{"serverless": "Serverless", "byoc": "BYOC"}

["serverless", "byoc"]

serverless

Union Serverless

The Union orchestrator empowers AI development teams to rapidly ship high-quality code to production by offering optimized performance, unparalleled resource efficiency, and a delightful workflow authoring experience.

• Run complex AI workloads with unparalleled performance, scale, and efficiency.

• Achieve millisecond-level execution times with reusable containers.

• Scale out to multiple regions, clusters, and clouds as needed for resource availability, scale or compliance.

Deployment models

Union is available in two deployment options: Serverless and BYOC (Bring Your Own Cloud).

• Union Serverless is a turnkey solution that takes care of all the infrastructure for you. All you need to do is sign up through your GitHub account and start running your workflows. To get started, follow the quick start guide below.

• Union BYOC lets you keep your data and workflow code on your infrastructure, but has Union manage it for you. It also offers more control over your hardware and other advanced features.

Quick start

In this section, we give a quick introduction to writing and running Union workflows.

Sign up for Union Serverless

First, sign up for Union Serverless:

Create an account

Once you’ve received confirmation that your sign up succeeded, navigate to the UI at serverless.union.ai. This is where you will be able to see your workflow executions and manage your projects:

Set up your Python environment

Set up a Python virtual environment with conda, venv or a similar tool.

Python 3.8 or higher is required. Python 3.11 is the current recommended version.

conda

Install conda using Miniconda, then run the following to create a new Python environment:

$ conda create -n union-env python=3.11
$ conda activate union-env

venv

Install Python 3.11 from your package manager or from Python.org, then run the following to create a virtual environment:

$ python -m venv .venv
$ source .venv/bin/activate

Install the union package

After setting up your virtual environment and activating it, install the union Python package:

$ pip install -U union

This will install:

• The union command-line tool

• The union SDK

• The flytekit SDK

Configure the union CLI

To register and run workflows on your Union instance using the union CLI, you will need to create a configuration file that contains your Union connection information. To do this, run the following command:

$ union create login --serverless

This will create the ~/.union/config.yaml with the configuration information to connect to Union Serverless.

Note

These directions apply to Union Serverless. To configure a connection to your Union instance in Union BYOC, see the BYOC version of this page.

By default, the union CLI will look for a configuration file at ~/.union/config.yaml. (See union CLI for more details.) You can override this behavior to specify a different configuration file by setting the UNION_CONFIG environment variable:

export UNION_CONFIG=~/.my-config-location/my-config.yaml

Alternatively, you can always specify the configuration file on the command line when invoking union by using the --config flag:

$ union --config ~/.my-config-location/my-config.yaml run my_script.py my_workflow

Warning

If you have previously used Union, you may have configuration files left over that will interfere with access to Union Serverless through the union CLI tool. Make sure to remove any files in ~/.unionai/ or ~/.union/ and unset the environment variables UNIONAI_CONFIG and UNION_CONFIG to avoid conflicts.

Create a “Hello, world!” workflow

To create an example workflow file, copy the following into a file called hello.py:

from flytekit import task, workflow

@task
def say_hello(name: str) -> str:
    return f"Hello, {name}!"

@workflow
def hello_world_wf(name: str = 'world') -> str:
    res = say_hello(name=name)
    return res

Tasks and workflows

The “Hello, world!” code contains a task and a workflow, which are Python functions decorated with the @task and @workflow decorators, respectively. For more information, see the task and workflow documentation.

Run the workflow locally in Python

You can run the workflow in your local Python environment with the union run command:

$ union run hello.py hello_world_wf

You should see the following output:

Running Execution on local.
Hello, world!

Since the @workflow function takes an argument called name, you can also pass that in as a command-line argument like this:

$ union run hello.py hello_world_wf --name Ada

You should see the following output:

Running Execution on local.
Hello, Ada!

Run the workflow remotely on Union

To run the workflow remotely on Union, add the --remote flag:

$ union run --remote hello.py hello_world_wf --name "Ada"

The output displays a URL that links to the workflow execution in the UI:

[✔] Go to https://serverless.union.ai/org/... to see execution in the UI.

Go to the UI to see the execution:

Note

When you use union create login --host <union-host-url> to configure the union CLI, this creates a config.yaml file configured for a Proof Key of Code Exchange (PKCE) mechanism. This is one of three authentication options, including DeviceFlow and ClientSecret. In short, PKCE opens a browser window allowing you to login, DeviceFlow returns a URL you can navigate to, and ClientSecret authenticates via a pre-configured secret. If you are using Union in a headless fashion, either on a VM, connecting to a machine via SSH, in CI/CD, etc., DeviceFlow and ClientSecret should be considered. See CLI authentication for more information.