Create a secure Vertex AI Workbench instance in a VPC network

Create and configure a standalone VPC

1. Create a VPC network named securevertex-vpc:

   gcloud compute networks create securevertex-vpc --subnet-mode=custom
   

2. Create a subnet named securevertex-subnet-a, with a primary IPv4 range of 10.10.10.0/29:

   gcloud compute networks subnets create securevertex-subnet-a --range=10.10.10.0/29 --network=securevertex-vpc --region=us-central1 --enable-private-ip-google-access
   

   You can supply a different value for the --range parameter. However, the minimum prefix length for a single notebook is 29. For more information, see IPv4 subnet ranges.

3. Create a regional Cloud Router named cloud-router-us-central1:

   gcloud compute routers create cloud-router-us-central1 --network securevertex-vpc --region us-central1
   

4. Create a regional Cloud NAT gateway named cloud-nat-us-central1:

   gcloud compute routers nats create cloud-nat-us-central1 --router=cloud-router-us-central1 --auto-allocate-nat-external-ips --nat-all-subnet-ip-ranges --region us-central1
   

Create a Cloud Storage bucket

In this section, you create a Cloud Storage bucket to hold a post-startup script that you can run when you create a new Vertex AI Workbench instance.

1. Create the Cloud Storage bucket:

   gcloud storage buckets create --location=us-central1 --uniform-bucket-level-access gs://BUCKET_NAME
   

   Replace BUCKET_NAME with a unique bucket name.

2. Set the BUCKET_NAME shell variable and verify that it was entered correctly:

   BUCKET_NAME=BUCKET_NAME
   echo $BUCKET_NAME
   

Create and upload a post-startup script

In this section, you create a post-startup script to clone a GitHub repository into a new Vertex AI Workbench instance.

1. To create the script, use a text editor such as vim or nano to create a poststartup.sh file. You need to prepend sudo in order to have permission to write to the file, for example:

   sudo vim poststartup.sh
   

2. Paste the following shell script into the file:

   #! /bin/bash
   echo "Current user: id" >> /tmp/notebook_config.log 2>&1
   echo "Changing dir to /home/jupyter" >> /tmp/notebook_config.log 2>&1
   cd /home/jupyter
   echo "Cloning generative-ai from github" >> /tmp/notebook_config.log 2>&1
   su - jupyter -c "git clone https://github.com/GoogleCloudPlatform/generative-ai.git" >> /tmp/notebook_config.log 2>&1
   echo "Current user: id" >> /tmp/notebook_config.log 2>&1
   echo "Installing python packages" >> /tmp/notebook_config.log 2&1
   su - jupyter -c "pip install --upgrade --no-warn-conflicts --no-warn-script-location --user \
        google-cloud-bigquery \
        google-cloud-pipeline-components \
        google-cloud-aiplatform \
        seaborn \
        kfp" >> /tmp/notebook_config.log 2>&1
   

3. Save the file as follows:

   • If you're using vim, press the Esc key, and then type :wq to save the file and exit.
   • If you're using nano, type Control+O and press Enter to save the file, and then type Control+X to exit.

4. Upload the file to your Cloud Storage bucket:

   gcloud storage cp poststartup.sh gs://BUCKET_NAME
   

Create a custom service account

When you create a Vertex AI Workbench instance, we strongly recommend that you clear the Use Compute Engine default service account checkbox and specify a custom service account. If your organization doesn't enforce the iam.automaticIamGrantsForDefaultServiceAccounts organization policy constraint, the Compute Engine default service account (and thus anyone you specify as an instance user) is granted the Editor role (roles/editor) on your project. To turn off this behavior, see Disable automatic role grants to default service accounts.

1. Create a custom service account named workbench-sa:

   gcloud iam service-accounts create workbench-sa \
       --display-name="workbench-sa"
   

2. Assign the Storage Object Viewer IAM role to the service account:

   gcloud projects add-iam-policy-binding $projectid \
       --member="serviceAccount:workbench-sa@$projectid.iam.gserviceaccount.com" \
       --role="roles/storage.objectViewer"
   

3. Assign the Monitoring Metric Writer IAM role to the service account:

   gcloud projects add-iam-policy-binding $projectid \
       --member="serviceAccount:workbench-sa@$projectid.iam.gserviceaccount.com" \
       --role="roles/monitoring.metricWriter"
   

4. Assign the Vertex AI User IAM role to the service account:

   gcloud projects add-iam-policy-binding $projectid \
       --member="serviceAccount:workbench-sa@$projectid.iam.gserviceaccount.com" \
       --role="roles/aiplatform.user"
   

Create a Vertex AI Workbench instance

In this section, you create the Vertex AI Workbench instance. When the instance is created, the post-startup script you created is run automatically.

