Install required dependencies
 78 | 
 79 |     ```
 80 |     $ pip install -r requirements.txt
 81 |     ```
 82 | 
 83 | 4. Bootstrap your environment (if required)
 84 | 
 85 |     If you have not previously used CDK, bootstrap your environment
 86 | 
 87 |     ```
 88 |     $ cdk bootstrap
 89 |     ```
 90 | 
 91 | 5. Define your configuration
 92 | 
 93 |     Setup your Amazon SageMaker Endpoint Manager and configure the models that you'd like to deploy by opening the [`config/config.json`](config/configs.json) file with a text editor. Configure the project settings such as your `project_prefix`, `region_name` - region to deploy the stack and `ddb_auth_table_name`. For more information on the configuration, refer to the [Endpoint manager Configuration section](#endpoint-manager-configurations).
 94 | 
 95 |     Next define the jumpstart models you'd like to deploy. For more information on the jumpstart model configuration, refer to [jumpstart model configuration section](#example-jumpstart-model-configuration). Below, you can find an example of how to deploy a real-time falcon40b and flan jumpstart model. You will also find additional examples in the [`config/example-configs.json`](config/example-configs.json) file.
 96 |     
 97 |     ### Example Jumpstart model configuration
 98 | 
 99 |     **Falcon 40B - Realtime *publicly accessible* Endpoint configuration using apigateway/lambda integration**
100 |     ```
101 |         {
102 |             "name" : "Falcon40BPublic",
103 |             "model_id" : "huggingface-llm-falcon-40b-instruct-bf16",
104 |             "inference_instance_count": 2,
105 |             "inference_instance_type" : "ml.g5.12xlarge",
106 |             "inference_type": "realtime",
107 |             "public": true,
108 |             "schedule": {
109 |                 "initial_provision_minutes": 90
110 |             },
111 |             "integration": {
112 |                 "type": "lambda",
113 |                 "properties": {
114 |                     "lambda_src": "functions/falcon",
115 |                     "api_resource_name": "falcon"
116 |                 }
117 |             }
118 |         }
119 |     ```
120 | 
121 |     **FLAN T5 - Realtime Endpoint configuration using apigateway/lambda integration**
122 |     ```
123 |     {
124 |             "name" : "FlanT5",
125 |             "model_id" : "huggingface-text2text-flan-t5-xxl",
126 |             "inference_instance_type" : "ml.g5.12xlarge",
127 |             "inference_type": "realtime",
128 |             "schedule": {
129 |                 "initial_provision_minutes": 0
130 |             },
131 |             "integration": {
132 |                 "type": "lambda",
133 |                 "properties": {
134 |                     "lambda_src": "functions/flan",
135 |                     "api_resource_name": "flan"
136 |                 }
137 |             }
138 |         }
139 |     ```
140 | 
141 |     **FLAN T5 - Realtime Endpoint configuration using API Gateway/AWS Integration**
142 |     ```
143 |     {
144 |             "name" : "FlanT5",
145 |             "model_id" : "huggingface-text2text-flan-t5-xxl",
146 |             "inference_instance_type" : "ml.g5.12xlarge",
147 |             "inference_type": "realtime",
148 |             "schedule": {
149 |                 "initial_provision_minutes": 0
150 |             },
151 |             "integration": {
152 |                 "type": "api",
153 |                 "properties": {
154 |                     "api_resource_name": "flan"
155 |                 }
156 |             }
157 |         }
158 |     ```
159 | 
160 |     **FLAT T5 - Asynchronous Endpoint configuration**
161 |     ```
162 |     {
163 |             "name" : "FlanT5Async",
164 |             "model_id" : "huggingface-text2text-flan-t5-xxl",
165 |             "inference_instance_type" : "ml.g5.8xlarge",
166 |             "inference_type": "async"
167 |     }
168 |     ```
169 | 
170 | 6. Deploy the SageMaker Endpoint Manager solution stack
171 | 
172 |     **Note** If you're upgrading from a previous version of Amazon SageMaker Endpoint Manager, you may need to delete the existing stacks. Please refer to the following instructions [upgrade instructions](#upgrade-instructions).
173 | 
174 | 
175 |     Deploy the end-to-end SageMaker Endpoint manager stack. This stack will deploy several nested stacks including: 
176 |     1. API gateway stack: The API Gateway which will provide us with an internet facing API to interact with our Amazon SageMaker Endpoint and manage our Amazon SageMaker endpoint. The API Gateway is backed by a basic lambda authorizer with the authorization tokens stored in an Amazon DynamoDB database. 
177 |     2. Endpoint Manager Stack: The stack includes several lambda functions that will be responsible for the automatic creation and deletion of your Amazon SageMaker real-time endpoints. 
178 |     3. Model Stack: Deploys the predefined Amazon SageMaker endpoints and passthrough lambda (if configured). This stack will deploy all models in the list of jumpstart models.
179 |     4. Step function Stack: Stack to support asynchronous invocation of the Amazon SageMaker real-time endpoint fronted by the API Gateway
180 | 
181 |      ```
182 |      cdk deploy SagemakerEndpointManagerStack
183 |      ```
184 | 
185 |      ### Upgrade instructions
186 |      If you're using an older version of Amazon SageMaker Endpoint Manager with the 3 separate stacks, you need to delete this stacks before deploying the new version. To do so, refer to the following instructions:
187 | 
188 |      1. Checkout the older Amazon SageMaker version
189 |         
190 |         ```
191 |         git checkout 049e21e
192 |         ```
193 |      2. Delete the stacks. 
194 |         ```
195 |         cdk destroy --all
196 |         ```
197 |         If you have any existing endpoints, make sure you manually delete the endpoints.
198 |     3. Switch back to the main branch
199 |         ```
200 |         git checkout main
201 |         ```
202 |      4. Deploy the end-to-end SageMaker Endpoint manager stack.
203 | 
204 |         ```
205 |         cdk deploy SagemakerEndpointManagerStack
206 |         ```  
207 | 
208 |  7.  Setup your auth
209 | 
210 |      In your AWS account, you will find a Dynamodb table `auth` which stores a token (or pass code) which you will use to as an authorization token to access the APIs. Create an item in the `auth` table with an attribute `token` and set the value to your pass code which you will use when calling the API.
211 | 
212 |  Now that you have setup your environment with a real-time endpoint, let's take a look at some of the functionality this solutions has to offer.
213 | 
214 | ---
215 | ## Real-time Endpoint Management Functions - Querying your real-time endpoint expiry time
216 | 
217 | When you create your endpoint for the first time, it will initialize it with the default provision time in minutes. You can check the available time left on your endpoint by either querying a specific endpoint or get a list of endpoint.
218 | 
219 | To check the time left for a specific endpoint run the following:
220 | 
221 | ```
222 | curl --location 'https://xxxxxxxxxx.execute-api.us-east-1.amazonaws.com/prod/endpoint-expiry?EndpointName=<YOUR ENDPOINT NAME>' \
223 | --header 'Authorization: <YOUR TOKEN VALUE>'
224 | ```
225 | 
226 | Example Request:
227 | ```
228 | curl --location 'https://xxxxxxxxxx.execute-api.us-east-1.amazonaws.com/prod/endpoint-expiry?EndpointName=demo-Falcon40B-Endpoint' \
229 | --header 'Authorization: <YOUR TOKEN VALUE>'
230 | ```
231 | 
232 | Example Response:
233 | ```
234 | {
235 |     "EndpointName": "demo-Falcon40B-Endpoint",
236 |     "EndpointExpiry ": "22-06-2023-08-24-12",
237 |     "TimeLeft": "00:00:10.21130"
238 | }
239 | ```
240 | 
241 | To check the time left for all endpoints configured, run the following:
242 | ```
243 | curl --location 'https://xxxxxxxxxx.execute-api.us-east-1.amazonaws.com/prod/endpoint-expiry' \
244 | --header 'Authorization: <YOUR TOKEN VALUE>'
245 | ```
246 | 
247 | Example Response
248 | ```
249 | [
250 |     {
251 |         "EndpointName": "demo-Falcon40B-Endpoint",
252 |         "EndpointExpiry ": "26-06-2023-12-39-47",
253 |         "TimeLeft": "0:24:05.157431"
254 |     },
255 |     {
256 |         "EndpointName": "another-ml-Endpoint",
257 |         "EndpointExpiry ": "26-06-2023-13-15-27",
258 |         "TimeLeft": "0:59:45.157396"
259 |     }
260 | ]
261 | ```
262 | ---
263 | ## Real-time Endpoint Management Functions - Extending your real-time endpoint expiry time
264 | 
265 | To extend the amount of time your endpoint will be kept alive, run the following:
266 | 
267 | ```
268 | curl --location 'https://xxxxxxxxxx.execute-api.us-east-1.amazonaws.com/prod/endpoint-expiry' \
269 | --header 'Authorization: <YOUR TOKEN VALUE>' \
270 | --header 'Content-Type: application/json' \
271 | --data '{
272 |     "EndpointName": "<YOUR ENDPOINT NAME>",
273 |     "minutes": <AMOUNT OF TIME TO EXTEND YOUR ENDPOINT>
274 | }'
275 | ```
276 | 
277 | The following example request will extend the endpoint uptime by 10 minutes:
278 | ```
279 | curl --location 'https://xxxxxxxxxx.execute-api.us-east-1.amazonaws.com/prod/endpoint-expiry' \
280 | --header 'Authorization: <YOUR TOKEN VALUE>' \
281 | --header 'Content-Type: application/json' \
282 | --data '{
283 |     "EndpointName": "demo-Falcon40B-Endpoint",
284 |     "minutes": 10
285 | }'
286 | ```
287 | 
288 | Expected Response:
289 | ```
290 | {
291 |     "EndpointName": "demo-Falcon40B-Endpoint",
292 |     "EndpointExpiry ": "26-06-2023-12-39-47",
293 |     "TimeLeft": "0:30:46.596924"
294 | }
295 | ```
296 | ---
297 | 
298 | ## Real-time Endpoint Management Functions - Adding a new real-time endpoint
299 | 
300 | With the endpoint management API, you can add a new real-time endpoint with pre-existing Amazon SageMaker endpoint configurations. 
301 | 
302 | **Note:** You can use the endpoint manager for any model regardless if it is jumpstart or not.
303 | 
304 | Run the following API call to create a new real-time endpoint:
305 | 
306 | ```
307 | curl --location 'https://xxxxxxxxxx.execute-api.us-east-1.amazonaws.com/prod/endpoint-expiry' \
308 | --header 'Authorization: <YOUR TOKEN VALUE>' \
309 | --header 'Content-Type: application/json' \
310 | --data '{
311 |     "EndpointName": "<YOUR ENDPOINT NAME>",
312 |     "EndpointConfigName": "<YOUR ENDPOINT CONFIGURATION NAME>",
313 |     "minutes": <PROVISION TIME IN MINUTES>
314 | }'
315 | ```
316 | 
317 | The following API call will create a new endpoint with the name `test-inpainting-Endpoint` using the endpoint configuration `jumpstart-example-model-inpainting-cfg` with an initial uptime of 30 minutes.
318 | ```
319 | curl --location 'https://xxxxxxxxxx.execute-api.us-east-1.amazonaws.com/prod/endpoint-expiry' \
320 | --header 'Authorization: <YOUR TOKEN VALUE>' \
321 | --header 'Content-Type: application/json' \
322 | --data '{
323 |     "EndpointName": "test-inpainting-Endpoint",
324 |     "EndpointConfigName": "jumpstart-example-model-inpainting-cfg",
325 |     "minutes": 30
326 | }'
327 | ```
328 | 
329 | Example Response:
330 | ```
331 | {
332 |     "EndpointName": "test-inpainting-Endpoint",
333 |     "EndpointExpiry ": "26-06-2023-13-15-27",
334 |     "TimeLeft": 30
335 | }
336 | ```
337 | 
338 | ---
339 | ## Interacting with your real-time endpoint via API
340 | With the deploy API Gateway and model lambda, you can interact with your Amazon SageMaker endpoint through the internet. Below are examples of how you can interact with the falcon and flan api.
341 | 
342 | **Falcon API**
343 | 
344 | Sample request to interact with the **Falcon API**:
345 | ```
346 | curl --location 'https://xxxxxxxxxx.execute-api.us-east-1.amazonaws.com/prod/falcon' \
