Skip to main content

Data - Compliance between AI Tools and S3™ * compatible Object Storage

Stephanie R.
Updated

Learn how to use S3™*-compatible buckets with OVHcloud AI Solutions via the OVHcloud Control Panel and ovhai CLI.

This will help you to manage your data when working with AI Notebooks, AI Training, and AI Deploy.

 

Topics

 

Requirements

 

Instructions

Create an S3™-compatible bucket

You can create your first S3™-compatible Object Storage bucket in the OVHcloud Control Panel by navigating to the Object Storage section via the Public Cloud menu.

On the next screen, you can see all the buckets you have already created on your Public Cloud project. You can create a new bucket by clicking the + Create an object container button, just above the bucket list.

objectStorage_serviceList.png

If you haven't created a bucket yet, the interface will be slightly different, prompting you to + Create an object container.

Once you have clicked this button, you will be asked to specify the type of bucket you want to create. In this guide, we will choose the S3™-compatible API:

objectStorage_order_solution.png

Then, specify the deployment mode you want to use:

objectStorage_order_deploymentMode.png

Select a region:

objectStorage_order_region.png

Enable or disable object versioning:

objectStorage_order_versioning.png

Once you have selected the deployment mode and region for your S3™-compatible bucket, the following steps will vary based on the deployment mode you have chosen:

For Region deployment mode:

You will need to link a user from your Public Cloud project to your new bucket.

To perform this step, you have two options. Firstly, you can choose from your existing users in your Public Cloud project. Alternatively, you can create a new user, which will be automatically linked to the bucket.

objectStorage_order_linkUser.png

NOTE: If you create a new user, make sure to save its credentials (username, access key, secret key). Of course, you can get them back later if you lose them.

For Local Zone deployment mode:

No additional steps are required to link a user to the bucket, as all users of your Public Cloud project will automatically have access to all containers in Local Zones.

Enable or disable data encryption:

objectStorage_order_encryption.png

Name your bucket. Note that the bucket name must be unique across all buckets in your Public Cloud account and must meet the following conditions:

  • Between 3 and 63 characters
  • Consist only of lowercase letters, numbers, dots (.), and hyphens (-)
  • Must start and end with lowercase alphanumeric characters (a to z and 0 to 9)

Once you have named your bucket and configured its settings, click the Create the container button to confirm S3™-compatible bucket creation.

objectStorage_order_containerName.png

You should then find it in the list of your existing Object Storage buckets.

By clicking on its name, you can see the objects it contains (empty for now), as well as information such as its endpoint and location (region), which we will use later:

objectStorage_containerDetails.png

 

Edit your S3™-compatible user roles

Once your S3™-compatible bucket has been created, you need to check that the associated user has the necessary rights to interact with your data and the OVHcloud AI Solutions (AI Notebooks, Training, and Deploy).

For Local Zone deployment, you did not need to associate a user with your bucket. However, it is important to check that you also have at least one user with the necessary rights.

To do that, navigate to the Users & Roles area in the Project Management section of the OVHcloud Control Panel.

Whether you have created a new user or use an existing one, check that this user has at least the following rights: AI Training Operator and ObjectStore Operator, as shown below:

manager_usersAndRoles_addUser_roles.png

For more information about editing user rights, you can check the dedicated documentation.

 

Retrieve user credentials

Before continuing, you need to ensure that you have the following information so that you can authenticate via the CLI and use your bucket:

  • Username
  • Password
  • S3™-compatible access key
  • S3™-compatible secret key

As we saw in the previous step, all your Public Cloud project users are displayed in the Users & Roles menu in the Project Management section. You will find there your existing usernames. The username we are interested in is the one to which you gave permissions in the previous step.

