Create Deployments

Once you have fully tested your BentoML Service locally, you can push it to BentoCloud for production deployment. This document explains how to create a Deployment on BentoCloud.

Prerequisites

  • Make sure you have logged in to BentoCloud using an API token with Developer Operations Access.

  • You have created a BentoML project that contains at least a service.py file and a bentofile.yaml file (or you have an available Bento either locally or on BentoCloud). You can use this Quickstart or any project in Examples.

Deploy a new project to BentoCloud

You can deploy a new project through either the command line interface (CLI) or Python API.

In your project directory where the bentofile.yaml file is stored, run the following command and use the -n flag to optionally set a name.

bentoml deploy . -n <deployment_name>

You can see the Deployment status in your terminal after running the command.

Specify the path to your BentoML project using the bento parameter and optionally set a name.

import bentoml

bentoml.deployment.create(bento = "./path_to_your_project", name = "my_deployment_name")

You can use the block function wait_until_ready to periodically check the status of a Deployment until it becomes ready or until a specified timeout is reached.

import bentoml

dep = bentoml.deployment.create(bento="./path_to_your_project", name = "my_deployment_name")

dep.wait_until_ready(timeout=3600)

BentoML does the following automatically during deployment:

  1. Build: Build your project into a Bento based on bentofile.yaml.

  2. Push: Push the Bento to BentoCloud.

  3. Deploy: Deploy the Bento on BentoCloud by performing the following steps in order:

    1. Containerize the Bento as an OCI-compliant image.

    2. Provision instances on BentoCloud.

    3. Start the BentoML Service on the instances based on the specified configuration.

Note

You DO NOT need to perform the above three steps (Build, Push, and Deploy) manually, which is a long-running automated process.

Deploy an existing Bento to BentoCloud

If you already have a Bento built locally (run bentoml list to view all the local Bentos), you can deploy it using either the BentoML CLI or Python API.

In your project directory where the bentofile.yaml file is stored, run the following command and use the -n flag to optionally set a name.

bentoml deploy bento_name:version -n <deployment_name>

Specify the Bento tag using the bento parameter and optionally set a name.

import bentoml

bentoml.deployment.create(bento = "bento_name:version", name = "my_deployment_name")

You can use the block function wait_until_ready to periodically check the status of a Deployment until it becomes ready or until a specified timeout is reached.

import bentoml

dep = bentoml.deployment.create(bento = "bento_name:version", name = "my_deployment_name")

dep.wait_until_ready(timeout=3600)

The bentoml deploy command and the bentoml.deployment.create function automatically push and deploy the Bento to BentoCloud. If you only need to share a Bento with your team and deploy it later, you can push the Bento to BentoCloud by running the following command:

$ bentoml push <bento_name:version>

╭──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╮
│ Successfully pushed Bento "bento_name:version"                                                                                                                                   │
╰──────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────╯
Pushing Bento "bento_name:version" ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100.0% • 51.2/51.2 kB • ? • 0:00:00

You can then view your pushed Bento on the BentoCloud console, which provides a web-based, graphical user interface (UI), and create a Deployment using the Bento.