347 | --header 'Authorization: <YOUR TOKEN VALUE>' \
348 | --header 'Content-Type: application/json' \
349 | --data '{
350 |     "inputs": "Write a program to compute factorial in python:",
351 |     "parameters": {"max_new_tokens": 200}
352 | }'
353 | ```
354 | 
355 | Sample Response:
356 | ```
357 | [
358 |     {
359 |         "generated_text": "\nYou can compute factorial in Python using the built-in function `math.factorial()`. Here's an example:\n\n```python\nimport math\n\nn = 5\nfactorial = math.factorial(n)\nprint(factorial)\n```\n\nThis will output `120`, which is the factorial of 5."
360 |     }
361 | ]
362 | ```
363 | **Flan API**
364 | 
365 | Sample request to interact with the **Flan API**:
366 | 
367 | ```
368 | curl --location 'https://xxxxxxxxxx.execute-api.us-east-1.amazonaws.com/prod/flan' \
369 | --header 'Authorization: <YOUR TOKEN VALUE>' \
370 | --header 'Content-Type: application/json' \
371 | --data '{
372 |     "text_inputs":"write a story about beautiful weather on a topical island.", 
373 |     "max_length": 50, 
374 |     "temperature": 0.0,
375 |     "seed": 321
376 | }'
377 | ```
378 | 
379 | Expected response
380 | ```
381 | {
382 |     "generated_texts": [
383 |         "..."
384 |     ]
385 | }
386 | ```
387 | ---
388 | ## Asynchronously interacting with your real-time endpoint via API
389 | 
390 | In some cases, the inference may take more than 30 seconds which exceeds the API gateway timeout limit. As such an alternative approach to invoking the real-time Amazon SageMaker endpoint is required. This solution provide you with a mechanism to asynchronously invoke the endpoint and retrieve the response when the invocation has been completed. To run inference asynchronously, you will call the API in two step. The first to send your prompt `https://xxxxxxxxxx.execute-api.us-east-1.amazonaws.com/prod/startexecution` and the second to retrieve the result `https://xxxxxxxxxx.execute-api.us-east-1.amazonaws.com/prod/describeexecution`.
391 | 
392 | 
393 | Send prompt to API request
394 | ```
395 | curl --location 'https://xxxxxxxxxx.execute-api.us-east-1.amazonaws.com/prod/startexecution' \
396 | --header 'Authorization: <YOUR TOKEN>' \
397 | --header 'Content-Type: application/json' \
398 | --data '{
399 |     "endpointname": "demo-Falcon40B-Endpoint",
400 |     "body": {"inputs": "Write a program to compute factorial in python:", "parameters": {"max_new_tokens": 200}}
401 | }'
402 | ```
403 | 
404 | Example Send prompt response
405 | ```
406 | {
407 |     "executionArn": "arn:aws:states:us-east-1:xxxxxxxxxxxxx:execution:SageMakerInvokeStepfunctionD7692275-BlnWWEoa6OY7:d7bde7bb-084a-4a29-b025-98"
408 | }
409 | ```
410 | 
411 | Once you've sent a prompt request, you will be provided with an executionARN. You will then use this response and send it to the retrieval API. Refer to the example below.
412 | 
413 | Retrieve an invocation API request
414 | ```
415 | curl --location 'https://xxxxxxxxxx.execute-api.us-east-1.amazonaws.com/prod/describeexecution' \
416 | --header 'Authorization: <YOUR TOKEN>' \
417 | --header 'Content-Type: application/json' \
418 | --data '{
419 |     "executionArn": "arn:aws:states:us-east-1:xxxxxxxxxxxxx:execution:SageMakerInvokeStepfunctionD7692275-BlnWWEoa6OY7:d7bde7bb-084a-4a29-b025-98d9da47ce44"
420 | }'
421 | ```
422 | 
423 | If successful, you will get a response with a the status parameter as `SUCCEEDED`. If the invocation has a status of `RUNNING`, run the request again until you get a `SUCCEEDED` state. The output of the response can be found in the `output` parameter.
424 |  
425 | Example retrieval response
426 | ```
427 | {
428 |     "executionArn": "arn:aws:states:us-east-1:xxxxxxxxxxxxx:execution:SageMakerInvokeStepfunctionD7692275-BlnWWEoa6OY7:d7bde7bb-084a-4a29-b025-98d9da47ce44",
429 |     "input": "{\"endpointname\":\"demo-Falcon40B-Endpoint\",\"body\":{\"inputs\":\"Write a program to compute factorial in python:\",\"parameters\":{\"max_new_tokens\":200}}}",
430 |     "inputDetails": {
431 |         "__type": "com.amazonaws.swf.base.model#CloudWatchEventsExecutionDataDetails",
432 |         "included": true
433 |     },
434 |     "name": "d7bde7bb-084a-4a29-b025-98d9da47ce44",
435 |     "output": "{\"Body\":\"[{\\\"generated_text\\\":\\\"\\\\nYou can compute factorial in Python using the built-in function `math.factorial()`. Here's an example:\\\\n\\\\n```python\\\\nimport math\\\\n\\\\nn = 5\\\\nfactorial = math.factorial(n)\\\\nprint(factorial)\\\\n```\\\\n\\\\nThis will output `120`, which is the factorial of 5.\\\"}]\",\"ContentType\":\"application/json\",\"InvokedProductionVariant\":\"AllTraffic\"}",
436 |     ...
437 |     "startDate": 1.691411436575E9,
438 |     "stateMachineArn": "arn:aws:states:us-east-1:xxxxxxxxxxxxx:stateMachine:SageMakerInvokeStepfunctionD7692275-BlnWWEoa6OY7",
439 |     "status": "SUCCEEDED",
440 |     ...
441 | }
442 | ```
443 | 
444 | ---
445 | ## Example Notebook
446 | 
447 | You can also interact with the API gateway via notebook. To do so, clone this repository or copy `00-api_gateway_managed_endpoint.ipynb` to your notebook environment (i.e. Amazon SageMaker Studio) and run through the instructions in the notebook.
448 | 
449 | The notebook will show you how to manage your endpoint, interact with your SageMaker endpoint API And how to use Langchain with APIGateway.
450 | 
451 | ---
452 | 
453 | ## Endpoint Manager Configurations
454 | Endpoint manager configurations:
455 | - `project_prefix`
456 |   - Description: Project prefix name to use for all resources
457 |   - Type: String
458 |   - Required: Yes
459 | - `region_name`
460 |   - Description: AWS region to deploy the CDK solution and endpoint
461 |   - Type: String
462 |   - Example values: `us-east-1` | `us-west-1`
463 | - `ddb_auth_table_name`
464 |   - Description: Name of the dynamodb table to store your auth token for your API Gateway
465 |   - Type: String
466 | - `jumpstart_models`
467 |   - Description: List of jumpstart models configurations 
468 |   - Type: Array of [Jumpstart model](#jumpstart-model)
469 | 
470 | ### **Jumpstart model**
471 | Jumpstart model configurations
472 |   - `name`
473 |     - Description: Name of model
474 |     - Type: String
475 |     - Required: Yes
476 |   - `model_id`
477 |     - Description: Jumpstart model ID (A list of model id can be found [here](https://sagemaker.readthedocs.io/en/stable/doc_utils/pretrainedmodels.html))
478 |     - Type: String
479 |     - Required: Yes
480 |   - `inference_instance_count`
481 |     - Description: Number of instances to use for inference
482 |     - Type: Integer
483 |     - Required: No
484 |     - Default: 1
485 |   - `inference_instance_type`
486 |     - Description: Size of the instance type to use
487 |     - Type: String
488 |     - Required: Yes
489 |   - `inference_type`
490 |     - Description: Type of inference endpoint to be deployed - Real-time or asynchronous
491 |     - Type: String
492 |     - Required: Yes
493 |     - Valid Options: `realtime` | `async`
494 |   - `public`
495 |     - Description: Whether to accept un-authenticated inference requests (without an API token) for this model
496 |     - Type: Boolean
497 |     - Required: No
498 |     - Default: false
499 |   - `schedule`
500 |     - Description: Schedule configuration for the endpoint
501 |     - Type: [Schedule Configuration](#schedule-config) object
502 |   - `integration`
503 |     - Description: Endpoint integration configurations.
504 |     - Type: [Integration](#integration-configuration) object.
505 | 
506 | 
507 | ### **Schedule Configuration**
508 | SageMaker Endpoint Schedule configuration (Currently supports expiring endpoints).
509 | - `initial_provision_minutes`
510 |     - Description: Initial time the endpoint will be provisioned for when the CDK stack is deployed in minutes.
511 |     - Type: Integer
512 | 
513 | 
514 | ### **Integration Configuration**
515 | Endpoint integration configurations
516 | - `type`
517 |     - Description: Indicates the type of API Gateway integration with the endpoint. If using lambda integration, ensure that you specify the `lambda_src` configuration in the integration properties.
518 |     - Type: String
519 |     - Required: Yes
520 |     - Valid Options: `lambda` | `api`
521 | - `properties`
522 |     - Description: Integration properties
523 |     - Type: [Integration Properties](#integration-properties)
524 | 
525 | ### **Integration Properties**
526 | Integration specific configurations
527 | - `lambda_src`
528 |     - Description: Path to the lambda handler for the API Gateway/Endpoint integration.
529 |     - Type: String
530 |     - Required: Yes for lambda integration type.
531 | - `api_resource_name`
532 |     - Description: API gateway resource name. For example, setting it to `falcon` will result in the API gateway path as `https://<api_gateway_id>.execute-api.us-east-1.amazonaws.com/prod/falcon`
533 |     - Type: String
534 |     - Required: Yes
535 | 
536 | ---
537 | ## How does the endpoint manager work?
538 | 
539 | 1. When the stack is provisioned for the first time, the user defined the initial required endpoint provision time in minutes (`initial_provision_time_minutes`) in the `app.py`
540 | 2. Once provisioned, a start/stop lambda will poll a list of Amazon SageMaker Parameter store parameter with the prefix `/sagemaker/endpoint/expiry/*` to check the expiry date/time for each endpoint. If the date/time is not expired and an endpoint has not been created, the lambda will create the model endpoint.
541 | 3. If the expiry datetime is less than the current time, the lambda will automatically delete the endpoint.
542 | 4. Users can check the time left on their endpoint by querying the `endpoint-expiry` API. For more information refer to [Real-time Endpoint Management Functions - Querying your real-time endpoint expiry time](#real-time-endpoint-management-functions---querying-your-real-time-endpoint-expiry-time).
543 | 5. Users can also extend the endpoint uptime by sending a request to the `endpoint-expiry` API by providing the time in minutes the request body. For more information, refer to [Real-time Endpoint Management Functions - Extending your real-time endpoint expiry](#real-time-endpoint-management-functions---extending-your-real-time-endpoint-expiry-time).
544 | 6. You can also add a new endpoint to be managed by the endpoint manager for pre-existing Amazon SageMaker endpoint configurations. For more information, refer to [Real-time Endpoint Management Functions - Adding a new real-time endpoint](#real-time-endpoint-management-functions---adding-a-new-real-time-endpoint).
545 | ---
546 | ## To Do 
547 | - [x] Bug - if time is expired, extending the time will need to be greater than the different of current time + time required. Will need to add a check to see if time is expired, add time from now + time required. 
548 | - [x] Add support for multiple endpoints
549 | - [x] Add support for managing existing endpoints
550 | - [x] Add support for adding new endpoints via API
551 | - [x] Improve lambda error handling
552 | - [x] Add API gateway/AWS integration
553 | - [x] Add support for asynchronous invocation of Amazon SageMaker real-time endpoint
554 | - [ ] Add support for scheduled endpoints (i.e. M-F 9-5)
555 | - [ ] Add support for cognito users
556 | - [ ] Add support for expiry notifications
557 | - [ ] Add UI to manage endpoint expiry
558 | - [ ] Automatic shutdown of endpoint based on endpoint activity (i.e. <X% of usage)
559 | - [ ] Add configuration to use pre-defined dynamodb auth table or create new dynamodb table.
560 | - [ ] Extend async solution with lambda
561 | 
562 | ## References
563 | 
564 | Building an API Gateway with CDK - https://docs.aws.amazon.com/cdk/api/v2/python/aws_cdk.aws_apigateway/README.html#lambda-based-request-authorizer
565 | 
566 | 


