• How the runner creates Kubernetes pods
• Connect to the Kubernetes API
  • Set the bearer token for Kubernetes API calls
  • Configure runner API permissions
• Configuration settings
  • CPU requests and limits
  • Memory requests and limits
  • Storage requests and limits
  • Other config.toml settings
  • Configuration example
• Configure the executor service account
• Pods and containers
  • Default annotations for job pods
  • Pod lifecycle
  • Overwrite pod tolerations
  • Overwrite pod labels
  • Overwrite pod annotations
  • Overwrite generated pod specifications
    • Merge patch strategy
    • JSON patch strategy
    • Strategic patch strategy
    • Best practices
    • Create a PVC for each build job by modifying the Pod Spec
  • Set a security policy for the pod
  • Remove old runner pods
  • Set a security policy for the container
  • Set a pull policy
  • Specify container capabilities
  • Overwrite container resources
  • Define a list of services
  • Overwrite service containers resources
  • Overwrite the Kubernetes default service account
  • Set the RuntimeClass
  • Change the base directory for build logs and scripts
  • User namespaces
• Operating system, architecture, and Windows kernel version
• Nodes
  • Specify the node to execute builds
    • Example for linux/arm64
    • Example for windows/amd64
  • Overwrite the node selector
  • Define a list of node affinities
  • Define nodes where pods are scheduled
• Networking
  • Configure a container lifecycle hook
    • KubernetesLifecycleExecAction
    • KubernetesLifecycleHTTPGet
    • KubernetesLifecycleHTTPGetHeader
    • KubernetesLifecycleTcpSocket
  • Configure pod DNS settings
    • KubernetesDNSConfigOption
    • Default list of dropped capabilities
  • Add extra host aliases
• Volumes
  • Using the cache with the Kubernetes executor
  • Configure volume types
    • hostPath volume
    • persistentVolumeClaim volume
    • configMap volume
    • secret volume
    • emptyDir volume
    • csi volume
  • Mount volumes on service containers
  • Custom volume mount
  • Persistent per-concurrency build volumes
  • Use a helper image
• Using Docker in builds
  • Exposed /var/run/docker.sock
  • Using docker:dind
  • Prevent host kernel exposure
  • Use kaniko to build Docker images
  • Restrict Docker images and services
  • Restrict Docker pull policies
• Job execution
  • Configure the number of request attempts to the Kubernetes API
  • Container entrypoint known issues
  • Restrict access to job variables
• Resources check during prepare step
  • Overwrite the Kubernetes namespace
• Fault tolerance in Kubernetes clusters

Kubernetes executor

Use the Kubernetes executor to use Kubernetes clusters for your builds. The executor calls the Kubernetes cluster API and creates a pod for each GitLab CI job.

The Kubernetes executor divides the build into multiple steps:

1. Prepare: Create the Pod against the Kubernetes Cluster. This creates the containers required for the build and services to run.
2. Pre-build: Clone, restore cache, and download artifacts from previous stages. This step runs on a special container as part of the pod.
3. Build: User build.
4. Post-build: Create cache, upload artifacts to GitLab. This step also uses the special container as part of the pod.

How the runner creates Kubernetes pods

The following diagram shows the interaction between a GitLab instance and a runner hosted on a Kubernetes cluster. The runner calls the Kubernetes API to create pods on the cluster.

The pod consists of the following containers for each service defined in the .gitlab-ci.yml or config.toml files:

• A build container defined as build.
• A helper container defined as helper.
• A services containers defined as svc-X, where X is [0-9]+.

Services and containers run in the same Kubernetes pod and share the same localhost address. The following restrictions apply:

• In GitLab Runner 12.8 and Kubernetes 1.7 and later, the services are accessible through their DNS names. If you use an older version, you must use localhost.
• You cannot use several services that use the same port. For example, you cannot have two mysql services at the same time.

The interaction in the diagram is valid for any Kubernetes cluster. For example, turnkey solutions hosted on the major public cloud providers, or self-managed Kubernetes installations.

Connect to the Kubernetes API

Use the following options to connect to the Kubernetes API. The user account provided must have permission to create, list, and attach to Pods in the specified namespace.

If you’re running GitLab Runner in the Kubernetes cluster, you should omit all of these fields so that the GitLab Runner auto-discovers the Kubernetes API.

If you’re running GitLab Runner externally to the Cluster, then you must set each of these settings and ensure that GitLab Runner has access to the Kubernetes API on the cluster.

