This is the Platform API for interacting with the AI4EOSC services, built using FastAPI. It aims at providing a stable UI, effectively decoupling the services offered by the project from the underlying tools we use to provide them (ie. Nomad, OSCAR).
The API is currently deployed here:
- production API (
master
branch) - development API (
dev
branch)
Images of both API are accessible in the project's Harbor registry:
registry.services.ai4os.eu/ai4os/ai4-papi:prod
registry.services.ai4os.eu/ai4os/ai4-papi:dev
The Dashboards pointing to those APIs are respectively:
Requirements To use this library you need to have Nomad installed to be able to interact with deployments. Once you have Nomad installed you have to export the following variables with the proper local paths and http address:
export NOMAD_ADDR=https://some-public-ip:4646
export NOMAD_CACERT=/path/to/tls/nomad-ca.pem
export NOMAD_CLIENT_CERT=/path/to/tls/nomad-cli.pem
export NOMAD_CLIENT_KEY=/path/to/tls/nomad-cli-key.pem
For this you will need to ask the administrator of the cluster for the proper certificates.
Once you are done you can proceed to install the module:
pip install git+https://github.com/ai4eosc/ai4-papi.git
If you plan to use the module to develop, install instead in editable mode:
git clone https://github.com/ai4eosc/ai4-papi
cd ai4-papi
pip install -e .
To deploy the API, the are several options:
-
Using entrypoints:
ai4papi-run --host 0.0.0.0 --port 8080
-
Using uvicorn directly (with the auto
reload
feature enabled if you are developing):uvicorn ai4papi.main:app --reload
-
Using our Makefile
make run
-
From Dockerhub
docker run -v /local-path-to/nomad-certs:/home/nomad-certs -p 8080:80 registry.services.ai4os.eu/ai4os/ai4-papi:prod
-
Building from our Dockerfile.
docker build -t ai4-papi:prod --build-arg papi_branch=master . docker run -v /local-path-to/nomad-certs:/home/nomad-certs -p 8080:80 ai4-papi:prod
Once the API is running, go to http://127.0.0.1:8080/docs to check the API methods in the Swagger UI.
Some of the API methods are authenticated (🔒) via OIDC tokens, so you will need to perform the following steps to access those methods.
- Create an EGI Check-In account.
- Enroll (
People > Enroll
) in one of the approved Virtual Organizations:vo.ai4eosc.eu
vo.imagine-ai.eu
You will have to wait until an administrator approves your request before proceeding with the next steps. Supported OIDC providers and Virtual Organizations are described in the configuration.
There are two ways of generating a valid refresh user token to access the methods: either via an UI or via the terminal.
If have a EGI Check-In account, you can generate a refresh user token with
EGI token: click Authorise
and sign-in with your account.
Then use the Access Token
to authenticate your calls.
-
Install the OIDC agent in your system.
-
Configure the OIDC agent:
eval `oidc-agent-service start` oidc-gen \ --issuer https://aai.egi.eu/auth/realms/egi \ --scope "openid profile offline_access eduperson_entitlement email" \ egi-checkin
It will open the browser so you can authenticate with your EGI account. Then go back to the terminal and finish by setting and encryption password.
-
Add the following line to your
.bashrc
to start the agent automatically at startup (ref):eval `oidc-agent-service use` > /dev/null
-
Generate the OIDC token
oidc-token egi-checkin
-
Optional
: You can check you have set everything up correctly by running:flaat-userinfo --oidc-agent-account egi-checkin
This should print you EGI user information.
To make authenticated calls, you have several options:
-
Using CURL calls:
curl --location 'http://localhost:8080' --header 'Authorization: Bearer <your-OIDC-token>'
-
From in the Swagger UI (http://localhost:8080/docs), click in the upper right corner button
Authorize
🔓 and input your token. From now on you will be authenticated when making API calls from the Swagger UI. -
From inside a Python script
from types import SimpleNamespace from ai4papi.routers.v1 import deployments deployments.get_deployments( vos=['vo.ai4eosc.eu'], authorization=SimpleNamespace( credentials='your-OIDC-token' ), )
Here follows an overall summary of the available routes. The 🔒 symbol indicates the method needs authentication to be accessed. More details can be found in the API docs.
-
/v1/catalog/
: interact with the metadata of the modules/tools in the marketplace.Notes: The catalog caches results for up to 6 hours to improve UX (see doctring).
-
/v1/deployments/
: (🔒) deploy modules/tools in the platform to perform trainings -
/v1/stats/deployments/
: (🔒) retrieve usage stats for users and overall platform.Requirements
For this you need to declare a ENV variable with the path of the Nomad cluster logs repo:
export ACCOUNTING_PTH="/your/custom/path/ai4-accounting"
It will serve the contents of the
ai4-accounting/summaries
folder.
The API methods can also be accessed by interacting directly with the Python package.
from types import SimpleNamespace
from ai4papi.routers.v1 import deployments
# Get all the user's deployments
deployments.modules.get_deployments(
vos=['vo.ai4eosc.eu'],
authorization=SimpleNamespace(
credentials='your-OIDC-token'
),
)
#
# [{'job_ID': 'example',
# 'status': 'running',
# 'owner': '4545898984949741@someprovider',
# 'submit_time': '2023-01-13 11:36:16',
# 'alloc_ID': 'e6b24722-e332-185a-a9b6-817ce8d26f48',
# 'resources': {
# 'cpu_num': 2,
# 'gpu_num': 0,
# 'memoryMB': 8000,
# 'diskMB': 300
# },
# 'endpoints': {
# 'deepaas': 'https://deepaas.xxx.xxx.xxx.xxx',
# 'monitor': 'https://monitor.xxx.xxx.xxx.xxx',
# 'ide': 'https://ide.xxx.xxx.xxx'
# }
# }]
These are the configuration files the API uses:
etc/main_conf.yaml
: main configuration file of the APIetc/modules
: configuration files for standard modulesetc/tools
: configuration files for toolsdeep-oc-federated-server
: federated server
The pattern for the subfolders follows:
user.yaml
: user customizable configuration to make a deployment in Nomad. Also contains the generic quotas for hardware (seerange
parameter).nomad.hcl
: additional non-customizable values (eg. ports)