--------------------------------------------------------------------------------
/app.py:
--------------------------------------------------------------------------------
 1 | """Deploy the SageMaker Endpoint Manager Stack"""
 2 | #!/usr/bin/env python3
 3 | import json
 4 | import aws_cdk as cdk
 5 | from stack.sagemaker_endpoint_manager_stack import SagemakerEndpointManagerStack
 6 | 
 7 | # Load model configurations from config file
 8 | with open('config/configs.json', encoding='UTF-8') as file:
 9 |     configs = json.load(file)
10 | 
11 | # Define region to deploy stack into
12 | environment = cdk.Environment(region=configs["region_name"])
13 | 
14 | app = cdk.App()
15 | 
16 | # Deploy the SageMaker Endpoint Manager Stack
17 | sem_stack = SagemakerEndpointManagerStack(app, "SagemakerEndpointManagerStack",
18 |                                          env=environment,
19 |                                          configs=configs
20 |                                          )
21 | 
22 | app.synth()
23 | 


--------------------------------------------------------------------------------
/cdk.json:
--------------------------------------------------------------------------------
 1 | {
 2 |   "app": "python3 app.py",
 3 |   "watch": {
 4 |     "include": [
 5 |       "**"
 6 |     ],
 7 |     "exclude": [
 8 |       "README.md",
 9 |       "cdk*.json",
10 |       "requirements*.txt",
11 |       "source.bat",
12 |       "**/__init__.py",
13 |       "python/__pycache__",
14 |       "tests"
15 |     ]
16 |   },
17 |   "context": {
18 |     "@aws-cdk/aws-lambda:recognizeLayerVersion": true,
19 |     "@aws-cdk/core:checkSecretUsage": true,
20 |     "@aws-cdk/core:target-partitions": [
21 |       "aws",
22 |       "aws-cn"
23 |     ],
24 |     "@aws-cdk-containers/ecs-service-extensions:enableDefaultLogDriver": true,
25 |     "@aws-cdk/aws-ec2:uniqueImdsv2TemplateName": true,
26 |     "@aws-cdk/aws-ecs:arnFormatIncludesClusterName": true,
27 |     "@aws-cdk/aws-iam:minimizePolicies": true,
28 |     "@aws-cdk/core:validateSnapshotRemovalPolicy": true,
29 |     "@aws-cdk/aws-codepipeline:crossAccountKeyAliasStackSafeResourceName": true,
30 |     "@aws-cdk/aws-s3:createDefaultLoggingPolicy": true,
31 |     "@aws-cdk/aws-sns-subscriptions:restrictSqsDescryption": true,
32 |     "@aws-cdk/aws-apigateway:disableCloudWatchRole": true,
33 |     "@aws-cdk/core:enablePartitionLiterals": true,
34 |     "@aws-cdk/aws-events:eventsTargetQueueSameAccount": true,
35 |     "@aws-cdk/aws-iam:standardizedServicePrincipals": true,
36 |     "@aws-cdk/aws-ecs:disableExplicitDeploymentControllerForCircuitBreaker": true,
37 |     "@aws-cdk/aws-iam:importedRoleStackSafeDefaultPolicyName": true,
38 |     "@aws-cdk/aws-s3:serverAccessLogsUseBucketPolicy": true,
39 |     "@aws-cdk/aws-route53-patters:useCertificate": true,
40 |     "@aws-cdk/customresources:installLatestAwsSdkDefault": false,
41 |     "@aws-cdk/aws-rds:databaseProxyUniqueResourceName": true,
42 |     "@aws-cdk/aws-codedeploy:removeAlarmsFromDeploymentGroup": true,
43 |     "@aws-cdk/aws-apigateway:authorizerChangeDeploymentLogicalId": true,
44 |     "@aws-cdk/aws-ec2:launchTemplateDefaultUserData": true,
45 |     "@aws-cdk/aws-secretsmanager:useAttachedSecretResourcePolicyForSecretTargetAttachments": true,
46 |     "@aws-cdk/aws-redshift:columnId": true,
47 |     "@aws-cdk/aws-stepfunctions-tasks:enableEmrServicePolicyV2": true,
48 |     "@aws-cdk/aws-ec2:restrictDefaultSecurityGroup": true,
49 |     "@aws-cdk/aws-apigateway:requestValidatorUniqueId": true,
50 |     "@aws-cdk/aws-kms:aliasNameRef": true
51 |   }
52 | }
53 | 


--------------------------------------------------------------------------------
/config/configs.json:
--------------------------------------------------------------------------------
 1 | {
 2 |     "project_prefix": "demo",
 3 |     "region_name": "us-east-1",
 4 |     "ddb_auth_table_name": "AuthTable",
 5 |     "jumpstart_models" :[
 6 |         {
 7 |             "name" : "Falcon40B",
 8 |             "model_id" : "huggingface-llm-falcon-40b-instruct-bf16",
 9 |             "inference_instance_type" : "ml.g5.12xlarge",
10 |             "inference_type": "realtime",
11 |             "schedule": {
12 |                 "initial_provision_minutes": 90
13 |             },
14 |             "integration": {
15 |                 "type": "lambda",
16 |                 "properties": {
17 |                     "lambda_src": "functions/falcon",
18 |                     "api_resource_name": "falcon"
19 |                 }
20 |             },
21 |             "async_api_enabled": true
22 |         },
23 |         {
24 |             "name" : "FlanT5",
25 |             "model_id" : "huggingface-text2text-flan-t5-xxl",
26 |             "inference_instance_type" : "ml.g5.12xlarge",
27 |             "inference_type": "realtime",
28 |             "schedule": {
29 |                 "initial_provision_minutes": 90
30 |             },
31 |             "integration": {
32 |                 "type": "api",
33 |                 "properties": {
34 |                     "api_resource_name": "flan"
35 |                 }
36 |             },
37 |             "async_api_enabled": true
38 |         }
39 |     ]
40 | }
41 | 


--------------------------------------------------------------------------------
/config/example-configs.json:
--------------------------------------------------------------------------------
 1 | {
 2 |     "project_prefix": "demo",
 3 |     "region_name": "us-east-1",
 4 |     "ddb_auth_table_name": "AuthTable",
 5 |     "jumpstart_models" :[
 6 |         {
 7 |             "name" : "Falcon40BPublic",
 8 |             "model_id" : "huggingface-llm-falcon-40b-instruct-bf16",
 9 |             "inference_instance_count": 2,
10 |             "inference_instance_type" : "ml.g5.12xlarge",
11 |             "inference_type": "realtime",
12 |             "public": true,
13 |             "schedule": {
14 |                 "initial_provision_minutes": 90
15 |             },
16 |             "integration": {
17 |                 "type": "lambda",
18 |                 "properties": {
19 |                     "lambda_src": "functions/falcon",
20 |                     "api_resource_name": "falcon"
21 |                 }
22 |             }
23 |         },
24 |         {
25 |             "name" : "FlanT5lambda",
26 |             "model_id" : "huggingface-text2text-flan-t5-xxl",
27 |             "inference_instance_type" : "ml.g5.12xlarge",
28 |             "inference_type": "realtime",
29 |             "schedule": {
30 |                 "initial_provision_minutes": 90
31 |             },
32 |             "integration": {
33 |                 "type": "lambda",
34 |                 "properties": {
35 |                     "lambda_src": "functions/flan",
36 |                     "api_resource_name": "flan"
37 |                 }
38 |             }
39 |         },
40 |         {
41 |             "name" : "FlanT5api",
42 |             "model_id" : "huggingface-text2text-flan-t5-xxl",
43 |             "inference_instance_type" : "ml.g5.12xlarge",
44 |             "inference_type": "realtime",
45 |             "schedule": {
46 |                 "initial_provision_minutes": 90
47 |             },
48 |             "integration": {
49 |                 "type": "api",
50 |                 "properties": {
51 |                     "api_resource_name": "flanapi"
52 |                 }
53 |             }
54 |         },
55 |         {
56 |             "name" : "FlanT5Async",
57 |             "model_id" : "huggingface-text2text-flan-t5-xxl",
58 |             "inference_instance_type" : "ml.g5.8xlarge",
59 |             "inference_type": "async"
60 |         }
61 |     ]
62 | }
63 | 


--------------------------------------------------------------------------------
/config/sagemaker-invoke.json:
--------------------------------------------------------------------------------
 1 | {
 2 |     "Comment": "Invoke SageMAker Endpoint",
 3 |     "StartAt": "InvokeEndpoint",
 4 |     "States": {
 5 |       "InvokeEndpoint": {
 6 |         "Type": "Task",
 7 |         "End": true,
 8 |         "Parameters": {
 9 |           "ContentType": "application/json",
10 |           "Body.$": "$.body",
11 |           "EndpointName.$": "$.endpointname"
12 |         },
13 |         "Resource": "arn:aws:states:::aws-sdk:sagemakerruntime:invokeEndpoint"
14 |       }
15 |     }
16 |   }


--------------------------------------------------------------------------------
/construct/sagemaker_async_endpoint_construct.py:
--------------------------------------------------------------------------------
 1 | """SageMaker Async Endpoint Construct"""
 2 | from typing import Optional
 3 | from aws_cdk import (
 4 |     aws_sagemaker as sagemaker,
 5 |     CfnOutput
 6 | )
 7 | from constructs import Construct
 8 | 
 9 | class SageMakerAsyncEndpointConstruct(Construct):
10 |     """Class representing a SageMaker Async Endpoint Construct"""
11 | 
12 |     def __init__(self, scope: Construct, construct_id: str,
13 |         project_prefix: str,
14 |         role_arn: str,
15 |         model_name: str,
16 |         model_bucket_name: Optional[str],
17 |         model_bucket_key: Optional[str],
18 |         model_docker_image: Optional[str],
19 |         variant_name: str,
20 |         variant_weight: int,
21 |         instance_count: int,
22 |         instance_type: str,
23 |         environment: dict,
24 |         success_topic: str,
25 |         error_topic: str,
26 |         s3_async_bucket: str,
27 |         deploy_enable: bool,
28 |         model_package_arn: Optional[str],
29 |         enable_network_isolation: Optional[bool]) -> None:
30 |         super().__init__(scope, construct_id)
31 | 
32 |         if model_package_arn is not None:
33 |             # Deploy model using model package arn
34 |             container = [
35 |                         sagemaker.CfnModel.ContainerDefinitionProperty(
36 |                                 model_package_name=model_package_arn
37 |                             )
38 |                         ]
39 |         else:
40 |             # Deploy model using docker image
41 |             container = [
42 |                         sagemaker.CfnModel.ContainerDefinitionProperty(
43 |                                     image= model_docker_image,
44 |                                     model_data_url= f"s3://{model_bucket_name}/{model_bucket_key}",
45 |                                     environment= environment
46 |                                 )
47 |                         ]
48 | 
49 |         model = sagemaker.CfnModel(self, f"{model_name}-Model",
50 |                     execution_role_arn= role_arn,
51 |                     containers=container,
52 |                     enable_network_isolation=enable_network_isolation
53 |             )
54 | 
55 |         config = sagemaker.CfnEndpointConfig(self, f"{model_name}-Config",
56 |             production_variants=[
57 |                 sagemaker.CfnEndpointConfig.ProductionVariantProperty(
58 |                     model_name= model.attr_model_name,
59 |                     variant_name= variant_name,
60 |                     initial_variant_weight= variant_weight,
61 |                     initial_instance_count= instance_count,
62 |                     instance_type= instance_type
63 |                 )
64 |             ],
65 |             async_inference_config=sagemaker.CfnEndpointConfig.AsyncInferenceConfigProperty(
66 |                 output_config=sagemaker.CfnEndpointConfig.AsyncInferenceOutputConfigProperty(
67 |                 notification_config=sagemaker.CfnEndpointConfig.AsyncInferenceNotificationConfigProperty(
68 |                     error_topic=error_topic,
69 |                     success_topic=success_topic
70 |                 ),
71 |                 s3_output_path=f"s3://{s3_async_bucket}/async_inference_output",
72 |                 s3_failure_path=f"s3://{s3_async_bucket}/async_inference_failure"
73 |                 ),
74 |                 # the properties below are optional
75 |                 client_config=sagemaker.CfnEndpointConfig.AsyncInferenceClientConfigProperty(
76 |                     max_concurrent_invocations_per_instance=4
77 |                 )
78 |             )
79 |         )
80 | 
81 |         self.deploy_enable = deploy_enable
82 |         if deploy_enable:
83 |             self.endpoint = sagemaker.CfnEndpoint(self, f"{model_name}-Endpoint",
84 |                                 endpoint_name= f"{project_prefix}-{model_name}-Endpoint",
85 |                                 endpoint_config_name= config.attr_endpoint_config_name
86 |             )
87 | 
88 |             CfnOutput(scope=self,id=f"{model_name}EndpointName", value=self.endpoint.endpoint_name)
89 | 
90 |     @property
91 |     def endpoint_name(self) -> str:
92 |         """Return endpoint name"""
93 |         return self.endpoint.attr_endpoint_name if self.deploy_enable else "not_yet_deployed"
94 | 


--------------------------------------------------------------------------------
/construct/sagemaker_endpoint_construct.py:
--------------------------------------------------------------------------------
 1 | """SageMaker Endpoint Construct"""
 2 | from typing import Optional
 3 | from aws_cdk import (
 4 |     aws_sagemaker as sagemaker,
 5 |     CfnOutput
 6 | )
 7 | from constructs import Construct
 8 | 
 9 | class SageMakerEndpointConstruct(Construct):
10 |     """Class representing a real-time SageMaker Endpoint Construct"""
11 | 
12 |     def __init__(self, scope: Construct, construct_id: str,
13 |         project_prefix: str,
14 |         role_arn: str,
15 |         model_name: str,
16 |         model_bucket_name: Optional[str],
17 |         model_bucket_key: Optional[str],
18 |         model_docker_image: Optional[str],
19 |         variant_name: str,
20 |         variant_weight: int,
21 |         instance_count: int,
22 |         instance_type: str,
23 |         environment: dict,
24 |         deploy_enable: bool,
25 |         model_package_arn: Optional[str],
26 |         enable_network_isolation: Optional[bool]) -> None:
27 |         super().__init__(scope, construct_id)
28 | 
29 |         # Do not use a custom named resource for models as these get replaced
30 |         if model_package_arn is not None:
31 |             # Deploy model using model package arn
32 |             container = [
33 |                         sagemaker.CfnModel.ContainerDefinitionProperty(
34 |                                 model_package_name=model_package_arn
35 |                             )
36 |                         ]
37 |         else:
38 |             # Deploy model using docker image
39 |             container = [
40 |                         sagemaker.CfnModel.ContainerDefinitionProperty(
41 |                                 image= model_docker_image,
42 |                                 model_data_url= f"s3://{model_bucket_name}/{model_bucket_key}",
43 |                                 environment= environment
44 |                             )
45 |                         ]
46 | 
47 |         model = sagemaker.CfnModel(self, f"{model_name}-Model",
48 |                 execution_role_arn= role_arn,
49 |                 containers=container,
50 |                 enable_network_isolation=enable_network_isolation
51 |             )
52 | 
53 |         self.config = sagemaker.CfnEndpointConfig(self, f"{model_name}-Config",
54 |             production_variants=[
55 |                 sagemaker.CfnEndpointConfig.ProductionVariantProperty(
56 |                     model_name= model.attr_model_name,
57 |                     variant_name= variant_name,
58 |                     initial_variant_weight= variant_weight,
59 |                     initial_instance_count= instance_count,
60 |                     instance_type= instance_type
61 |                 )
62 |             ]
63 |         )
64 | 
65 |         self.deploy_enable = deploy_enable
66 |         if deploy_enable:
67 |             self.endpoint = sagemaker.CfnEndpoint(self, f"{model_name}-Endpoint",
68 |                                 endpoint_name= f"{project_prefix}-{model_name}-Endpoint",
69 |                                 endpoint_config_name= self.config.attr_endpoint_config_name
70 |             )
71 | 
72 |             CfnOutput(scope=self,id=f"{model_name}EndpointName", value=self.endpoint.endpoint_name)
73 | 
74 |     @property
75 |     def endpoint_name(self) -> str:
76 |         """Return endpoint name"""
77 |         return self.endpoint.attr_endpoint_name if self.deploy_enable else "not_yet_deployed"
78 | 