Set the bearer token for Kubernetes API calls

To set the bearer token for API calls to create pods, use the KUBERNETES_BEARER_TOKEN variable. This allows project owners to use project secret variables to specify a bearer token.

When specifying the bearer token, you must set the Host configuration setting.

variables:
  KUBERNETES_BEARER_TOKEN: thebearertokenfromanothernamespace

Configure runner API permissions

To configure permissions for the core API group, update the values.yml file for GitLab Runner Helm charts.

You can either:

• Set rbac.create to true.
• Specify a service account rbac.serviceAccountName: <service_account_name> with the following permissions in the values.yml file.

• The serviceAccount permission is needed only:

  • For GitLab 15.0 and 15.1.
  • For GitLab 15.0.1, 15.1.1, and 15.2 when resource_availability_check_max_attempts is set to a value higher than 0.

• As of GitLab Runner 15.8 the configmaps permission is no longer needed.

• The event permission is needed only:

  • For GitLab 16.2.1 and later.

• The namespace permission is needed only:

  • When enabling namespace isolation via namespace_per_job.

• The pods/log permission is only needed when one of the following scenarios are true:

  • The FF_KUBERNETES_HONOR_ENTRYPOINT feature flag is enabled.

  • The FF_USE_LEGACY_KUBERNETES_EXECUTION_STRATEGY feature flag is disabled when the CI_DEBUG_SERVICES variable is set to true.

  • The FF_WAIT_FOR_POD_TO_BE_REACHABLE feature flag is enabled.

Configuration settings

Use the following settings in the config.toml file to configure the Kubernetes executor.

CPU requests and limits

Memory requests and limits

Storage requests and limits

Other config.toml settings

Configuration example

The following sample shows an example configuration of the config.toml file for the Kubernetes executor.

concurrent = 4

[[runners]]
  name = "myRunner"
  url = "https://gitlab.com/ci"
  token = "......"
  executor = "kubernetes"
  [runners.kubernetes]
    host = "https://45.67.34.123:4892"
    cert_file = "/etc/ssl/kubernetes/api.crt"
    key_file = "/etc/ssl/kubernetes/api.key"
    ca_file = "/etc/ssl/kubernetes/ca.crt"
    namespace = "gitlab"
    namespace_overwrite_allowed = "ci-.*"
    bearer_token_overwrite_allowed = true
    privileged = true
    cpu_limit = "1"
    memory_limit = "1Gi"
    service_cpu_limit = "1"
    service_memory_limit = "1Gi"
    helper_cpu_limit = "500m"
    helper_memory_limit = "100Mi"
    poll_interval = 5
    poll_timeout = 3600
    dns_policy = "cluster-first"
    priority_class_name = "priority-1"
    logs_base_dir = "/tmp"
    scripts_base_dir = "/tmp"
    [runners.kubernetes.node_selector]
      gitlab = "true"
    [runners.kubernetes.node_tolerations]
      "node-role.kubernetes.io/master" = "NoSchedule"
      "custom.toleration=value" = "NoSchedule"
      "empty.value=" = "PreferNoSchedule"
      "onlyKey" = ""

Configure the executor service account

To configure the executor service account, you can set the KUBERNETES_SERVICE_ACCOUNT environment variable or use the --kubernetes-service-account flag.

Pods and containers

You can configure pods and containers to control how jobs are executed.

Default annotations for job pods

The following annotations are added by default on the Pod running the jobs:

To overwrite default annotations, use the pod_annotations in the GitLab Runner configuration. You can also overwrite annotations for each CI/CD job in the .gitlab-ci.yml file.

Pod lifecycle

A pod’s lifecycle can be affected by:

• Setting the pod_termination_grace_period_seconds property in the TOML configuration file. The process running on the pod can run for the given duration after the TERM signal is sent. A kill signal is sent if the Pod is not successfully terminated after this period of time.
• Enabling the FF_USE_POD_ACTIVE_DEADLINE_SECONDS feature flag. When enabled and the job times out, the pod running the CI/CD job is marked as failed and all associated containers are killed. To have the job time out on GitLab first, activeDeadlineSeconds is set to configured timeout + 1 second.

