GitLab agent configuration

DETAILS: Tier: Premium, Ultimate Offering: GitLab.com, Self-managed, GitLab Dedicated

• Introduced in GitLab 15.11 with a flag named remote_development_feature_flag. Disabled by default.
• Enabled on GitLab.com and self-managed in GitLab 16.0.
• Generally available in GitLab 16.7. Feature flag remote_development_feature_flag removed.

When you set up workspace infrastructure, you must configure a GitLab agent to support workspaces. This guide assumes that a GitLab agent is already installed in the Kubernetes cluster.

Prerequisites:

• The agent configuration must have the remote_development module enabled, and the required fields of this module must be correctly set. For more information, see workspace settings.
• The agent must be allowed in a group for the purpose of creating workspaces. During workspace creation, users can select allowed agents that are associated with any parent group of the workspace project.
• The workspace creator must have the Developer role to the project of the agent.

Agent authorization in a group for creating workspaces

• New authorization strategy introduced in GitLab 17.2.

With the new authorization strategy that replaces the legacy authorization strategy, group owners and administrators can control which cluster agents can be used for hosting workspaces in a group.

For example, if the path to your workspace project is top-group/subgroup-1/subgroup-2/workspace-project, you can use any configured agent for either top-group, subgroup-1 or subgroup-2 group.

%%{init: {'theme':'neutral'}}%%

graph TD;

    classDef active fill:lightgreen, stroke:#green, color:green, stroke-width:1px;

    topGroup[Top Group, allowed Agent 1]
    subgroup1[Subgroup 1, allowed Agent 2]
    subgroup2[Subgroup 2, allowed Agent 3]
    wp(Workspace Project, Agent 1, 2 & 3 all available)

    topGroup --> subgroup1
    subgroup1 --> subgroup2
    subgroup2 --> wp

    class wp active;

If a cluster agent is allowed for one group, for example subgroup-1, then the cluster agent is available for all projects under subgroup-1 (including nested projects) to create a workspace. Therefore, you must consider the group where a cluster agent is allowed, as it might affect the scope within which a cluster agent may be available for hosting workspaces.

Allow a cluster agent for workspaces in a group

Prerequisites:

• You must set up workspace infrastructure.
• You must have administrator access to the instance or the Owner role for the group.

To allow a cluster agent for workspaces in a group:

1. On the left sidebar, select Search or go to and find your group.
2. On the left sidebar, select Settings > Workspaces.
3. In the Group agents section, select the All agents tab.
4. From the list of available agents, find the agent with status Blocked, and select Allow.
5. On the confirmation dialog, select Allow agent.

The status of the selected agent is updated to Allowed, and the agent is displayed in the Allowed agents tab.

Remove an allowed cluster agent for workspaces within a group

Prerequisites:

• You must set up workspace infrastructure.
• You must have administrator access to the instance or the Owner role for the group.

To remove an allowed cluster agent from a group:

1. On the left sidebar, select Search or go to and find your group.
2. On the left sidebar, select Settings > Workspaces.
3. In the Group agents section, select the Allowed agents tab.
4. From the list of allowed agents, find the agent you want to remove, and select Block.
5. On the confirmation dialog, select Block agent.

The status of the selected agent is updated to Blocked, and the agent is removed from the Allowed agents tab.

Removing an allowed cluster agent from a group does not immediately stop running workspaces using this agent. Running workspaces stop when they are automatically terminated or manually stopped.

Legacy agent authorization strategy

In GitLab 17.1 and earlier, an agent doesn't have to be allowed to be available in a group for creating workspaces. You can use an agent present anywhere in the top-level group (or the root group) of a workspace project to create a workspace, as long as the remote development module is enabled and you have at least the Developer role for the root group. For example, if the path to your workspace project is top-group/subgroup-1/subgroup-2/workspace-project, you can use any configured agent in top-group and in any of its subgroups.

Workspace settings

NOTE: If a setting has an invalid value, it's not possible to update any setting until you fix that value. Updating any of these settings (except enabled) does not affect existing workspaces.

enabled

Use this setting to define whether:

• The GitLab agent can communicate with the GitLab instance.
• You can create a workspace with the GitLab agent.

The default value is false.