--------------------------------------------------------------------------------
/docs/diagram.png:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/dalacan/sagemaker-endpoint-manager/eaefa6c702bf7c1633bcb69e41862bba04de0f44/docs/diagram.png


--------------------------------------------------------------------------------
/functions/auth/auth.py:
--------------------------------------------------------------------------------
  1 | """
  2 | Copyright 2015-2016 Amazon.com, Inc. or its affiliates. All Rights Reserved.
  3 | 
  4 | Licensed under the Apache License, Version 2.0 (the "License"). You may not use this file except in compliance with the License. A copy of the License is located at
  5 | 
  6 |      http://aws.amazon.com/apache2.0/
  7 | 
  8 | or in the "license" file accompanying this file. This file is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.
  9 | """
 10 | from __future__ import print_function
 11 | 
 12 | import os
 13 | import re
 14 | import boto3
 15 | import json
 16 | 
 17 | dynamodb = boto3.resource('dynamodb')
 18 | 
 19 | 
 20 | def handler(event, context):
 21 |     table = os.environ.get("TABLE_NAME")
 22 |     """Do not print the auth token unless absolutely necessary """
 23 |     #print("Client token: " + event['authorizationToken'])
 24 |     print("Method ARN: " + event['methodArn'])
 25 |     """validate the incoming token"""
 26 |     """and produce the principal user identifier associated with the token"""
 27 | 
 28 |     """this could be accomplished in a number of ways:"""
 29 |     """1. Call out to OAuth provider"""
 30 |     """2. Decode a JWT token inline"""
 31 |     """3. Lookup in a self-managed DB"""
 32 |     principalId = "user"
 33 | 
 34 |     """you can send a 401 Unauthorized response to the client by failing like so:"""
 35 |     """raise Exception('Unauthorized')"""
 36 | 
 37 |     """if the token is valid, a policy must be generated which will allow or deny access to the client"""
 38 | 
 39 |     """if access is denied, the client will recieve a 403 Access Denied response"""
 40 |     """if access is allowed, API Gateway will proceed with the backend integration configured on the method that was called"""
 41 | 
 42 |     """this function must generate a policy that is associated with the recognized principal user identifier."""
 43 |     """depending on your use case, you might store policies in a DB, or generate them on the fly"""
 44 | 
 45 |     """keep in mind, the policy is cached for 5 minutes by default (TTL is configurable in the authorizer)"""
 46 |     """and will apply to subsequent calls to any method/resource in the RestApi"""
 47 |     """made with the same token"""
 48 | 
 49 |     """the example policy below denies access to all resources in the RestApi"""
 50 |     tmp = event['methodArn'].split(':')
 51 |     apiGatewayArnTmp = tmp[5].split('/')
 52 |     awsAccountId = tmp[4]
 53 | 
 54 |     policy = AuthPolicy(principalId, awsAccountId)
 55 |     policy.restApiId = apiGatewayArnTmp[0]
 56 |     policy.region = tmp[3]
 57 |     policy.stage = apiGatewayArnTmp[1]
 58 | 
 59 |     # Check here
 60 |     token = event['headers']['Authorization']
 61 | 
 62 |     try:
 63 |         auth_table = dynamodb.Table(table)
 64 |         response = auth_table.get_item(
 65 |             Key={"token": token}
 66 |             )
 67 |     except Exception as e:
 68 |         print("Exception")
 69 |         policy.denyAllMethods()
 70 |     else:
 71 |         if 'Item' in response:
 72 |             print("Allowing access")
 73 |             policy.allowAllMethods()
 74 |         else:
 75 |             print("Not found, access denied.")
 76 |             policy.denyAllMethods()
 77 | 
 78 |     # Finally, build the policy
 79 |     authResponse = policy.build()
 80 | 
 81 |     # return authResponse
 82 |     return json.loads(json.dumps(authResponse, default=str))
 83 | 
 84 | class HttpVerb:
 85 |     GET     = "GET"
 86 |     POST    = "POST"
 87 |     PUT     = "PUT"
 88 |     PATCH   = "PATCH"
 89 |     HEAD    = "HEAD"
 90 |     DELETE  = "DELETE"
 91 |     OPTIONS = "OPTIONS"
 92 |     ALL     = "*"
 93 | 
 94 | class AuthPolicy(object):
 95 |     awsAccountId = ""
 96 |     """The AWS account id the policy will be generated for. This is used to create the method ARNs."""
 97 |     principalId = ""
 98 |     """The principal used for the policy, this should be a unique identifier for the end user."""
 99 |     version = "2012-10-17"
100 |     """The policy version used for the evaluation. This should always be '2012-10-17'"""
101 |     pathRegex = "^[/.a-zA-Z0-9-\*]+$"
102 |     """The regular expression used to validate resource paths for the policy"""
103 | 
104 |     """these are the internal lists of allowed and denied methods. These are lists
105 |     of objects and each object has 2 properties: A resource ARN and a nullable
106 |     conditions statement.
107 |     the build method processes these lists and generates the approriate
108 |     statements for the final policy"""
109 |     allowMethods = []
110 |     denyMethods = []
111 | 
112 |     restApiId = "<<restApiId>>"
113 |     """ Replace the placeholder value with a default API Gateway API id to be used in the policy. 
114 |     Beware of using '*' since it will not simply mean any API Gateway API id, because stars will greedily expand over '/' or other separators. 
115 |     See https://docs.aws.amazon.com/IAM/latest/UserGuide/reference_policies_elements_resource.html for more details. """    
116 | 
117 |     region = "<<region>>"
118 |     """ Replace the placeholder value with a default region to be used in the policy. 
119 |     Beware of using '*' since it will not simply mean any region, because stars will greedily expand over '/' or other separators. 
120 |     See https://docs.aws.amazon.com/IAM/latest/UserGuide/reference_policies_elements_resource.html for more details. """    
121 | 
122 |     stage = "<<stage>>"
123 |     """ Replace the placeholder value with a default stage to be used in the policy. 
124 |     Beware of using '*' since it will not simply mean any stage, because stars will greedily expand over '/' or other separators. 
125 |     See https://docs.aws.amazon.com/IAM/latest/UserGuide/reference_policies_elements_resource.html for more details. """
126 | 
127 |     def __init__(self, principal, awsAccountId):
128 |         self.awsAccountId = awsAccountId
129 |         self.principalId = principal
130 |         self.allowMethods = []
131 |         self.denyMethods = []
132 | 
133 |     def _addMethod(self, effect, verb, resource, conditions):
134 |         """Adds a method to the internal lists of allowed or denied methods. Each object in
135 |         the internal list contains a resource ARN and a condition statement. The condition
136 |         statement can be null."""
137 |         if verb != "*" and not hasattr(HttpVerb, verb):
138 |             raise NameError("Invalid HTTP verb " + verb + ". Allowed verbs in HttpVerb class")
139 |         resourcePattern = re.compile(self.pathRegex)
140 |         if not resourcePattern.match(resource):
141 |             raise NameError("Invalid resource path: " + resource + ". Path should match " + self.pathRegex)
142 | 
143 |         if resource[:1] == "/":
144 |             resource = resource[1:]
145 | 
146 |         resourceArn = ("arn:aws:execute-api:" +
147 |             self.region + ":" +
148 |             self.awsAccountId + ":" +
149 |             self.restApiId + "/" +
150 |             self.stage + "/" +
151 |             verb + "/" +
152 |             resource)
153 | 
154 |         if effect.lower() == "allow":
155 |             self.allowMethods.append({
156 |                 'resourceArn' : resourceArn,
157 |                 'conditions' : conditions
158 |             })
159 |         elif effect.lower() == "deny":
160 |             self.denyMethods.append({
161 |                 'resourceArn' : resourceArn,
162 |                 'conditions' : conditions
163 |             })
164 | 
165 |     def _getEmptyStatement(self, effect):
166 |         """Returns an empty statement object prepopulated with the correct action and the
167 |         desired effect."""
168 |         statement = {
169 |             'Action': 'execute-api:Invoke',
170 |             'Effect': effect[:1].upper() + effect[1:].lower(),
171 |             'Resource': []
172 |         }
173 | 
174 |         return statement
175 | 
176 |     def _getStatementForEffect(self, effect, methods):
177 |         """This function loops over an array of objects containing a resourceArn and
178 |         conditions statement and generates the array of statements for the policy."""
179 |         statements = []
180 | 
181 |         if len(methods) > 0:
182 |             statement = self._getEmptyStatement(effect)
183 | 
184 |             for curMethod in methods:
185 |                 if curMethod['conditions'] is None or len(curMethod['conditions']) == 0:
186 |                     statement['Resource'].append(curMethod['resourceArn'])
187 |                 else:
188 |                     conditionalStatement = self._getEmptyStatement(effect)
189 |                     conditionalStatement['Resource'].append(curMethod['resourceArn'])
190 |                     conditionalStatement['Condition'] = curMethod['conditions']
191 |                     statements.append(conditionalStatement)
192 | 
193 |             statements.append(statement)
194 | 
195 |         return statements
196 | 
197 |     def allowAllMethods(self):
198 |         """Adds a '*' allow to the policy to authorize access to all methods of an API"""
199 |         self._addMethod("Allow", HttpVerb.ALL, "*", [])
200 | 
201 |     def denyAllMethods(self):
202 |         """Adds a '*' allow to the policy to deny access to all methods of an API"""
203 |         self._addMethod("Deny", HttpVerb.ALL, "*", [])
204 | 
205 |     def allowMethod(self, verb, resource):
206 |         """Adds an API Gateway method (Http verb + Resource path) to the list of allowed
207 |         methods for the policy"""
208 |         self._addMethod("Allow", verb, resource, [])
209 | 
210 |     def denyMethod(self, verb, resource):
211 |         """Adds an API Gateway method (Http verb + Resource path) to the list of denied
212 |         methods for the policy"""
213 |         self._addMethod("Deny", verb, resource, [])
214 | 
215 |     def allowMethodWithConditions(self, verb, resource, conditions):
216 |         """Adds an API Gateway method (Http verb + Resource path) to the list of allowed
217 |         methods and includes a condition for the policy statement. More on AWS policy
218 |         conditions here: http://docs.aws.amazon.com/IAM/latest/UserGuide/reference_policies_elements.html#Condition"""
219 |         self._addMethod("Allow", verb, resource, conditions)
220 | 
221 |     def denyMethodWithConditions(self, verb, resource, conditions):
222 |         """Adds an API Gateway method (Http verb + Resource path) to the list of denied
223 |         methods and includes a condition for the policy statement. More on AWS policy
224 |         conditions here: http://docs.aws.amazon.com/IAM/latest/UserGuide/reference_policies_elements.html#Condition"""
225 |         self._addMethod("Deny", verb, resource, conditions)
226 | 
227 |     def build(self):
228 |         """Generates the policy document based on the internal lists of allowed and denied
229 |         conditions. This will generate a policy with two main statements for the effect:
230 |         one statement for Allow and one statement for Deny.
231 |         Methods that includes conditions will have their own statement in the policy."""
232 |         if ((self.allowMethods is None or len(self.allowMethods) == 0) and
233 |             (self.denyMethods is None or len(self.denyMethods) == 0)):
234 |             raise NameError("No statements defined for the policy")
235 | 
236 |         policy = {
237 |             'principalId' : self.principalId,
238 |             'policyDocument' : {
239 |                 'Version' : self.version,
240 |                 'Statement' : []
241 |             }
242 |         }
243 | 
244 |         policy['policyDocument']['Statement'].extend(self._getStatementForEffect("Allow", self.allowMethods))
245 |         policy['policyDocument']['Statement'].extend(self._getStatementForEffect("Deny", self.denyMethods))
246 | 
247 |         return policy
248 | 


--------------------------------------------------------------------------------
/functions/falcon/app.py:
--------------------------------------------------------------------------------
 1 | import os
 2 | import boto3
 3 | 
 4 | # grab environment variables
 5 | ENDPOINT_NAME = os.environ['ENDPOINT_NAME']
 6 | client = boto3.client('runtime.sagemaker')
 7 | 
 8 | def handler(event, context):
 9 | 
10 |     payload = event['body']
11 | 
12 |     try:
13 |         response = client.invoke_endpoint(
14 |             EndpointName=ENDPOINT_NAME, 
15 |             ContentType='application/json',
16 |             Accept='application/json',
17 |             Body=payload
18 |         )
19 | 
20 |         result = {
21 |             "statusCode": 200,
22 |             "headers": {
23 |                     'Content-Type': 'text/json'
24 |                         },
25 |             "body": response["Body"].read()
26 |         }
27 |     except Exception as e:
28 |         result = {
29 |             "statusCode": 500,
30 |             "headers": {
31 |                     'Content-Type': 'text/json'
32 |                         },
33 |             "body": str(e)
34 |         }
35 | 
36 |     return result
37 | 