Overwrite pod tolerations

To overwrite Kubernetes pod tolerations:

1. In the config.toml or Helm values.yaml file, to enable the overwrite of CI job pod tolerations, define a regular expression for node_tolerations_overwrite_allowed. This regular expression validates the values of CI variable names that start with KUBERNETES_NODE_TOLERATIONS_.

   runners:
    ...
    config: |
      [[runners]]
        [runners.kubernetes]
          node_tolerations_overwrite_allowed = ".*"
   

2. In the .gitlab-ci.yml file, define one or more CI variables to overwrite CI job pod tolerations.

   variables:
     KUBERNETES_NODE_TOLERATIONS_1: 'node-role.kubernetes.io/master:NoSchedule'
     KUBERNETES_NODE_TOLERATIONS_2: 'custom.toleration=value:NoSchedule'
     KUBERNETES_NODE_TOLERATIONS_3: 'empty.value=:PreferNoSchedule'
     KUBERNETES_NODE_TOLERATIONS_4: 'onlyKey'
     KUBERNETES_NODE_TOLERATIONS_5: '' # tolerate all taints
   

Overwrite pod labels

To overwrite Kubernetes pod labels for each CI/CD job:

1. In the .config.yaml file, define a regular expression for pod_labels_overwrite_allowed.

2. In the .gitlab-ci.yml file, set the KUBERNETES_POD_LABELS_* variables with values of key=value. The pod labels are overwritten to the key=value. You can apply multiple values:

    variables:
      KUBERNETES_POD_LABELS_1: "Key1=Val1"
      KUBERNETES_POD_LABELS_2: "Key2=Val2"
      KUBERNETES_POD_LABELS_3: "Key3=Val3"
   

Overwrite pod annotations

To overwrite Kubernetes pod annotations for each CI/CD job:

1. In the .config.yaml file, define a regular expression for pod_annotations_overwrite_allowed.

2. In the .gitlab-ci.yml file, set the KUBERNETES_POD_ANNOTATIONS_* variables and use key=value for the value. Pod annotations are overwritten to the key=value. You can specify multiple annotations:

   variables:
     KUBERNETES_POD_ANNOTATIONS_1: "Key1=Val1"
     KUBERNETES_POD_ANNOTATIONS_2: "Key2=Val2"
     KUBERNETES_POD_ANNOTATIONS_3: "Key3=Val3"
   

In the example below, the pod_annotations and the pod_annotations_overwrite_allowed are set. This configuration allows overwrite of any of the pod_annotations configured in the config.toml.

[[runners]]
  # usual configuration
  executor = "kubernetes"
  [runners.kubernetes]
    image = "alpine"
    pod_annotations_overwrite_allowed = ".*"
    [runners.kubernetes.pod_annotations]
      "Key1" = "Val1"
      "Key2" = "Val2"
      "Key3" = "Val3"
      "Key4" = "Val4"

Overwrite generated pod specifications

This feature is in beta. We strongly recommend that you use this feature on a test Kubernetes cluster before you use it on a production cluster. To use this feature, you must enable the FF_USE_ADVANCED_POD_SPEC_CONFIGURATION feature flag.

To add feedback for improvements before the feature is made generally available, use this issue.

To modify the PodSpec generated by the runner manager, use the pod_spec setting in the config.toml file.

The pod_spec setting:

• Overwrites and completes fields for the generated pod specification.
• Overwrites configuration values that might have been set in your config.toml under [runners.kubernetes].

You can configure multiple pod_spec settings.

You cannot set the patch_path and patch in the same pod_spec configuration, otherwise an error occurs.

Example of multiple pod_spec configurations in the config.toml:

[[runners]]
  [runners.kubernetes]
    [[runners.kubernetes.pod_spec]]
      name = "hostname"
      patch = '''
        hostname: "custom-pod-hostname"
      '''
      patch_type = "merge"
    [[runners.kubernetes.pod_spec]]
      name = "subdomain"
      patch = '''
        subdomain: "subdomain"
      '''
      patch_type = "strategic"
    [[runners.kubernetes.pod_spec]]
      name = "terminationGracePeriodSeconds"
      patch = '''
        [{"op": "replace", "path": "/terminationGracePeriodSeconds", "value": 60}]
      '''
      patch_type = "json"

Merge patch strategy

The merge patch strategy applies a key-value replacement on the existing PodSpec. If you use this strategy, the pod_spec configuration in the config.toml overwrites the values in the final PodSpec object before it is generated. Because the values are completely overwritten, you should use this patch strategy with caution.