To enable remote development in the agent configuration, set enabled to true:

remote_development:
  enabled: true

If remote development is disabled, an administrator must manually delete any running workspaces to remove those workspaces from the Kubernetes cluster.

dns_zone

Use this setting to define the DNS zone of the URL where workspaces are available.

Example configuration:

remote_development:
  dns_zone: "<workspaces.example.dev>"

gitlab_workspaces_proxy

Use this setting to define the namespace where gitlab-workspaces-proxy is installed. The default value for gitlab_workspaces_proxy.namespace is gitlab-workspaces.

Example configuration:

remote_development:
  gitlab_workspaces_proxy:
    namespace: "<custom-gitlab-workspaces-proxy-namespace>"

network_policy

Use this setting to define the network policy for each workspace. This setting controls network traffic for workspaces.

The default value is:

remote_development:
  network_policy:
    enabled: true
    egress:
      - allow: "0.0.0.0/0"
        except:
          - "10.0.0.0/8"
          - "172.16.0.0/12"
          - "192.168.0.0/16"

In this configuration:

• The network policy is generated for each workspace because enabled is true.
• The egress rules allow all traffic to the internet (0.0.0.0/0) except to the IP CIDR ranges 10.0.0.0/8, 172.16.0.0/12, and 192.168.0.0/16.

The behavior of the network policy depends on the Kubernetes network plugin. For more information, see the Kubernetes documentation.

network_policy.enabled

Use this setting to define whether the network policy is generated for each workspace. The default value for network_policy.enabled is true.

network_policy.egress

• Introduced in GitLab 16.7.

Use this setting to define a list of IP CIDR ranges to allow as egress destinations from a workspace.

Define egress rules when:

• The GitLab instance is on a private IP range.
• The workspace must access a cloud resource on a private IP range.

Each element of the list defines an allow attribute with an optional except attribute. allow defines an IP range to allow traffic from. except lists IP ranges to exclude from the allow range.

Example configuration:

remote_development:
  network_policy:
    egress:
      - allow: "0.0.0.0/0"
        except:
          - "10.0.0.0/8"
          - "172.16.0.0/12"
          - "192.168.0.0/16"
      - allow: "172.16.123.1/32"

In this example, traffic from the workspace is allowed if:

• The destination IP is any range except 10.0.0.0/8, 172.16.0.0/12, or 192.168.0.0/16.
• The destination IP is 172.16.123.1/32.

default_resources_per_workspace_container

• Introduced in GitLab 16.8.

Use this setting to define the default requests and limits for CPU and memory per workspace container. Any resources you define in your devfile override this setting.

For default_resources_per_workspace_container, requests and limits are required. For more information about possible CPU and memory values, see Resource units in Kubernetes.

Example configuration:

remote_development:
  default_resources_per_workspace_container:
    requests:
      cpu: "0.5"
      memory: "512Mi"
    limits:
      cpu: "1"
      memory: "1Gi"

max_resources_per_workspace

• Introduced in GitLab 16.8.

Use this setting to define the maximum requests and limits for CPU and memory per workspace.

For max_resources_per_workspace, requests and limits are required. For more information about possible CPU and memory values, see:

• Resource units in Kubernetes
• Resource quotas

Workspaces fail when they exceed the values you set for requests and limits.

Example configuration:

remote_development:
  max_resources_per_workspace:
    requests:
      cpu: "1"
      memory: "1Gi"
    limits:
      cpu: "2"
      memory: "2Gi"

The maximum resources you define must include any resources required for init containers to perform bootstrapping operations such as cloning the project repository.

workspaces_quota

• Introduced in GitLab 16.9.

Use this setting to set the maximum number of workspaces for the GitLab agent.

You cannot create new workspaces for an agent when:

• The number of workspaces for the agent has reached the defined workspaces_quota.
• workspaces_quota is set to 0.

If workspaces_quota is set to a value below the number of non-terminated workspaces for an agent, the agent's workspaces are not terminated automatically.

The default value is -1 (unlimited). Possible values are greater than or equal to -1.

Example configuration:

remote_development:
  workspaces_quota: 10

workspaces_per_user_quota

• Introduced in GitLab 16.9.