--------------------------------------------------------------------------------
/functions/flan/app.py:
--------------------------------------------------------------------------------
 1 | import os
 2 | import json
 3 | import boto3
 4 | 
 5 | # grab environment variables
 6 | ENDPOINT_NAME = os.environ['ENDPOINT_NAME']
 7 | runtime= boto3.client('runtime.sagemaker')
 8 | 
 9 | def handler(event, context):
10 |     payload = {'text_inputs':'write a sentence to suggest providing a custom input for the model inference', 'max_length': 50, 'temperature': 0.0, 'seed': 321}
11 |     if event['body'] is not None :
12 |         body = event['body']
13 |     else:
14 |         body = json.dumps(payload)
15 | 
16 |     try:
17 |         response = runtime.invoke_endpoint(
18 |             EndpointName=ENDPOINT_NAME,
19 |             Body=body,
20 |             ContentType='application/json',
21 |             Accept='application/json'    )
22 | 
23 |         response = response["Body"].read().decode('utf-8')
24 | 
25 |         result = {
26 |             "statusCode": 200,
27 |             "headers": {
28 |                     'Content-Type': 'text/json'
29 |                         },
30 |             "body": response
31 |         }
32 |     except Exception as e:
33 |         result = {
34 |             "statusCode": 500,
35 |             "headers": {
36 |                     'Content-Type': 'text/json'
37 |                         },
38 |             "body": str(e)
39 |         }
40 |         
41 |     return result
42 | 


--------------------------------------------------------------------------------
/functions/start_stop_endpoint/app.py:
--------------------------------------------------------------------------------
 1 | import boto3
 2 | import botocore
 3 | from datetime import datetime
 4 | import json
 5 | 
 6 | sagemaker_client = boto3.client('sagemaker')
 7 | ssm_client = boto3.client("ssm")
 8 | 
 9 | def start_stop_endpoint(expiry_parameter_values):
10 |     expiry = datetime.strptime(expiry_parameter_values['expiry'], '%d-%m-%Y-%H-%M-%S')
11 |     now = datetime.utcnow()
12 |     
13 |     # Expired, delete endpoint
14 |     if expiry < now:
15 |         # Delete endpoint
16 |         print("Endpoint has expired, deleting endpoint")
17 |         try:
18 |             sagemaker_client.delete_endpoint(EndpointName=expiry_parameter_values['endpoint_name'])
19 |         except botocore.exceptions.ClientError as error:
20 |             print("Error deleting endpoint")
21 |             print(error)
22 |     else:
23 |         # Check if endpoint is expiring
24 |         print("Endpoint is not expiring")
25 |         try:
26 |             print("Checking if endpoint exists")
27 |             # Check endpoint exist
28 |             describe_response = sagemaker_client.describe_endpoint(
29 |                 EndpointName=expiry_parameter_values['endpoint_name']
30 |             )
31 | 
32 |             # Check if endpoint creation failed, it it has, delete it so that it can be created
33 |             if describe_response['EndpointStatus'] == 'Failed':
34 |                 print("Endpoint creation failed, deleting endpoint")
35 |                 sagemaker_client.delete_endpoint(EndpointName=expiry_parameter_values['endpoint_name'])
36 |         except botocore.exceptions.ClientError as error:
37 |             # Endpoint does not exist, create endpoint
38 |             if error.response['Error']['Code'] == 'ValidationException':
39 |                 print("Creating endpoint")
40 |                 try:
41 |                     create_endpoint_response = create_endpoint(expiry_parameter_values['endpoint_name'], expiry_parameter_values['endpoint_config_name'])
42 |                 except botocore.exceptions.ClientError as error:
43 |                     print("Error creating endpoint")
44 |             else:
45 |                 print("Error describing endpoint")
46 |                 print(error)
47 | 
48 | def create_endpoint(endpoint_name, endpoint_config_name):
49 |     return sagemaker_client.create_endpoint(
50 |                                         EndpointName=endpoint_name, 
51 |                                         EndpointConfigName=endpoint_config_name)
52 | 
53 | def handler(event, context):
54 |     # Get a list of endpoint expiry parameters
55 |     response = ssm_client.get_parameters_by_path(
56 |         Path="/sagemaker/endpoint/expiry/",
57 |         Recursive=True)
58 |     result = response["Parameters"]
59 | 
60 |     
61 |     while "NextToken" in response:
62 |         response = ssm_client.get_parameters_by_path(
63 |             Path="/sagemaker/endpoint/expiry/",
64 |             Recursive=True,
65 |             NextToken=result["NextToken"])
66 |         result.extend(response["Parameters"])
67 | 
68 |     # Process each endpoint expiry configuration
69 |     for parameter in result:
70 |         print("Processing endpoint")
71 |         # endpoint_name = parameter['Name'].split("/")[-1]
72 |         parameter_values = json.loads(parameter['Value'])
73 |         start_stop_endpoint(parameter_values) 


--------------------------------------------------------------------------------
/functions/update_expiry/app.py:
--------------------------------------------------------------------------------
  1 | import boto3
  2 | import botocore
  3 | import json
  4 | 
  5 | from datetime import datetime, timedelta
  6 | 
  7 | ssm_client = boto3.client("ssm")
  8 | 
  9 | def get_expiry(expiry_parameter_values):
 10 |     expiry = datetime.strptime(expiry_parameter_values['expiry'], '%d-%m-%Y-%H-%M-%S')
 11 |     now = datetime.utcnow()
 12 |     time_left = expiry - now
 13 | 
 14 | 
 15 |     endpoint_expiry_info = {
 16 |         "EndpointName": expiry_parameter_values['endpoint_name'],
 17 |         "EndpointExpiry ": expiry_parameter_values['expiry'],
 18 |         "TimeLeft": str(time_left)
 19 |     }
 20 | 
 21 |     return endpoint_expiry_info
 22 | 
 23 | def get_endpoint_expiry_info(event):
 24 |     if event['queryStringParameters'] is not None and 'EndpointName' in event['queryStringParameters']:
 25 |         print("Getting specific endpoint")
 26 |         # Get expiry
 27 |         expiry_parameter = ssm_client.get_parameter(
 28 |             Name=f"/sagemaker/endpoint/expiry/{event['queryStringParameters']['EndpointName']}",
 29 |             WithDecryption=False)
 30 |         expiry_parameter_values = json.loads(expiry_parameter['Parameter']['Value'])
 31 |         
 32 |         endpoint_expiry_info = get_expiry(expiry_parameter_values)
 33 | 
 34 |         response =  {
 35 |             "statusCode": 200,
 36 |             "headers": {
 37 |                 "Content-Type": "application/json"
 38 |             },
 39 |             "body": json.dumps(endpoint_expiry_info)
 40 |         }
 41 |     else:
 42 |         print("Getting list of endpoint expiry")
 43 |         # Get a list of endpoint expiry parameters
 44 |         ssm_response = ssm_client.get_parameters_by_path(
 45 |             Path="/sagemaker/endpoint/expiry/",
 46 |             Recursive=True)
 47 |         result = ssm_response["Parameters"]
 48 | 
 49 |         
 50 |         while "NextToken" in ssm_response:
 51 |             ssm_response = ssm_client.get_parameters_by_path(
 52 |                 Path="/sagemaker/endpoint/expiry/",
 53 |                 Recursive=True,
 54 |                 NextToken=result["NextToken"])
 55 |             result.extend(ssm_response["Parameters"])
 56 | 
 57 |         # Process each endpoint expiry configuration
 58 |         endpoint_expiry_info = []
 59 |         for parameter in result:
 60 |             print("Processing endpoint")
 61 |             parameter_values = json.loads(parameter['Value'])
 62 |             endpoint_expiry_info.append(get_expiry(parameter_values))
 63 | 
 64 |         response =  {
 65 |             "statusCode": 200,
 66 |             "headers": {
 67 |                 "Content-Type": "application/json"
 68 |             },
 69 |             "body": json.dumps(endpoint_expiry_info)
 70 |         }
 71 | 
 72 |     return response
 73 | 
 74 | def create_endpoint_config(endpoint_name, endpoint_config_name, provision_minutes):
 75 |     now = datetime.utcnow()
 76 |     expiry = now + timedelta(minutes=provision_minutes)
 77 |     expiry_str = expiry.strftime("%d-%m-%Y-%H-%M-%S")
 78 | 
 79 |     expiry_ssm_value = {
 80 |         "expiry": expiry_str,
 81 |         "endpoint_name": endpoint_name,
 82 |         "endpoint_config_name": endpoint_config_name
 83 |     }
 84 | 
 85 |     # Add new parameter
 86 |     ssm_response = ssm_client.put_parameter(
 87 |         Name=f"/sagemaker/endpoint/expiry/{endpoint_name}",
 88 |         Type="String",
 89 |         Overwrite=True,
 90 |         Value=json.dumps(expiry_ssm_value)
 91 |     )
 92 | 
 93 |     return provision_minutes, expiry_str
 94 | 
 95 | def update_endpoint_config(endpoint_name, provision_minutes, expiry_parameter_values):
 96 |     current_expiry = datetime.strptime(expiry_parameter_values['expiry'], '%d-%m-%Y-%H-%M-%S')
 97 | 
 98 |     # Check if current expiry is in the past
 99 |     now = datetime.utcnow()
100 |     if current_expiry < now:
101 |         current_expiry = now
102 | 
103 |     expiry = current_expiry + timedelta(minutes=provision_minutes)
104 |     expiry_str = expiry.strftime("%d-%m-%Y-%H-%M-%S")
105 | 
106 |     time_left = expiry - now
107 | 
108 |     expiry_parameter_values['expiry'] = expiry_str
109 | 
110 |     # Update parameter
111 |     ssm_response = ssm_client.put_parameter(
112 |         Name=f"/sagemaker/endpoint/expiry/{endpoint_name}",
113 |         Overwrite=True,
114 |         Value=json.dumps(expiry_parameter_values)
115 |     )
116 | 
117 |     return time_left, expiry_str
118 | 
119 | def create_update_endpoint_expiry(event):
120 |     if event['body'] is not None :
121 |         # Update expiry
122 |         body = json.loads(event["body"])
123 | 
124 |         if 'EndpointName' not in body:
125 |             response = {
126 |                 "statusCode": 400,
127 |                 "body": json.dumps({"error": "EndpointName required"})
128 |             }
129 |             return response
130 | 
131 |         try:
132 |             endpoint_name = body['EndpointName']
133 |             expiry_parameter = ssm_client.get_parameter(
134 |                     Name=f"/sagemaker/endpoint/expiry/{endpoint_name}",
135 |                     WithDecryption=False)
136 |             
137 |             expiry_parameter_values = json.loads(expiry_parameter['Parameter']['Value'])
138 | 
139 |             print("Updating endpoint")
140 |             time_left, expiry_str = update_endpoint_config(endpoint_name, body['minutes'], expiry_parameter_values)
141 | 
142 |             response =  {
143 |                         "statusCode": 200,
144 |                         "headers": {
145 |                             "Content-Type": "application/json"
146 |                         },
147 |                         "body": json.dumps({
148 |                             "EndpointName": endpoint_name,
149 |                             "EndpointExpiry ": expiry_str,
150 |                             "TimeLeft": str(time_left)
151 |                         })
152 |                     }
153 |         except botocore.exceptions.ClientError as error:
154 |             if error.response['Error']['Code'] == 'ParameterNotFound':
155 |                 if 'EndpointConfigName' in body:
156 |                     print("Creating new endpoint config")
157 |                     time_left, expiry_str = create_endpoint_config(endpoint_name, body['EndpointConfigName'], body['minutes'])
158 | 
159 |                     response =  {
160 |                         "statusCode": 200,
161 |                         "headers": {
162 |                             "Content-Type": "application/json"
163 |                         },
164 |                         "body": json.dumps({
165 |                             "EndpointName": endpoint_name,
166 |                             "EndpointExpiry ": expiry_str,
167 |                             "TimeLeft": str(time_left)
168 |                         })
169 |                     }
170 |                 else:               
171 |                     response = {
172 |                         "statusCode": 400,
173 |                         "body": json.dumps({"error": "EndpointName not found/Endpoint config name required"})
174 |                     }
175 |             else:
176 |                 response = {
177 |                     "statusCode": 400,
178 |                     "body": json.dumps({"error": "Error retrieving endpoint"})
179 |                 }
180 |     else:
181 |         response = {
182 |             "statusCode": 400,
183 |             "body": json.dumps({"error": "Body required"})
184 |         }
185 | 
186 |     return response
187 | 
188 | def handler(event, context):
189 |     http_method = event['httpMethod']
190 | 
191 |     if http_method == "GET":
192 |         response = get_endpoint_expiry_info(event)
193 |     elif http_method == "POST":
194 |         response = create_update_endpoint_expiry(event)
195 |     else:
196 |         # Return an error message for unsupported methods
197 |         response = {
198 |             "statusCode": 405,
199 |             "body": json.dumps({"error": "Method not allowed"})
200 |         }
201 |     return response


--------------------------------------------------------------------------------
/requirements-dev.txt:
--------------------------------------------------------------------------------
1 | pytest==6.2.5
2 | 