Example of a pod_spec configuration with the merge patch strategy:

concurrent = 1
check_interval = 1
log_level = "debug"
shutdown_timeout = 0

[session_server]
  session_timeout = 1800

[[runners]]
  name = ""
  url = "https://gitlab.example.com"
  id = 0
  token = "__REDACTED__"
  token_obtained_at = 0001-01-01T00:00:00Z
  token_expires_at = 0001-01-01T00:00:00Z
  executor = "kubernetes"
  shell = "bash"
  environment = ["FF_USE_ADVANCED_POD_SPEC_CONFIGURATION=true", "CUSTOM_VAR=value"]
  [runners.kubernetes]
    image = "alpine"
    ...
    [[runners.kubernetes.pod_spec]]
      name = "build envvars"
      patch = '''
        containers:
        - env:
          - name: env1
            value: "value1"
          - name: env2
            value: "value2"
          name: build
      '''
      patch_type = "merge"

With this configuration, the final PodSpec has only one container called build with two environment variables env1 and env2. The example above make the related CI Job failed as:

• The helper container specification is removed.
• The build container specification lost all necessary configuration set by GitLab Runner.

To prevent the job from failing, in this example, the pod_spec must contain the untouched properties generated by GitLab Runner.

JSON patch strategy

The json patch strategy uses the JSON Patch specification to give control over the PodSpec objects and arrays to update. You cannot use this strategy on array properties.

Example of a pod_spec configuration with the json patch strategy. In this configuration, a new key: value pair is added to the existing nodeSelector. The existing values are not overwritten.

concurrent = 1
check_interval = 1
log_level = "debug"
shutdown_timeout = 0

[session_server]
  session_timeout = 1800

[[runners]]
  name = ""
  url = "https://gitlab.example.com"
  id = 0
  token = "__REDACTED__"
  token_obtained_at = 0001-01-01T00:00:00Z
  token_expires_at = 0001-01-01T00:00:00Z
  executor = "kubernetes"
  shell = "bash"
  environment = ["FF_USE_ADVANCED_POD_SPEC_CONFIGURATION=true", "CUSTOM_VAR=value"]
  [runners.kubernetes]
    image = "alpine"
    ...
    [[runners.kubernetes.pod_spec]]
      name = "val1 node"
      patch = '''
        { "op": "add", "path": "/nodeSelector", "value": { key1: "val1" } }
      '''
      patch_type = "json"

Strategic patch strategy

This strategic patch strategy uses the existing patchStrategy applied to each field of the PodSpec object.

Example of a pod_spec configuration with the strategic patch strategy. In this configuration, a resource request is set to on the build container.

concurrent = 1
check_interval = 1
log_level = "debug"
shutdown_timeout = 0

[session_server]
  session_timeout = 1800

[[runners]]
  name = ""
  url = "https://gitlab.example.com"
  id = 0
  token = "__REDACTED__"
  token_obtained_at = 0001-01-01T00:00:00Z
  token_expires_at = 0001-01-01T00:00:00Z
  executor = "kubernetes"
  shell = "bash"
  environment = ["FF_USE_ADVANCED_POD_SPEC_CONFIGURATION=true", "CUSTOM_VAR=value"]
  [runners.kubernetes]
    image = "alpine"
    ...
    [[runners.kubernetes.pod_spec]]
      name = "cpu request 500m"
      patch = '''
        containers:
        - name: build
          resources:
            requests:
              cpu: "500m"
      '''
      patch_type = "strategic"

With this configuration, a resource request is set to on the build container.

Best practices

• Test the added pod_spec in a test environment before deployment in a production environment.
• Make sure that the pod_spec configuration does not negatively impact the GitLab Runner generated specification.
• Do not use the merge patch strategy for complex pod specification updates.
• Where possible, use the config.toml when the configuration is available. For example, the following configuration replaces the first environment variables set by GitLab Runner by the one set in the custom pod_spec instead of adding the environment variable set to the existing list.

concurrent = 1
check_interval = 1
log_level = "debug"
shutdown_timeout = 0

[session_server]
  session_timeout = 1800

[[runners]]
  name = ""
  url = "https://gitlab.example.com"
  id = 0
  token = "__REDACTED__"
  token_obtained_at = 0001-01-01T00:00:00Z
  token_expires_at = 0001-01-01T00:00:00Z
  executor = "kubernetes"
  shell = "bash"
  environment = ["FF_USE_ADVANCED_POD_SPEC_CONFIGURATION=true", "CUSTOM_VAR=value"]
  [runners.kubernetes]
    image = "alpine"
    ...
    [[runners.kubernetes.pod_spec]]
      name = "build envvars"
      patch = '''
        containers:
        - env:
          - name: env1
            value: "value1"
          name: build
      '''
      patch_type = "strategic"

Create a PVC for each build job by modifying the Pod Spec

To create a PersistentVolumeClaim for each build job make sure to check out how to enable the Pod Spec functionality.

