Troubleshoot Colab Enterprise

This page shows you how to resolve issues with Colab Enterprise.

Unable to access Colab Enterprise

The following error occurs when you try to access Colab Enterprise in the Google Cloud console. 

Authentication error: 7; Error Details: User not allowed to access service.

The Google Workspace setting Additional services without individual control is turned off.

To resolve this issue, set Additional services without individual control to On.

Unable to connect to a runtime

This issue occurs due to several reasons. See also Unable to connect to a default runtime. The most common causes are:

Browser blocks third party cookies

The browser that you are using is blocking a third party cookie that Colab Enterprise uses to establish an HTTPS connection with the runtime.

To resolve this issue, configure your browser's settings to allow the DATALAB_TUNNEL_TOKEN third party cookie from the domain *.aiplatform-notebook.googleusercontent.com.

Network blocks outbound traffic to notebook domains

Your network's firewall rules block outbound traffic to *.aiplatform-notebook.cloud.google.com or *aiplatform.googleapis.com.

To resolve this issue, configure your network's firewall rules to allow outbound traffic to *.aiplatform-notebook.cloud.google.com and *aiplatform.googleapis.com.

Unable to connect to a default runtime

The following error occurs when you try to connect to a default runtime when there isn't an existing default runtime template. 

Failed to connect to runtime

When you connect to a default runtime that hasn't been created yet, Colab Enterprise first creates a default runtime template, and then uses that default runtime template to generate the default runtime. However, if you don't have the permissions required to create a runtime template, then Colab Enterprise can't create the default runtime template nor the default runtime.

To resolve this issue, ask your administrator to grant you a role that includes the aiplatform.notebookRuntimeTemplates.create permission.

Notebook slow to connect to runtime

A few minutes after starting a connection to a runtime, the Connecting message is still in the top right corner of the notebook, and Colab Enterprise hasn't connected to the runtime yet.

To resolve this issue, wait a few more minutes. Colab Enterprise is still connecting to a runtime. Connecting to a runtime that isn't already running can take a few minutes because Vertex AI is allocating a new virtual machine (VM) instance.

Unable to create a runtime

This issue occurs when you're unable to create a runtime. See also Unable to create a default runtime. The most common causes are:

Insufficient quota

If you are unable to create a runtime, you might have exceeded your Compute Engine runtime quota.

Colab Enterprise uses Compute Engine quota for runtimes. For more information, see the Compute Engine quota and limits overview.

To resolve this issue, request a higher quota limit.

Unavailable resources

The following error occurs when you try to create a runtime.

No available zone found for runtime RUNTIME_ID
for machine type MACHINE_TYPE
with accelerator type: ACCELERATOR. Please try again later.

This error occurs if there are no resources available for your machine type configuration within your notebook's region.

To resolve this issue, try a runtime with a different machine type configuration, or try again at a later time.

Colab Enterprise unable to create a runtime template

This section shows you how to resolve issues with creating runtime templates.

Unable to select enough GPUs

While creating a runtime template, you can't select the number of GPUs that you want. This might be because you don't have enough quota.

Colab Enterprise uses Compute Engine quota for GPUs. For more information, see the Compute Engine quota and limits overview.

To resolve this issue, request a higher quota limit.

Default runtime already exists

The following error occurs when you try to create a runtime from the default runtime template when the default runtime already exists. 

Failed to create runtime

One click runtime already exists.

If you try to create a runtime from a default runtime template, Colab Enterprise tries to create a default runtime. There can only be one default runtime per user, project, and region. If the default runtime already exists, Colab Enterprise is unable to create another default runtime.

To resolve this issue, connect to the existing default runtime or create a runtime from a non-default runtime template.

Unable to create a default runtime

When Colab Enterprise creates a default runtime, it first creates a default runtime template that it uses to generate the default runtime. If you try to create a default runtime without the permissions required to create a runtime template, then Colab Enterprise can't create the default runtime.

To resolve this issue, ask your administrator to grant you a role that includes the aiplatform.notebookRuntimeTemplates.create permission.

User credentials not found when running code

This issue occurs when you try to run code in a notebook that interacts with Google Cloud services and APIs, but you haven't granted Colab Enterprise access to your user credentials.

The error message might look like one of the following:

Request had invalid authentication credentials.
Expected OAuth 2 access token, login cookie or other valid authentication credential

DefaultCredentialsError: Your default credentials were not found.

See the following common reasons for this issue and their resolutions:

  • You didn't complete a Sign in dialog that appears when you first connect to a runtime that has end-user credentials enabled. By completing this dialog, you grant Colab Enterprise access to your user credentials.

    To resolve this issue, try connecting to the runtime again and grant access.

    To validate that access has been granted:

    1. Click Account > Managed Google Account > Data and Privacy > Third-party Apps & Services.

    2. Verify that Colab Enterprise is listed.

  • In the Sign in dialog (consent screen) that appears when you first connect to a runtime that has end-user credentials enabled, you didn't select your user account to grant Colab Enterprise access to your user credentials.

    To resolve this issue:

    1. In the Google Cloud console, click your account profile image, and then click Google Account.

    2. Click Data & privacy.

    3. In Data from apps and services you use, click Third-party apps & services.

    4. Click Colab Enterprise.

    5. In Colab Enterprise has some access to your Google Account, click See details.

    6. Click Remove access.

    7. Click Confirm.

      This removes your current access settings.

    8. The next time you connect to a runtime that has end-user credentials enabled, make sure to select the correct user account when you complete the Sign in dialog.

  • A pop-up blocker might be preventing the Colab Enterprise Sign in dialog (consent screen) from appearing.

    To resolve this issue, enable pop-ups temporarily in your browser and try to connect to the runtime again.

  • Additional services in Google Workspace have turned off Colab.

    To resolve this issue, enable the Colab service in Google Workspace.

    Even though this refers to Colab, it affects Colab Enterprise.

    Colab should be ON for everyone.

  • Enterprise service restrictions might be blocking access to Colab Enterprise.

    To isolate this issue, verify that your organization's service restrictions aren't blocking access to Google services.

    Service Status should be ON for everyone.

  • You aren't using a runtime with end-user credentials enabled, and you haven't provided your user credentials to Application Default Credentials (ADC) by using the Google Cloud CLI.

    To resolve this issue, see Provide your user credentials to ADC by running code in your notebook.

Gemini features not working

When you type in a Colab Enterprise notebook, code completion suggestions don't appear. Or, the pen_spark Help me code button doesn't appear above a code cell.

In your Colab Enterprise settings, you might have turned off the Automatically trigger code completions setting.

To resolve this issue, do the following:

  1. Click  Toggle header visibility.
  2. On the Tools menu, click Settings.
  3. Click Editor.
  4. Select Automatically trigger code completions.
  5. Click Save.