--------------------------------------------------------------------------------
/requirements.txt:
--------------------------------------------------------------------------------
1 | aws-cdk-lib==2.90.0
2 | constructs>=10.0.0,<11.0.0
3 | 
4 | sagemaker>=2.173.0
5 | boto3


--------------------------------------------------------------------------------
/source.bat:
--------------------------------------------------------------------------------
 1 | @echo off
 2 | 
 3 | rem The sole purpose of this script is to make the command
 4 | rem
 5 | rem     source .venv/bin/activate
 6 | rem
 7 | rem (which activates a Python virtualenv on Linux or Mac OS X) work on Windows.
 8 | rem On Windows, this command just runs this batch file (the argument is ignored).
 9 | rem
10 | rem Now we don't need to document a Windows command for activating a virtualenv.
11 | 
12 | echo Executing .venv\Scripts\activate.bat for you
13 | .venv\Scripts\activate.bat
14 | 


--------------------------------------------------------------------------------
/stack/__init__.py:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/dalacan/sagemaker-endpoint-manager/eaefa6c702bf7c1633bcb69e41862bba04de0f44/stack/__init__.py


--------------------------------------------------------------------------------
/stack/api_stack.py:
--------------------------------------------------------------------------------
 1 | from aws_cdk import (
 2 |     RemovalPolicy,
 3 |     NestedStack,
 4 |     aws_lambda as _lambda,
 5 |     aws_apigateway as apigateway,
 6 |     aws_dynamodb as dynamodb,
 7 |     aws_iam as iam
 8 | )
 9 | 
10 | from constructs import Construct
11 | 
12 | class APIStack(NestedStack):
13 | 
14 |     def __init__(self, scope: Construct, construct_id: str, configs, **kwargs) -> None:
15 |         super().__init__(scope, construct_id, **kwargs)
16 | 
17 |         # DynamoDB table for authentication
18 |         table_name = configs.get("ddb_auth_table_name", "AuthTable")
19 | 
20 |         auth_db = dynamodb.Table(self, "AuthTable",
21 |                                  table_name=table_name,
22 |                                  partition_key=dynamodb.Attribute(name="token", type=dynamodb.AttributeType.STRING),
23 |                                  removal_policy=RemovalPolicy.DESTROY)
24 | 
25 |         # Auth handler function
26 |         auth_handler = _lambda.Function(self, "AuthHandler",
27 |                                         code=_lambda.Code.from_asset('functions/auth'),
28 |                         runtime=_lambda.Runtime.PYTHON_3_9,
29 |                         handler='auth.handler',
30 |                         environment={
31 |                             "TABLE_NAME": table_name
32 |                             })
33 |         
34 |         auth_db.grant_read_data(auth_handler)
35 |                                         
36 |         self.api_authorizer = apigateway.RequestAuthorizer(self, "APIAuthorizer",
37 |             handler=auth_handler,
38 |             identity_sources=[apigateway.IdentitySource.header("Authorization")]
39 |         )
40 | 
41 |         self.api = apigateway.RestApi(self, "FoundationModelAPI",
42 |                   rest_api_name="Foundation Model API Service",
43 |                   description="This service serves all the foundation models.")
44 | 
45 |         self.api_gateway_role = iam.Role(
46 |             self,
47 |             "ApiGatewayRole",
48 |             assumed_by=iam.ServicePrincipal("apigateway.amazonaws.com"),
49 |         )
50 | 


--------------------------------------------------------------------------------
/stack/endpoint_manager_stack.py:
--------------------------------------------------------------------------------
 1 | from aws_cdk import (
 2 |     NestedStack,
 3 |     Duration,
 4 |     aws_iam as iam,
 5 |     aws_ssm as ssm,
 6 |     aws_lambda as _lambda,
 7 |     aws_events as events,
 8 |     aws_events_targets as targets,
 9 |     aws_apigateway as apigateway
10 | )
11 | 
12 | from constructs import Construct
13 | 
14 | class EndpointManagerStack(NestedStack):
15 | 
16 |     def __init__(self, scope: Construct, construct_id: str, api_stack, **kwargs) -> None:
17 |         super().__init__(scope, construct_id, **kwargs)
18 | 
19 |         # Create endpoint manager lambdas
20 |         start_endpoint_handler = _lambda.Function(self, f"StartEndpointHandler",
21 |                 runtime=_lambda.Runtime.PYTHON_3_9,
22 |                 code=_lambda.Code.from_asset("functions/start_stop_endpoint"),
23 |                 handler="app.handler",
24 |                 timeout=Duration.seconds(30))
25 | 
26 |         # Add policy to lambda to create endpoint
27 |         start_endpoint_handler.add_to_role_policy(iam.PolicyStatement(
28 |             effect=iam.Effect.ALLOW,
29 |             actions=["sagemaker:CreateEndpoint", "sagemaker:DeleteEndpoint", "sagemaker:DescribeEndpoint"],
30 |             resources=[
31 |                 "*"
32 |             ],
33 |         ))
34 | 
35 |         ssm_arn = f"arn:aws:ssm:{self.region}:{self.account}:parameter/sagemaker/endpoint/expiry/*"
36 |         
37 |         # Add SSM read access
38 |         start_endpoint_handler.add_to_role_policy(iam.PolicyStatement(
39 |             effect=iam.Effect.ALLOW,
40 |             actions=["ssm:DescribeParameters", "ssm:GetParameter", "ssm:GetParameterHistory", "ssm:GetParameters", "ssm:GetParametersByPath"],
41 |             resources=[
42 |                 ssm_arn
43 |             ],
44 |         ))
45 | 
46 |         start_stop_endpoint_rule = events.Rule(self, 'eventStartStopLambdaRule',
47 |                                            description='Start/Stop Endpoint Lambda Rule',
48 |                                            schedule=events.Schedule.rate(Duration.minutes(1)),
49 |                                            targets=[targets.LambdaFunction(handler=start_endpoint_handler)])
50 |         
51 |         update_expiry_handler = _lambda.Function(self, f"UpdateExpiryHandler",
52 |                 runtime=_lambda.Runtime.PYTHON_3_9,
53 |                 code=_lambda.Code.from_asset("functions/update_expiry"),
54 |                 handler="app.handler",
55 |                 timeout=Duration.seconds(30))
56 | 
57 |         # Add SSM read/write policy
58 |         update_expiry_handler.add_to_role_policy(iam.PolicyStatement(
59 |             effect=iam.Effect.ALLOW,
60 |             actions=["ssm:DescribeParameters", "ssm:GetParameter", "ssm:GetParameterHistory", "ssm:GetParameters", "ssm:PutParameter", "ssm:GetParametersByPath"],
61 |             resources=[
62 |                 ssm_arn
63 |             ],
64 |         ))
65 | 
66 |         # Add lambda to api gateway
67 |         post_update_expiry_integration = apigateway.LambdaIntegration(update_expiry_handler,
68 |                                                                   request_templates={"application/json": '{ "statusCode": "200" }'})
69 |         # Add lambda to api
70 |         resource = api_stack.api.root.add_resource('endpoint-expiry')
71 |         resource.add_method("POST", post_update_expiry_integration, authorizer=api_stack.api_authorizer)
72 |         resource.add_method("GET", post_update_expiry_integration, authorizer=api_stack.api_authorizer)