Kubernetes allows to create an ephemeral PersistentVolumeClaim attached to the lifecycle of a Pod. This will work if dynamic provisioning is enabled on your Kubernetes cluster allowing each PVC to request a new Volume, the volume too will be tied to the lifetime of the Pod.

After dynamic provisioning is enabled, the config.toml can be modified as follows to create an ephemeral PVC:

[[runners.kubernetes.pod_spec]]
  name = "ephemeral-pvc"
  patch = '''
    containers:
    - name: build
      volumeMounts:
      - name: builds
        mountPath: /builds
    - name: helper
      volumeMounts:
      - name: builds
        mountPath: /builds
    volumes:
    - name: builds
      ephemeral:
        volumeClaimTemplate:
          spec:
            storageClassName: <The Storage Class that will dynamically provision a Volume>
            accessModes: [ ReadWriteOnce ]
            resources:
              requests:
                storage: 1Gi
  '''

Set a security policy for the pod

Configure the security context in the config.toml to set a security policy for the build pod.

Use the following options:

Example of a pod security context in the config.toml:

concurrent = %(concurrent)s
check_interval = 30
  [[runners]]
    name = "myRunner"
    url = "gitlab.example.com"
    executor = "kubernetes"
    [runners.kubernetes]
      helper_image = "gitlab-registry.example.com/helper:latest"
      [runners.kubernetes.pod_security_context]
        run_as_non_root = true
        run_as_user = 59417
        run_as_group = 59417
        fs_group = 59417

Remove old runner pods

Sometimes old runner pods are not cleared. This can happen when the runner manager is incorrectly shut down.

To handle this situation, you can use the GitLab Runner Pod Cleanup application to schedule cleanup of old pods. For more information, see:

• The GitLab Runner Pod Cleanup project README.
• GitLab Runner Pod Cleanup documentation.

Set a security policy for the container

Configure the container security context in the config.toml executor to set a container security policy for the build, helper, or service pods.

Use the following options:

In the following example in the config.toml, the security context configuration:

• Sets a pod security context.
• Overrides run_as_user and run_as_group for the build and helper containers.
• Specifies that all service containers inherit run_as_user and run_as_group from the pod security context.

concurrent = 4
check_interval = 30
  [[runners]]
    name = "myRunner"
    url = "gitlab.example.com"
    executor = "kubernetes"
    [runners.kubernetes]
      helper_image = "gitlab-registry.example.com/helper:latest"
      [runners.kubernetes.pod_security_context]
        run_as_non_root = true
        run_as_user = 59417
        run_as_group = 59417
        fs_group = 59417
      [runners.kubernetes.init_permissions_container_security_context]
        run_as_user = 1000
        run_as_group = 1000
      [runners.kubernetes.build_container_security_context]
        run_as_user = 65534
        run_as_group = 65534
        [runners.kubernetes.build_container_security_context.capabilities]
          add = ["NET_ADMIN"]
      [runners.kubernetes.helper_container_security_context]
        run_as_user = 1000
        run_as_group = 1000
      [runners.kubernetes.service_container_security_context]
        run_as_user = 1000
        run_as_group = 1000

Set a pull policy

Use the pull_policy parameter in the config.toml file to specify a single or multiple pull policies. The policy controls how an image is fetched and updated, and applies to the build image, helper image, and any services.

To determine which policy to use, see the Kubernetes documentation about pull policies.

For a single pull policy:

[runners.kubernetes]
  pull_policy = "never"

For multiple pull policies:

[runners.kubernetes]
  # use multiple pull policies
  pull_policy = ["always", "if-not-present"]

When you define multiple policies, each policy is attempted until the image is obtained successfully. For example, when you use [ always, if-not-present ], the policy if-not-present is used if the always policy fails due to a temporary registry problem.

To retry a failed pull:

[runners.kubernetes]
  pull_policy = ["always", "always"]

The GitLab naming convention is different to the Kubernetes one.

Specify container capabilities

You can specify the Kubernetes capabilities to use in the container.

To specify the container capabilities, use the cap_add and cap_drop options in the config.toml. Container runtimes can also define a default list of capabilities, like those in Docker or the container.

There is a list of capabilities that the runner drops by default. Capabilities that you list in cap_add option are excluded from being dropped.