If you have lost the password associated with this username, you can regenerate it by clicking the more options ... button, and then on Generate a password, which will revoke the old one.

manager_usersAndRoles_moreOptions.png

Then, you need to retrieve the access key and secret key associated to this user. To do that, return to the Object Storage section. Click on the Object Storage Users tab. Here you will find the Access key for each of your existing users. You can also view the secret key by clicking the more options ... button, then on View the secret key.

objectStorage_users_moreOptions.png

 

Set up proper access permissions

It is also important to ensure that this user has the appropriate permissions to access all the files in the bucket. Otherwise, you may encounter permission issues during the AI Solution deployment process.

To grant access to your bucket for your user, click the more options ... button located next to your bucket name. Then, click the Add a user to a container option and select the same user you used during bucket creation.

objectStorage_serviceList_moreOptions.png

By following these steps, you can grant the necessary permissions to your user, ensuring a smooth deployment process for your notebook, job, or app.

 

Add a datastore via the ovhai CLI

Now that your bucket has been created, and the associated user has the necessary permissions, you can move on to the ovhai CLI. If you have not installed it yet, please refer to the ovhai CLI installation guide.

The first step is to authenticate with the user that you have granted access to your bucket. To do this, use the following command:

ovhai login

We recommend that you choose the Terminal authentication mode, and enter your username and associated password to log in.

If you encounter an authentication error, please recheck this step and make sure you have the AI right.

Once authenticated, you can list your existing datastores with the following command:

ovhai datastore list

You should obtain the following result:

ALIAS  STORE_TYPE OWNER    ENDPOINT
US-EAST-VA    swift      ovhcloud ~
US-WEST-OR    swift      ovhcloud ~

Here you can see that there are no S3™-compatible data stores in the list yet. That is why you will add your own, to then access the buckets of a given region, and for a specific access key and secret key combination!

To add this datastore, you will use the ovhai datastore add s3 command pattern. You can use the --help flag to know more about adding S3™-compatible datastores:

ovhai datastore add s3 --help

This will give us the different arguments to fill in:

Add an S3 compatible data store

Usage: ovhai datastore add s3 [OPTIONS] <ALIAS> <ENDPOINT_URL> <REGION> <ACCESS_KEY> <SECRET_KEY>

Arguments:
  <ALIAS>         Alias for the data store
  <ENDPOINT_URL>  Data store connection URL
  <REGION>        Data store region
  <ACCESS_KEY>    Connection access key
  <SECRET_KEY>    Connection secret key

Options:
      --store-credentials-locally  Whether or not to store the data store credentials locally when creating or updating a data store
      --token <TOKEN>              Authentication using Token rather than OAuth
      --no-color                   Remove colors from output
  -h, --help                       Print help

Here is the command pattern to add an OVHcloud S3™-compatible datastore:

ovhai datastore add s3 <ALIAS> <endpoint_url> <region> <my-access-key> <my-secret-key> --store-credentials-locally

You will need to replace: - <alias> by the name you want to give for your datastore - <endpoint_url> by the Endpoint value you have retrieved at the end of bucket creation step. This endpoint should start with: https://s3.... - <region> by the region where you have created your bucket. This region is indicated as the Location of your bucket, on the last screenshot of the bucket creation step. Make sure you use lowercase letters. - <my-access-key> by the access key of the associated user - my-secret-key by the secret key of the associated user

This endpoint also allows you to choose between Standard and High Performance storage classes. The storage class you select will determine the storage class used during file upload to the bucket when using the ovhai CLI. If you plan to upload files using the Control Panel (UI), you will have the option to choose between Standard and High Performance storage. However, when using the CLI, the default endpoint will use Standard storage. If you require High Performance storage, please ensure that the endpoint ends with .perf.cloud.ovh.us.

For more information on storage classes, please consult this documentation.