--------------------------------------------------------------------------------
/stack/foundation_model_stack.py:
--------------------------------------------------------------------------------
  1 | from aws_cdk import (
  2 |     NestedStack,
  3 |     Duration,
  4 |     aws_iam as iam,
  5 |     aws_ssm as ssm,
  6 |     aws_lambda as _lambda,
  7 |     aws_apigateway as apigateway,
  8 |     aws_sns as sns,
  9 |     aws_s3 as s3
 10 | )
 11 | 
 12 | from constructs import Construct
 13 | from construct.sagemaker_endpoint_construct import SageMakerEndpointConstruct
 14 | from construct.sagemaker_async_endpoint_construct import SageMakerAsyncEndpointConstruct
 15 | 
 16 | import json
 17 | from datetime import datetime, timedelta
 18 | from stack.util import merge_env
 19 | 
 20 | from utils.sagemaker_helper import (
 21 |     get_sagemaker_uris,
 22 |     sagemaker_env,
 23 |     get_model_spec,
 24 |     get_model_package_arn,
 25 |     enable_network_isolation
 26 | )
 27 | 
 28 | from stack.stepfunction_stack import StepFunctionStack
 29 | 
 30 | class FoundationModelStack(NestedStack):
 31 | 
 32 |     def __init__(self, scope: Construct, construct_id: str, configs, api_stack, **kwargs) -> None:
 33 |         super().__init__(scope, construct_id, **kwargs)
 34 | 
 35 |         step_function_enabled_endpoints = []
 36 | 
 37 |         # Create policies for model
 38 |         role = iam.Role(self, "Gen-AI-SageMaker-Policy", assumed_by=iam.ServicePrincipal("sagemaker.amazonaws.com"))
 39 |         role.add_managed_policy(iam.ManagedPolicy.from_aws_managed_policy_name("AmazonS3FullAccess"))
 40 |         
 41 |         sts_policy = iam.Policy(self, "sm-deploy-policy-sts",
 42 |                                     statements=[iam.PolicyStatement(
 43 |                                         effect=iam.Effect.ALLOW,
 44 |                                         actions=[
 45 |                                             "sts:AssumeRole"
 46 |                                           ],
 47 |                                         resources=["*"]
 48 |                                     )]
 49 |                                 )
 50 | 
 51 |         logs_policy = iam.Policy(self, "sm-deploy-policy-logs",
 52 |                                     statements=[iam.PolicyStatement(
 53 |                                         effect=iam.Effect.ALLOW,
 54 |                                         actions=[
 55 |                                             "cloudwatch:PutMetricData",
 56 |                                             "logs:CreateLogStream",
 57 |                                             "logs:PutLogEvents",
 58 |                                             "logs:CreateLogGroup",
 59 |                                             "logs:DescribeLogStreams",
 60 |                                             "ecr:GetAuthorizationToken"
 61 |                                           ],
 62 |                                         resources=["*"]
 63 |                                     )]
 64 |                                 )
 65 |         
 66 |         ecr_policy = iam.Policy(self, "sm-deploy-policy-ecr",
 67 |                                     statements=[iam.PolicyStatement(
 68 |                                         effect=iam.Effect.ALLOW,
 69 |                                         actions=[
 70 |                                             "ecr:*",
 71 |                                           ],
 72 |                                         resources=["*"]
 73 |                                     )]
 74 |                                 )
 75 |                                 
 76 |         role.attach_inline_policy(sts_policy)
 77 |         role.attach_inline_policy(logs_policy)
 78 |         role.attach_inline_policy(ecr_policy)
 79 | 
 80 |         # Deploy jumpstart models
 81 |         for model in configs.get("jumpstart_models", []):
 82 |             # Get model info
 83 |             model_info = get_sagemaker_uris(model_id=model["model_id"],
 84 |                                         instance_type=model["inference_instance_type"],
 85 |                                         region_name=configs["region_name"])
 86 | 
 87 |             # Get model default environment parameters
 88 |             model_env = sagemaker_env(model_id=model["model_id"],
 89 |                                     region=configs["region_name"],
 90 |                                     model_version="*")
 91 | 
 92 |             # Get jumpstart model package arn if available
 93 |             model_specs = get_model_spec(
 94 |                                 model_id=model["model_id"],
 95 |                                 model_version="*",
 96 |                                 region=configs["region_name"]
 97 |             )
 98 | 
 99 |             model_package_arn = get_model_package_arn(
100 |                                 model_specs=model_specs,
101 |                                 region=configs["region_name"]
102 |             )
103 | 
104 |             is_network_isolation_enabled = enable_network_isolation(model_specs=model_specs)
105 | 
106 |             if model["inference_type"] == "realtime":
107 |                 # Create real-time endpoint
108 |                 # environment = {
109 |                 #                     "MODEL_CACHE_ROOT": "/opt/ml/model",
110 |                 #                     "SAGEMAKER_ENV": "1",
111 |                 #                     "SAGEMAKER_MODEL_SERVER_TIMEOUT": "3600",
112 |                 #                     "SAGEMAKER_MODEL_SERVER_WORKERS": "1",
113 |                 #                     "SAGEMAKER_CONTAINER_LOG_LEVEL": "20",
114 |                 #                     "SAGEMAKER_PROGRAM": "inference.py",
115 |                 #                     "SAGEMAKER_REGION": model_info["region_name"],
116 |                 #                     "SAGEMAKER_SUBMIT_DIRECTORY": "/opt/ml/model/code",
117 |                 #                 }
118 |                 environment = {
119 |                                     "MODEL_CACHE_ROOT": "/opt/ml/model",
120 |                                     "SAGEMAKER_ENV": "1",
121 |                                     "SAGEMAKER_MODEL_SERVER_TIMEOUT": "3600",
122 |                                     "SAGEMAKER_CONTAINER_LOG_LEVEL": "20",
123 |                                     "SAGEMAKER_PROGRAM": "inference.py",
124 |                                     "SAGEMAKER_REGION": model_info["region_name"],
125 |                                     "SAGEMAKER_SUBMIT_DIRECTORY": "/opt/ml/model/code",
126 |                                 }
127 | 
128 |                 environment = merge_env(environment, model_env)
129 |                 
130 |                 endpoint = SageMakerEndpointConstruct(self, f'FoundationModelEndpoint-{model["name"]}',
131 |                                             project_prefix = configs["project_prefix"],
132 |                                             
133 |                                             role_arn= role.role_arn,
134 | 
135 |                                             model_name = model["name"],
136 |                                             model_bucket_name = model_info["model_bucket_name"],
137 |                                             model_bucket_key = model_info["model_bucket_key"],
138 |                                             model_docker_image = model_info["model_docker_image"],
139 | 
140 |                                             variant_name = "AllTraffic",
141 |                                             variant_weight = 1,
142 |                                             instance_count = model.get("inference_instance_count", 1),
143 |                                             instance_type = model_info["instance_type"],
144 | 
145 |                                             environment = environment,
146 |                                             deploy_enable = False,
147 |                                             model_package_arn=model_package_arn,
148 |                                             enable_network_isolation=is_network_isolation_enabled
149 |                 )
150 |                 
151 |                 endpoint.node.add_dependency(role)
152 |                 endpoint.node.add_dependency(sts_policy)
153 |                 endpoint.node.add_dependency(logs_policy)
154 |                 endpoint.node.add_dependency(ecr_policy)
155 | 
156 |                 endpoint_name = f'{configs["project_prefix"]}-{model["name"]}-Endpoint'
157 | 
158 |                 # Set endpoint expiry
159 |                 now = datetime.utcnow()
160 |                 expiry = now + timedelta(minutes=model["schedule"]["initial_provision_minutes"])
161 | 
162 |                 expiry_ssm_value = {
163 |                     "expiry": expiry.strftime("%d-%m-%Y-%H-%M-%S"),
164 |                     "endpoint_name": endpoint_name,
165 |                     "endpoint_config_name": endpoint.config.attr_endpoint_config_name
166 |                 }
167 | 
168 |                 # Create default SSM parameter to manage endpoint
169 |                 expiry_ssm = ssm.StringParameter(self, 
170 |                                                 f"{endpoint_name}-expiry", 
171 |                                                 parameter_name=f"/sagemaker/endpoint/expiry/{endpoint_name}", 
172 |                                                 string_value=json.dumps(expiry_ssm_value))
173 |                 
174 |                 endpoint_arn = f'arn:aws:sagemaker:{self.region}:{self.account}:endpoint/{endpoint_name.lower()}'
175 |                 resource_name = model["integration"]["properties"]["api_resource_name"]
176 | 
177 |                 if model.get("async_api_enabled", False):
178 |                     step_function_enabled_endpoints.append(endpoint_arn)
179 |                 
180 |                 # Check integration type
181 |                 if model["integration"]["type"] == "lambda":
182 |                     # Add lambda/api integration
183 |                     app_handler = _lambda.Function(self, f"{resource_name}Handler",
184 |                     runtime=_lambda.Runtime.PYTHON_3_9,
185 |                     code=_lambda.Code.from_asset(model["integration"]["properties"]["lambda_src"]),
186 |                     handler="app.handler",
187 |                     timeout=Duration.seconds(180),
188 |                     environment={
189 |                         "ENDPOINT_NAME": endpoint_name,
190 |                     })
191 |             
192 |                     # Add sagemaker invoke permissions    
193 |                     app_handler.add_to_role_policy(iam.PolicyStatement(
194 |                         effect=iam.Effect.ALLOW,
195 |                         actions=["sagemaker:InvokeEndpoint"],
196 |                         resources=[endpoint_arn],
197 |                     ))
198 | 
199 |                     post_model_integration = apigateway.LambdaIntegration(app_handler,
200 |                                                                             request_templates={"application/json": '{ "statusCode": "200" }'})
201 |                     # Add lambda to api
202 |                     resource = api_stack.api.root.add_resource(resource_name)
203 |                     resource.add_method(
204 |                         "POST",
205 |                         post_model_integration,
206 |                         authorizer=None if model.get("public") else api_stack.api_authorizer,
207 |                     )
208 |                 elif model["integration"]["type"] == "api":
209 |                     # Add permission to invoke endpoint
210 |                     api_stack.api_gateway_role.add_to_policy(iam.PolicyStatement(
211 |                                                         effect=iam.Effect.ALLOW,
212 |                                                         actions=["sagemaker:InvokeEndpoint"],
213 |                                                         resources=[endpoint_arn],
214 |                                                     )
215 |                                                 )
216 |                     
217 |                     # Add api integration/aws integration
218 |                     resource = api_stack.api.root.add_resource(resource_name)
219 |                     post_model_integration = apigateway.AwsIntegration(
220 |                                                         service="runtime.sagemaker",
221 |                                                         integration_http_method="POST",
222 |                                                         path=f"endpoints/{endpoint_name}/invocations",
223 |                                                         options=apigateway.IntegrationOptions(
224 |                                                             request_parameters={
225 |                                                                 "integration.request.header.Content-Type": "method.request.header.Content-Type",
226 |                                                                 "integration.request.header.Accept": "method.request.header.Accept",
227 |                                                                 "integration.request.header.X-Amzn-SageMaker-Custom-Attributes": "method.request.header.X-Amzn-SageMaker-Custom-Attributes"
228 |                                                             },
229 |                                                             credentials_role=api_stack.api_gateway_role,
230 |                                                             integration_responses=[
231 |                                                                 apigateway.IntegrationResponse(
232 |                                                                     status_code="200",
233 |                                                                     response_templates={
234 |                                                                         "application/json": "$input.json('$')"
235 |                                                                     },
236 |                                                                 ),
237 |                                                                 apigateway.IntegrationResponse(
238 |                                                                     status_code="400",
239 |                                                                     selection_pattern="4\d{2}",
240 |                                                                     response_templates={
241 |                                                                         "application/json": '{ "error": $input.path("$.OriginalMessage") }'
242 |                                                                     },
243 |                                                                 ),
244 |                                                                 apigateway.IntegrationResponse(
245 |                                                                     status_code="500",
246 |                                                                     selection_pattern="5\d{2}",
247 |                                                                     response_templates={
248 |                                                                         "application/json": '{ "error": $input.path("$.OriginalMessage") }'
249 |                                                                     },
250 |                                                                 ),
251 |                                                             ],
252 |                                                         ),
253 |                                                     )
254 |                     resource.add_method("POST", 
255 |                                         post_model_integration, 
256 |                                         authorizer=None if model.get("public") else api_stack.api_authorizer,
257 |                                         request_parameters={
258 |                                                                 "method.request.header.Content-Type": True,
259 |                                                                 "method.request.header.Accept": True,
260 |                                                                 "method.request.header.X-Amzn-SageMaker-Custom-Attributes": False
261 |                                                             },
262 |                                                             method_responses=[
263 |                                                                 apigateway.MethodResponse(status_code="200"),
264 |                                                                 apigateway.MethodResponse(
265 |                                                                     status_code="400",
266 |                                                                     response_models={
267 |                                                                         "application/json": apigateway.Model.ERROR_MODEL
268 |                                                                     },
269 |                                                                 ),
270 |                                                                 apigateway.MethodResponse(
271 |                                                                     status_code="500",
272 |                                                                     response_models={
273 |                                                                         "application/json": apigateway.Model.ERROR_MODEL
274 |                                                                     },
275 |                                                                 ),
276 |                                                             ]
277 |                                             )
278 |                     
279 |             elif model["inference_type"] == "async":
280 |                 # Create async endpoint
281 | 
282 |                 # Create sns success and error topic
283 |                 success_topic = sns.Topic(self, f'{model["name"]}-SuccessTopic',
284 |                                             display_name=f'{model["name"]}-SuccessTopic')
285 |                 
286 |                 error_topic = sns.Topic(self, f'{model["name"]}-ErrorTopic',
287 |                                             display_name=f'{model["name"]}-ErrorTopic')
288 |                 
289 |                 sns_policy = iam.Policy(self, "sm-deploy-policy-sns",
290 |                                             statements=[iam.PolicyStatement(
291 |                                                 effect=iam.Effect.ALLOW,
292 |                                                 actions=["sns:Publish"],
293 |                                                 resources=[success_topic.topic_arn, error_topic.topic_arn]
294 |                                             )]
295 |                 )
296 | 
297 |                 # Create async output bucket
298 |                 s3_async = s3.Bucket(self, f'{model["name"]}-S3Async')
299 | 
300 |                 s3_policy = iam.Policy(self, "sm-deploy-policy-s3",
301 |                                             statements=[iam.PolicyStatement(
302 |                                                 effect=iam.Effect.ALLOW,
303 |                                                 actions=["s3:*"],
304 |                                                 resources=[s3_async.bucket_arn]
305 |                                             )]
306 |                 )
307 | 
308 |                 role.attach_inline_policy(sns_policy)
309 |                 role.attach_inline_policy(s3_policy)
310 | 
311 |                 environment = {
312 |                                     "MODEL_CACHE_ROOT": "/opt/ml/model",
313 |                                     "SAGEMAKER_ENV": "1",
314 |                                     "SAGEMAKER_MODEL_SERVER_TIMEOUT": "3600",
315 |                                     "SAGEMAKER_CONTAINER_LOG_LEVEL": "20",
316 |                                     "SAGEMAKER_PROGRAM": "inference.py",
317 |                                     "SAGEMAKER_REGION": model_info["region_name"],
318 |                                     "SAGEMAKER_SUBMIT_DIRECTORY": "/opt/ml/model/code",
319 |                                 }
320 | 
321 |                 environment = merge_env(environment, model_env)
322 | 
323 |                 endpoint = SageMakerAsyncEndpointConstruct(self, "FoundationModelEndpoint",
324 |                                 project_prefix = configs["project_prefix"],
325 |                                 
326 |                                 role_arn= role.role_arn,
327 | 
328 |                                 model_name = model["name"],
329 |                                 model_bucket_name = model_info["model_bucket_name"],
330 |                                 model_bucket_key = model_info["model_bucket_key"],
331 |                                 model_docker_image = model_info["model_docker_image"],
332 | 
333 |                                 variant_name = "AllTraffic",
334 |                                 variant_weight = 1,
335 |                                 instance_count = model.get("inference_instance_count", 1),
336 |                                 instance_type = model_info["instance_type"],
337 | 
338 |                                 environment = environment,
339 |                                 deploy_enable = True,
340 | 
341 |                                 success_topic=success_topic.topic_arn,
342 |                                 error_topic=error_topic.topic_arn,
343 |                                 s3_async_bucket=s3_async.bucket_name,
344 |                                 model_package_arn=model_package_arn,
345 |                                 enable_network_isolation=is_network_isolation_enabled
346 |                 )
347 |                 endpoint.node.add_dependency(role)
348 |                 endpoint.node.add_dependency(sts_policy)
349 |                 endpoint.node.add_dependency(logs_policy)
350 |                 endpoint.node.add_dependency(ecr_policy)
351 | 
352 |         if len(step_function_enabled_endpoints) > 0:
353 |             stepfunction_stack = StepFunctionStack(self, "StepFunctionStack",
354 |                                             api_stack = api_stack,
355 |                                             step_function_enabled_endpoints=step_function_enabled_endpoints)


--------------------------------------------------------------------------------
/stack/lambda_stack.py:
--------------------------------------------------------------------------------
 1 | from aws_cdk import (
 2 |     Duration,
 3 |     Stack,
 4 |     aws_lambda as _lambda,
 5 |     aws_iam as iam
 6 | )
 7 | 
 8 | from constructs import Construct
 9 | 
10 | class LambdaStack(Stack):
11 | 
12 |     def __init__(self, scope: Construct, construct_id: str, resource_name, asset_dir, endpoint_name, **kwargs) -> None:
13 |         super().__init__(scope, construct_id, **kwargs)
14 | 
15 |         # Create lambda
16 |         self.app_handler = _lambda.Function(self, f"{resource_name}Handler",
17 |                 runtime=_lambda.Runtime.PYTHON_3_9,
18 |                 code=_lambda.Code.from_asset(asset_dir),
19 |                 handler="app.handler",
20 |                 timeout=Duration.seconds(180),
21 |                 environment={
22 |                     "ENDPOINT_NAME": endpoint_name,
23 |                 })
24 |         
25 |         # 
26 |         endpoint_arn = f'arn:aws:sagemaker:{self.region}:{self.account}:endpoint/{endpoint_name.lower()}'
27 |     
28 |         self.app_handler.add_to_role_policy(iam.PolicyStatement(
29 |             effect=iam.Effect.ALLOW,
30 |             actions=["sagemaker:InvokeEndpoint"],
31 |             resources=[endpoint_arn],
32 |         ))
33 | 
34 |         self.resource_name = resource_name
35 | 
36 | 


--------------------------------------------------------------------------------
/stack/sagemaker_endpoint_manager_stack.py:
--------------------------------------------------------------------------------
 1 | from aws_cdk import (
 2 |     Stack,
 3 |     CfnOutput
 4 | )
 5 | 
 6 | from constructs import Construct
 7 | 
 8 | from stack.api_stack import APIStack
 9 | from stack.foundation_model_stack import FoundationModelStack
10 | from stack.endpoint_manager_stack import EndpointManagerStack
11 | from stack.stepfunction_stack import StepFunctionStack
12 | 
13 | class SagemakerEndpointManagerStack(Stack):
14 |     def __init__(self, scope: Construct, construct_id: str, configs, **kwargs) -> None:
15 |         super().__init__(scope, construct_id, **kwargs)
16 | 
17 |         # Deploy api stack
18 |         api_stack = APIStack(self, "APIStack", 
19 |                     configs=configs
20 |         )
21 | 
22 |         # Deploy endpoint manager stack
23 |         endpoint_manager_stack = EndpointManagerStack(self, "EndpointManagerStack", 
24 |                     api_stack = api_stack
25 |         )
26 | 
27 |         # Deploy model stack
28 |         fm_stack = FoundationModelStack(self, "ModelStack", 
29 |                                 configs=configs,
30 |                                 api_stack=api_stack
31 |         )
32 | 
33 |         CfnOutput(self, "APIURL",
34 |             value=f"https://{api_stack.api.rest_api_id}.execute-api.{self.region}.amazonaws.com/prod"
35 |         )