Use this setting to set the maximum number of workspaces per user.

You cannot create new workspaces for a user when:

• The number of workspaces for the user has reached the defined workspaces_per_user_quota.
• workspaces_per_user_quota is set to 0.

If workspaces_per_user_quota is set to a value below the number of non-terminated workspaces for a user, the user's workspaces are not terminated automatically.

The default value is -1 (unlimited). Possible values are greater than or equal to -1.

Example configuration:

remote_development:
  workspaces_per_user_quota: 3

use_kubernetes_user_namespaces

• Introduced in GitLab 17.4.

Use this setting to specify whether to use the user namespaces feature in Kubernetes.

User namespaces isolate the user running inside the container from the user in the host. In Kubernetes 1.30, this feature is in beta.

The default value is false. Before you set the value to true, ensure your Kubernetes cluster supports user namespaces.

Example configuration:

remote_development:
  use_kubernetes_user_namespaces: true

For more information about use_kubernetes_user_namespaces, see user namespaces.

default_runtime_class

• Introduced in GitLab 17.4.

Use this setting to select the container runtime configuration used to run the containers in the workspace.

The default value is "", which denotes the absence of a value.

Example configuration:

remote_development:
  default_runtime_class: "example-runtime-class-name"

A valid value:

• Contains 253 characters or less.
• Contains only lowercase letters, numbers, -, or ..
• Starts with an alphanumeric character
• Ends with an alphanumeric character.

For more information about default_runtime_class, see Runtime Class.

allow_privilege_escalation

• Introduced in GitLab 17.4.

Use this setting to control whether a process can gain more privileges than its parent process.

This setting directly controls whether the no_new_privs flag gets set on the container process.

The default value is false. The value can be set to true only if either:

• default_runtime_class is set to a non-empty value.
• use_kubernetes_user_namespaces is set to true.

Example configuration:

remote_development:
  default_runtime_class: "example-runtime-class-name"
  allow_privilege_escalation: true

For more information about allow_privilege_escalation, see Configure a Security Context for a Pod or Container.

annotations

• Introduced in GitLab 17.4.

Use this setting to attach arbitrary non-identifying metadata to the Kubernetes objects.

The default value is {}.

Example configuration:

remote_development:
  annotations:
    "example.com/key": "value"

A valid annotation key is a string made of two parts:

• Optional. A prefix. The prefix must be 253 characters or less, and contain period-separated DNS labels. The prefix must end with a slash (/).
• A name. The name must be 63 characters or less and contain only alphanumeric characters, dashes (-), underscores (_), and periods (.). The name must begin and end with an alphanumeric character.

You shouldn't use prefixes that end with kubernetes.io and k8s.io because they are reserved for Kubernetes core components. Prefixes that end with gitlab.com are also reserved.

A valid annotation value is a string.

For more information about annotations, see Annotations.

labels

• Introduced in GitLab 17.4.

Use this setting to attach arbitrary identifying metadata to the Kubernetes objects.

The default value is {}.

Example configuration:

remote_development:
  labels:
    "example.com/key": "value"

A label key is a string made of two parts:

• Optional. A prefix. The prefix must be 253 characters or less, and contain period-separated DNS labels. The prefix must end with a slash (/).
• A name. The name must be 63 characters or less and contain only alphanumeric characters, dashes (-), underscores (_), and periods (.). The name must begin and end with an alphanumeric character.

You shouldn't use prefixes that end with kubernetes.io and k8s.io because they are reserved for Kubernetes core components. Prefixes that end with gitlab.com are also reserved.

A valid label value:

• Contains 63 characters or less. The value can be empty.
• Begins and ends with an alphanumeric character.
• Can contain dashes (-), underscores (_), and periods (.).

For more information about labels, see Labels.

Configuring user access with remote development

You can configure the user_access module to access the connected Kubernetes cluster with your GitLab credentials. This module is configured and runs independently of the remote_development module.

Be careful when configuring both user_access and remote_development in the same GitLab agent. The remote_development clusters manage user credentials (such as personal access tokens) as Kubernetes Secrets. Any misconfiguration in user_access might cause this private data to be accessible over the Kubernetes API.

For more information about configuring user_access, see Configure Kubernetes access.