Here is an example to add a datastore named ai-training that will handle Region buckets created in Vint Hill (us-east-va) region (since they share the same endpoint_url):

ovhai datastore add s3 ai-training https://s3.us-east-va.io.cloud.ovh.us/ us-east-va <my-access-key> <my-secret-key> --store-credentials-locally

You can now check the datastore list:

ovhai datastore list

You should see your new datastore in the list:

ALIAS        STORE_TYPE OWNER    ENDPOINT
US-EAST-VA   swift      ovhcloud ~
US-WEST-OR   swift      ovhcloud ~
ai-training  s3         customer https://s3.us-east-va.io.cloud.ovh.us/

Now that you have a datastore dedicated to the specified region, you can list the buckets associated to this region and access and secret keys combination by running the following command (replace by your datastore name, s3-ai-training in previous example):

ovhai bucket list <alias> 
DATE                     NAME
2024-12-25T09:00:00.000Z s3-ai-training

You can see that the bucket s3-ai-training you created using the Control Panel (UI) has been retrieved.

 

Create new buckets linked to this datastore using CLI (Optional)

This step is optional and for information only.

Now that your datastore is configured, you can simply create new buckets for the region and user associated with your datastore using the following command, rather than using the OVHcloud Control Panel (UI):

ovhai bucket create <alias> <new_bucket-name>

Keep in mind that the bucket name must be between 3 and 63 characters, can consist only of lowercase letters, numbers, dots (.), and hyphens (-), and must start and end with lowercase alphanumeric characters (a to z and 0 to 9).

Then you can check that your S3™-compatible bucket has been created correctly, and linked to your datastore (replace by your datastore name, 1azgra in previous example):

ovhai bucket list <alias>

You should see your bucket in the list:

DATE                     NAME
2024-12-25T09:00:00.000Z s3-ai-training
2024-12-25T10:00:00.000Z new-bucket-name

 

Upload files to your bucket

Before connecting your bucket to the AI solutions, you are going to upload a few objects to it (images, codes, models, etc.) so that you can then retrieve these files from the AI Solutions. You can do this using the Control Panel (UI) or the ovhai CLI:

Go to the Object Storage section (in the Storage category) and click on your bucket name. Then, click the + Add objects button:

objectStorage_containerDetails.png

Select the files you want to upload to your bucket. If needed, you can prefix their names. Then, choose between Standard or High performance storage (find more information about storage classes here). Once you have made your selection, confirm by clicking on Import.

manager_aiNotebooks_addS3Object.png

In this example, the python file train-first-model.py is added and prefixed by /scripts, resulting in the file path /scripts/train-first-model.py.

Once the upload is completed, you should see uploaded files in the object list.

manager_aiNotebooks_S3_containerWithObject.png

Here is the basic command pattern to upload an object to your bucket:

ovhai bucket object upload <bucket-name>@<datastore-alias> <file_path_to_upload>

For example, you can use the following command to upload a file named file.py to a bucket named ai-training-container, connected to the s3-ai-training datastore. You will prefix the filename with /scripts, resulting in the file path /scripts/file.py:

ovhai bucket object upload ai-training-container@s3-ai-training file.py --add-prefix /scripts/

NOTE: When using this command, the current directory will be included in the resulting file path. It is recommended to change to the desired directory using the cd command before uploading objects, to avoid any unexpected file paths. We recommend using the --dry-run flag to preview the actions that will be taken during the upload process.

Once your file has been uploaded, you can list the content of your bucket:

ovhai bucket object list <bucket-name>@<datastore-alias>

Where you should see your uploaded object (file.py in this example):

DATE                     BYTES    NAME               DESCRIPTION
2024-12-26T16:00:00.000Z 10.4 KiB /scripts/file.py   STANDARD

To see the full list of options, you can also use the --help flag:

ovhai bucket object upload --help

 

Using S3™-compatible buckets with AI Tools

Now that your S3™-compatible bucket has been created and populated with some objects, you can mount it on any OVHcloud AI Solution to access your data.

Please note that for the moment this S3™-compatible feature can be used through ovhai CLI only. It will soon be available from the OVHcloud Control Panel.

To launch an AI notebook with an S3™-compatible bucket mounted as remote data, you can use the following command pattern:

ovhai notebook run <framework-id> <editor-id> \
   --volume <bucket-name>@<datastore-alias>:<mount_path>:<permission>

Just make sure to replace the framework and editor by those of your choice. Also, specify the right <bucket-name> and <datastore-alias, and designate the destination where you want the bucket data to be stored (<mount_path>) and its access permission (rw for read & write, or ro for read only).