--------------------------------------------------------------------------------
/stack/stepfunction_stack.py:
--------------------------------------------------------------------------------
  1 | from aws_cdk import (
  2 |     NestedStack,
  3 |     aws_apigateway as apigateway,
  4 |     aws_stepfunctions as sfn,
  5 |     aws_iam as iam,
  6 | )
  7 | 
  8 | from constructs import Construct
  9 | 
 10 | class StepFunctionStack(NestedStack):
 11 |     def __init__(self, scope: Construct, construct_id: str, api_stack, step_function_enabled_endpoints, **kwargs) -> None:
 12 |         super().__init__(scope, construct_id, **kwargs)
 13 | 
 14 |         # Create a step function execution role
 15 |         role = iam.Role(self, "StepfunctionExecutionRole", assumed_by=iam.ServicePrincipal("states.amazonaws.com"))
 16 | 
 17 |         # Add permission to role to invoke Amazon SageMaker real time endpoint
 18 |         role.add_to_policy(
 19 |             iam.PolicyStatement(
 20 |                 effect=iam.Effect.ALLOW,
 21 |                 actions=["sagemaker:InvokeEndpoint"],
 22 |                 resources=step_function_enabled_endpoints
 23 |             )
 24 |         )
 25 | 
 26 |         # Create an Amazon SageMaker invoke state machine
 27 |         sagemaker_invoke_fnc = sfn.StateMachine(self, "SageMakerInvokeStepfunction",
 28 |             definition_body=sfn.DefinitionBody.from_file("config/sagemaker-invoke.json"),
 29 |             role=role
 30 |         )
 31 |        
 32 |         # Add permission to invoke step function
 33 |         api_stack.api_gateway_role.add_to_policy(
 34 |             iam.PolicyStatement(
 35 |             effect=iam.Effect.ALLOW,
 36 |             actions=["states:StartExecution"],
 37 |             resources=[sagemaker_invoke_fnc.state_machine_arn]
 38 |             )
 39 |         )
 40 | 
 41 |         # Add permission to get describe step function execution
 42 |         stepfunction_execution_arn = f'arn:aws:states:{self.region}:{self.account}:execution:{sagemaker_invoke_fnc.state_machine_name}:*'
 43 | 
 44 |         api_stack.api_gateway_role.add_to_policy(
 45 |             iam.PolicyStatement(
 46 |                 effect=iam.Effect.ALLOW,
 47 |                 actions=["states:DescribeExecution"],
 48 |                 resources=[stepfunction_execution_arn]
 49 |             )
 50 |         )
 51 | 
 52 |         # Add api integration with step function
 53 |         start_execution_resource = api_stack.api.root.add_resource("startexecution")
 54 | 
 55 |         # Add apigateway integration with step function using with sagemaker_invoke_fnc state machine arn
 56 |         start_execution_integration = apigateway.AwsIntegration(
 57 |             service="states",
 58 |             integration_http_method="POST",
 59 |             action="StartExecution",
 60 |             options=apigateway.IntegrationOptions(
 61 |                 credentials_role=api_stack.api_gateway_role,
 62 |                 integration_responses=[
 63 |                                         apigateway.IntegrationResponse(
 64 |                                             status_code="200",
 65 |                                         )
 66 |                                     ],
 67 |                 request_templates={"application/json": "{ \"stateMachineArn\": \""+sagemaker_invoke_fnc.state_machine_arn+"\", \"input\": \"$util.escapeJavaScript($input.json('$'))\" }" }
 68 |             )
 69 |         )
 70 | 
 71 |         start_execution_resource.add_method("POST",
 72 |                                             start_execution_integration,
 73 |                                             authorizer=api_stack.api_authorizer,
 74 |                                             method_responses=[
 75 |                                                                 apigateway.MethodResponse(
 76 |                                                                     status_code="200",
 77 |                                                                     response_models={
 78 |                                                                         "application/json": apigateway.Model.EMPTY_MODEL
 79 |                                                                     },
 80 |                                                                 ),
 81 |                                                                 apigateway.MethodResponse(
 82 |                                                                     status_code="400",
 83 |                                                                     response_models={
 84 |                                                                         "application/json": apigateway.Model.ERROR_MODEL
 85 |                                                                     },
 86 |                                                                 ),
 87 |                                                                 apigateway.MethodResponse(
 88 |                                                                     status_code="500",
 89 |                                                                     response_models={
 90 |                                                                         "application/json": apigateway.Model.ERROR_MODEL
 91 |                                                                     },
 92 |                                                                 ),
 93 |                                                             ]
 94 |         )
 95 | 
 96 |         # Add api integration with step function to get step function execution status
 97 |         describe_execution_resource = api_stack.api.root.add_resource("describeexecution")
 98 | 
 99 |         # Add apigateway integration with step function to get step function execution status
100 |         describe_execution_integration = apigateway.AwsIntegration(
101 |             service="states",
102 |             integration_http_method="POST",
103 |             action="DescribeExecution",
104 |             options=apigateway.IntegrationOptions(
105 |                 credentials_role=api_stack.api_gateway_role,
106 |                 integration_responses=[
107 |                                         apigateway.IntegrationResponse(
108 |                                             status_code="200",
109 |                                         )
110 |                 ]
111 |             )
112 |         )
113 | 
114 |         describe_execution_resource.add_method("POST",
115 |                                             describe_execution_integration,
116 |                                             authorizer=api_stack.api_authorizer,
117 |                                             method_responses=[
118 |                                                                 apigateway.MethodResponse(
119 |                                                                     status_code="200",
120 |                                                                     response_models={
121 |                                                                         "application/json": apigateway.Model.EMPTY_MODEL
122 |                                                                     },
123 |                                                                 ),
124 |                                                                 apigateway.MethodResponse(
125 |                                                                     status_code="400",
126 |                                                                     response_models={
127 |                                                                         "application/json": apigateway.Model.ERROR_MODEL
128 |                                                                     },
129 |                                                                 ),
130 |                                                                 apigateway.MethodResponse(
131 |                                                                     status_code="500",
132 |                                                                     response_models={
133 |                                                                         "application/json": apigateway.Model.ERROR_MODEL
134 |                                                                     },
135 |                                                                 ),
136 |                                                             ]
137 |         )
138 | 
139 |             
140 | 
141 | 
142 | 
143 | 
144 | 


--------------------------------------------------------------------------------
/stack/util.py:
--------------------------------------------------------------------------------
 1 | def merge_env(env, extra_env_vars):
 2 |     """Sets env based on default or override, returns envs."""
 3 |     if env is None:
 4 |         env = {}
 5 | 
 6 |     for key, value in extra_env_vars.items():
 7 | 
 8 |         if key not in env:
 9 |             env[key] = value
10 | 
11 |     if env == {}:
12 |         env = None
13 | 
14 |     return env


--------------------------------------------------------------------------------
/tests/__init__.py:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/dalacan/sagemaker-endpoint-manager/eaefa6c702bf7c1633bcb69e41862bba04de0f44/tests/__init__.py


--------------------------------------------------------------------------------
/tests/unit/__init__.py:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/dalacan/sagemaker-endpoint-manager/eaefa6c702bf7c1633bcb69e41862bba04de0f44/tests/unit/__init__.py


--------------------------------------------------------------------------------
/tests/unit/test_sagemaker_jumpstart_generative_ai_app_stack.py:
--------------------------------------------------------------------------------
 1 | import aws_cdk as core
 2 | import aws_cdk.assertions as assertions
 3 | 
 4 | from sagemaker_jumpstart_generative_ai_app.sagemaker_jumpstart_generative_ai_app_stack import SagemakerJumpstartGenerativeAiAppStack
 5 | 
 6 | # example tests. To run these tests, uncomment this file along with the example
 7 | # resource in sagemaker_jumpstart_generative_ai_app/sagemaker_jumpstart_generative_ai_app_stack.py
 8 | def test_sqs_queue_created():
 9 |     app = core.App()
10 |     stack = SagemakerJumpstartGenerativeAiAppStack(app, "sagemaker-jumpstart-generative-ai-app")
11 |     template = assertions.Template.from_stack(stack)
12 | 
13 | #     template.has_resource_properties("AWS::SQS::Queue", {
14 | #         "VisibilityTimeout": 300
15 | #     })
16 | 


--------------------------------------------------------------------------------
/utils/__init__.py:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/dalacan/sagemaker-endpoint-manager/eaefa6c702bf7c1633bcb69e41862bba04de0f44/utils/__init__.py


--------------------------------------------------------------------------------
/utils/sagemaker_helper.py:
--------------------------------------------------------------------------------
  1 | import sagemaker
  2 | from sagemaker import script_uris
  3 | from sagemaker import image_uris 
  4 | from sagemaker import model_uris
  5 | from sagemaker.jumpstart.notebook_utils import list_jumpstart_models
  6 | from sagemaker import environment_variables
  7 | from sagemaker.jumpstart.utils import verify_model_region_and_return_specs
  8 | from sagemaker.jumpstart.enums import JumpStartScriptScope
  9 | from sagemaker.jumpstart.constants import (
 10 |     JUMPSTART_DEFAULT_REGION_NAME,
 11 | )
 12 | import boto3
 13 | from typing import Optional
 14 | 
 15 | # session = sagemaker.Session()
 16 | 
 17 | def sagemaker_env(model_id, region, model_version="*"):
 18 |     extra_env_vars = environment_variables.retrieve_default(
 19 |         model_id=model_id,
 20 |         model_version=model_version,
 21 |         region=region,
 22 |         include_aws_sdk_env_vars=False
 23 |     )
 24 | 
 25 |     return extra_env_vars
 26 | 
 27 | def get_sagemaker_uris(model_id,instance_type,region_name):
 28 |     
 29 |     
 30 |     MODEL_VERSION = "*"  # latest
 31 |     SCOPE = "inference"
 32 | 
 33 |     inference_image_uri = image_uris.retrieve(region=region_name, 
 34 |                                           framework=None,
 35 |                                           model_id=model_id, 
 36 |                                           model_version=MODEL_VERSION, 
 37 |                                           image_scope=SCOPE, 
 38 |                                           instance_type=instance_type)
 39 |     
 40 |     inference_model_uri = model_uris.retrieve(
 41 |                                           region=region_name,
 42 |                                           model_id=model_id, 
 43 |                                           model_version=MODEL_VERSION, 
 44 |                                           model_scope=SCOPE)
 45 |     
 46 |     inference_source_uri = script_uris.retrieve(
 47 |                                             region=region_name,
 48 |                                             model_id=model_id, 
 49 |                                             model_version=MODEL_VERSION, 
 50 |                                             script_scope=SCOPE)
 51 | 
 52 |     model_bucket_name = inference_model_uri.split("/")[2]
 53 |     model_bucket_key = "/".join(inference_model_uri.split("/")[3:])
 54 |     model_docker_image = inference_image_uri
 55 | 
 56 |     return {"model_bucket_name":model_bucket_name, "model_bucket_key": model_bucket_key, \
 57 |             "model_docker_image":model_docker_image, "instance_type":instance_type, \
 58 |                 "inference_source_uri":inference_source_uri, "region_name":region_name}
 59 | 
 60 | def get_model_spec(
 61 |     model_id: str,
 62 |     model_version: str,
 63 |     region: Optional[str],
 64 |     scope: Optional[str] = None,
 65 | ):
 66 |     if region is None:
 67 |         region = JUMPSTART_DEFAULT_REGION_NAME
 68 | 
 69 |     # Default to inference scope
 70 |     if scope is None:
 71 |         scope = JumpStartScriptScope.INFERENCE
 72 | 
 73 |     model_specs = verify_model_region_and_return_specs(
 74 |             model_id=model_id,
 75 |             version=model_version,
 76 |             scope=scope,
 77 |             region=region
 78 |         )
 79 | 
 80 |     if model_specs is None:
 81 |         return None
 82 |     
 83 |     return model_specs
 84 | 
 85 | def get_model_package_arn(
 86 |     model_specs,
 87 |     region: Optional[str]):
 88 |     if region is None:
 89 |         region = JUMPSTART_DEFAULT_REGION_NAME
 90 | 
 91 |     if model_specs.hosting_model_package_arns is None:
 92 |         return None
 93 | 
 94 |     # Return regional arn
 95 |     regional_arn = model_specs.hosting_model_package_arns.get(region)
 96 | 
 97 |     return regional_arn
 98 | 
 99 | def enable_network_isolation(model_specs):
100 |     if model_specs.inference_enable_network_isolation is None:
101 |         return False
102 | 
103 |     # Return regional arn
104 |     inference_enable_network_isolation = model_specs.inference_enable_network_isolation
105 | 
106 |     return inference_enable_network_isolation
107 |     


--------------------------------------------------------------------------------