1. In the Google Cloud console, go to the Instances tab in the Vertex AI Workbench page.

   Go to Vertex AI Workbench

2. Click add_box Create new, and then click Advanced options.

   The New instance page opens.

3. On the New instance page, in the Details section, provide the following information for your new instance and then click Continue:

   • Name: Provide a name for your new instance, or accept the default.
   • Region: Select us-central1.
   • Zone: Select us-central1-a.

4. In the Environment section, provide the following and then click Continue:

   • Post-startup script: Click Browse, click navigate_next View child resources next to the bucket name, click poststartup.sh, and then click Select.

5. In the Machine type section, provide the following and then click Continue:

   • Shielded VM: Select the following checkboxes:

     • Secure Boot
     • Virtual Trusted Platform Module (vTPM)
     • Integrity monitoring

6. In the Disks section, make sure that Google-managed encryption key is selected, and then click Continue:

7. In the Networking section, provide the following and then click Continue:

   • Networking: Select Network in this project and complete the following steps:

     1. In the Network field, select securevertex-vpc.

     2. In the Subnetwork field, select securevertex-subnet-a.

     3. Clear the Assign external IP address checkbox. Not assigning an external IP address prevents the instance from receiving unsolicited communication from the internet or other VPC networks.

     4. Select the Allow proxy access checkbox.

8. In the IAM and security section, provide the following and then click Continue:

   • IAM and security: To grant a single user access to the instance's JupyterLab interface, complete the following steps:

     1. Select Single user.
     2. In the User email field, enter the email address for a single user account. If you're creating the instance for someone else, the following conditions apply:
        • You (the instance creator) don't have access to the instance's JupyterLab interface. But you still control the instance, and you can start, stop, or delete it.
        • After you create the instance, you need to grant the user the Service Account User role (roles/iam.serviceAccountUser) on the instance's service account. See Optional: Grant the Service Account User role to the instance user.
     3. Clear the Use Compute Engine default service account checkbox. This step is important, because the Compute Engine default service account (and thus the single user you just specified) could have the Editor role (roles/editor) on your project.

     4. In the Service account email field, enter the following, replacing PROJECT_ID with the project ID:

        workbench-sa@PROJECT_ID.iam.gserviceaccount.com
        

        (This is the custom service account email address that you created earlier.) This service account has limited permissions.

        To learn more about granting access, see Manage access to a Vertex AI Workbench instance's JupyterLab interface.

   • Security options: Clear the following checkbox:

     • Root access to the instance

     Select the following checkbox:

     • nbconvert: nbconvert lets users export and download a notebook file as a different file type, such as HTML, PDF, or LaTeX. This setting is required by some of the notebooks in the Google Cloud Generative AI GitHub repository.

     Clear the following checkbox:

     • File downloading

     Select the following checkbox, unless you're in a production environment:

     • Terminal access: This enables terminal access to your instance from within the JupyterLab user interface.

9. In the System health section, select Environment auto-upgrade and provide the following:

   • In Reporting, select the following checkboxes:

     • Report system health
     • Report custom metrics to Cloud Monitoring
     • Install Cloud Monitoring
     • Report DNS status for required Google domains

10. Click Create and wait a few minutes for the Vertex AI Workbench instance to be created.

Optional: Grant the Service Account User role to the instance user

If you're creating the Vertex AI Workbench instance for another user, you must grant them the Service Account User role (roles/iam.serviceAccountUser) on the workbench-sa custom service account as follows:

gcloud iam service-accounts add-iam-policy-binding \
    workbench-sa@PROJECT_ID.iam.gserviceaccount.com \
    --member="user:USER_EMAIL" \
    --role="roles/iam.serviceAccountUser"

Replace the following values:

• PROJECT_ID: the project ID
• USER_EMAIL: the email address for the user

Verify that the Vertex AI Workbench instance was created

Vertex AI Workbench creates a Vertex AI Workbench instance based on your specified properties and automatically starts the instance.

When the instance is ready to use, Vertex AI Workbench activates an Open JupyterLab link. This link is accessible only to the single user that you specified at instance creation time.

Open the instance in JupyterLab and verify that the cloned Google Cloud Generative AI GitHub repository is present.