For example, to launch a basic Jupyter notebook on 2 CPUs, with a Conda pre-packaged environment, and the bucket ai-training-container from the s3-ai-training datastore mounted as remote data on /workspace/my-codes with read and write (RW) permissions, use the following command:

ovhai notebook run conda jupyterlab \
   --volume ai-training-container@s3-ai-training/:/workspace/my-codes:rw \
   --cpu 2

This command will create an AI notebook with the specified framework and editor, and will mount the specified bucket at the specified destination with the specified permissions. After logging in to your notebook interface (using the same credentials you have used to ovhai login), you should see the uploaded file in your notebook's workspace:

PaaS_AI_general_dataS3compliance15.png

For more information on the capabilities of the ovhai notebook run command, you can run the following command:

ovhai notebook run --help

You can also consult the dedicated CLI - Launch an AI notebook documentation to know more about the possible options, frameworks, and editors available.

To launch an AI Training job with an S3™-compatible bucket mounted as remote data, you can use the following command pattern:

ovhai job run <docker-image> \
   --volume <bucket-name>@<datastore-alias>:<mount_path>:<permission>

Just make sure to replace the <docker-image> by yours. Also, specify the right <bucket-name> and <datastore-alias, and designate the destination where you want the bucket data to be stored (<mount_path>) and its access permission (rw for read & write, or ro for read only).

For example, to launch a basic job on 1 L40S GPU and the bucket ai-training-container from the s3-ai-training datastore mounted as remote data on /workspace/my-codes with read and write (RW) permissions, use the following command:

ovhai job run ovhcom/ai-training-pytorch:2.4.0 \
   --volume ai-training-container@s3-ai-training/:/workspace/my-codes:rw \
   --gpu 1 \
   --flavor l40s-1-gpu

This command will create an AI Training job with the specified Docker image, and will mount the specified bucket at the specified destination with the specified permissions.

For more information on the capabilities of the ovhai job run command, you can run the following command:

ovhai job run --help

You can also consult the dedicated CLI - Launch an AI Training job documentation to know more about the possible job commands.

To launch an AI Deploy app with an S3™-compatible bucket mounted as remote data, you can use the following command pattern:

ovhai app run <docker-image> \
   --volume <bucket-name>@<datastore-alias>:<mount_path>:<permission>

Just make sure to replace the <docker-image> by yours. Also, specify the right <bucket-name> and <datastore-alias, and designate the destination where you want the bucket data to be stored (<mount_path>) and its access permission (rw for read & write, or ro for read only).

For example, to launch a basic app on 2x A100 GPUs and the bucket ai-training-container from the s3-ai-training datastore mounted as remote data on /workspace/my-codes with read and write (RW) permissions, use the following command:

ovhai app run ovhcom/ai-training-pytorch:2.4.0 \
   --volume ai-training-container@s3-ai-training/:/workspace/my-codes:rw \
   --gpu 2 \
   --flavor a100-1-gpu

This command will create an AI Deploy app with the specified Docker image, and will mount the specified bucket at the specified destination with the specified permissions.

For more information on the capabilities of the ovhai app run command, you can run the following command:

ovhai app run --help

You can also consult the dedicated CLI - Launch an AI Deploy app documentation to know more about the possible job commands.

If you want to use the boto3 library to manage your S3™-compatible bucket objects, here is a notebook that contains basic commands.

 

Go further

For more information and tutorials, please see our other AI & Machine Learning support guides or explore the guides for other OVHcloud products and services.

If you need training or technical assistance to implement our solutions, contact your sales representative or click on this link to get a quote and ask our Professional Services experts for a custom analysis of your project.

*S3 is a trademark filed by Amazon Technologies, Inc. OVHcloud's service is not sponsored by, endorsed by, or otherwise affiliated with Amazon Technologies, Inc.

Related articles