1. In the Google Cloud console, go to the Vertex AI Workbench page.

   Go to Vertex AI Workbench

2. In the list of Vertex AI Workbench instances, click the Open JupyterLab link for the instance you created.

   In the folder list, you'll see a generative-ai folder. This folder contains the cloned GitHub repository.

Monitor health status through Monitoring

You can monitor the system and application metrics for your Vertex AI Workbench instances by using the Google Cloud console. To learn more about instance monitoring and about creating custom metrics, see Monitor health status.

1. In the Google Cloud console, go to the Vertex AI Workbench page.

   Go to Vertex AI Workbench

2. Click the name of the Vertex AI Workbench instance that you want to view the metrics for.

3. On the Instance details page, click the Monitoring tab. Review the CPU Utilization and Network Bytes for your notebook instance. To learn how to interpret these metrics, see Review resource metrics.

   If you just created the instance, you won't see any data right away. Wait a few minutes and refresh the console tab.

Create a VM instance schedule for your Vertex AI Workbench instance

Because a Vertex AI Workbench instance is a Compute Engine VM instance, you can use Compute Engine APIs to create a VM instance schedule for it.

Use a VM instance schedule to start and stop your Vertex AI Workbench instance. During the hours when the instance is stopped, you pay only for Cloud Storage costs.

You can attach an instance schedule to any VM instance that's in the same region, so you can use the same instance schedule to control all your Vertex AI Workbench instances in the region.

To learn more about VM instance schedules, see Scheduling a VM instance to start and stop.

Create a custom IAM role

As a security best practice, we recommend creating a custom IAM role that has only the following permissions and assigning it to the Compute Engine default service account:

• compute.instances.start
• compute.instances.stop

1. Inside Cloud Shell, create a custom role named Vm_Scheduler and include the necessary permissions:

   gcloud iam roles create Vm_Scheduler \
       --project=$projectid \
       --title=vm-scheduler-notebooks \
       --permissions="compute.instances.start,compute.instances.stop" --stage=ga 
   

2. Describe the custom role:

   gcloud iam roles describe Vm_Scheduler \
   --project=$projectid
   

Assign the role to the Compute Engine default service account

To give the Compute Engine default service account permission to start and stop your Vertex AI Workbench instances, you need to assign the Vm_Scheduler custom role to it.

The Compute Engine default service account for your project has the following email address: PROJECT_NUMBER-compute@developer.gserviceaccount.com, where PROJECT_NUMBER is your project number.

1. Identify your project number and store it in the project_number shell variable:

   project_number=$(gcloud projects describe $projectid --format 'get(projectNumber)')
   echo $project_number
   

2. Assign the custom role to the default service account:

   gcloud projects add-iam-policy-binding $projectid \
       --member="serviceAccount:service-$project_number@compute-system.iam.gserviceaccount.com" \
       --role="projects/$projectid/roles/Vm_Scheduler"
   

Create and attach the schedule

To create an instance schedule that starts your Vertex AI Workbench instance at 7 AM and stops it at 6 PM:

1. Create a start and stop schedule named optimize-notebooks:

   gcloud compute resource-policies create instance-schedule optimize-notebooks \
       --region=us-central1 \
       --vm-start-schedule='0 7 * * *' \
       --vm-stop-schedule='0 18 * * *' \
       --timezone=TIME_ZONE
   

   Replace TIME_ZONE with the location-based IANA time zone for this instance schedule, for example, America/Chicago. If omitted, the default value UTC is used. For more information, see time zone.

2. Identify the name of your Vertex AI Workbench instance by running the following command and noting the NAME value that it returns:

   gcloud compute instances list
   

3. Store the name in the notebook_vm shell variable:

   notebook_vm=NOTEBOOK_VM_NAME
   echo $notebook_vm
   

   Replace NOTEBOOK_VM_NAME with your Vertex AI Workbench instance name.

4. Attach the instance schedule to your Vertex AI Workbench instance:

   gcloud compute instances add-resource-policies $notebook_vm \
       --resource-policies=optimize-notebooks \
       --zone=us-central1-a
   

5. Describe the instance schedule:

   gcloud compute resource-policies describe optimize-notebooks \
       --region=us-central1
   

You can verify if the instance schedule runs successfully by checking the Compute Engine audit logs for the instance schedule resource policy and the attached VM instance. You might need to wait for up to 15 minutes after the scheduled time for each operation.