Example configuration in the config.toml file:

concurrent = 1
check_interval = 30
[[runners]]
  name = "myRunner"
  url = "gitlab.example.com"
  executor = "kubernetes"
  [runners.kubernetes]
    # ...
    cap_add = ["SYS_TIME", "IPC_LOCK"]
    cap_drop = ["SYS_ADMIN"]
    # ...

When you specify the capabilities:

• User-defined cap_drop has priority over user-defined cap_add. If you define the same capability in both settings, only the capability from cap_drop is passed to the container.
• Remove the CAP_ prefix from capability identifiers passed to the container configuration. For example, if you want to add or drop the CAP_SYS_TIME capability, in the configuration file, enter the string, SYS_TIME.
• The owner of the Kubernetes cluster can define a PodSecurityPolicy, where specific capabilities are allowed, restricted, or added by default. These rules take precedence over any user-defined configuration.

Overwrite container resources

You can overwrite Kubernetes CPU and memory allocations for each CI/CD job. You can apply settings for requests and limits for the build, helper, and service containers.

To overwrite container resources, use the following variables in the .gitlab-ci.yml file.

The values for the variables are restricted to the maximum overwrite setting for that resource. If the maximum overwrite has not been set for a resource, the variable is not used.

 variables:
   KUBERNETES_CPU_REQUEST: "3"
   KUBERNETES_CPU_LIMIT: "5"
   KUBERNETES_MEMORY_REQUEST: "2Gi"
   KUBERNETES_MEMORY_LIMIT: "4Gi"
   KUBERNETES_EPHEMERAL_STORAGE_REQUEST: "512Mi"
   KUBERNETES_EPHEMERAL_STORAGE_LIMIT: "1Gi"

   KUBERNETES_HELPER_CPU_REQUEST: "3"
   KUBERNETES_HELPER_CPU_LIMIT: "5"
   KUBERNETES_HELPER_MEMORY_REQUEST: "2Gi"
   KUBERNETES_HELPER_MEMORY_LIMIT: "4Gi"
   KUBERNETES_HELPER_EPHEMERAL_STORAGE_REQUEST: "512Mi"
   KUBERNETES_HELPER_EPHEMERAL_STORAGE_LIMIT: "1Gi"

   KUBERNETES_SERVICE_CPU_REQUEST: "3"
   KUBERNETES_SERVICE_CPU_LIMIT: "5"
   KUBERNETES_SERVICE_MEMORY_REQUEST: "2Gi"
   KUBERNETES_SERVICE_MEMORY_LIMIT: "4Gi"
   KUBERNETES_SERVICE_EPHEMERAL_STORAGE_REQUEST: "512Mi"
   KUBERNETES_SERVICE_EPHEMERAL_STORAGE_LIMIT: "1Gi"

Define a list of services

Define a list of services in the config.toml.

concurrent = 1
check_interval = 30
  [[runners]]
    name = "myRunner"
    url = "gitlab.example.com"
    executor = "kubernetes"
    [runners.kubernetes]
      helper_image = "gitlab-registy.example.com/helper:latest"
      [[runners.kubernetes.services]]
        name = "postgres:12-alpine"
        alias = "db1"
      [[runners.kubernetes.services]]
        name = "registry.example.com/svc1"
        alias = "svc1"
        entrypoint = ["entrypoint.sh"]
        command = ["executable","param1","param2"]
        environment = ["ENV=value1", "ENV2=value2"]

If the service environment includes HEALTHCHECK_TCP_PORT, GitLab Runner waits until the service responds on that port before starting user CI scripts. You can also configure the HEALTHCHECK_TCP_PORT environment variable in a services section of .gitlab-ci.yml.

Overwrite service containers resources

If a job has multiple service containers, you can set explicit resource requests and limits to each service container. Use the variables attribute in each service to overwrite container resources specified in .gitlab-ci.yml.

  services:
    - name: redis:5
      alias: redis5
      variables:
        KUBERNETES_SERVICE_CPU_REQUEST: "3"
        KUBERNETES_SERVICE_CPU_LIMIT: "6"
        KUBERNETES_SERVICE_MEMORY_REQUEST: "3Gi"
        KUBERNETES_SERVICE_MEMORY_LIMIT: "6Gi"
        KUBERNETES_EPHEMERAL_STORAGE_REQUEST: "2Gi"
        KUBERNETES_EPHEMERAL_STORAGE_LIMIT: "3Gi"
    - name: postgres:12
      alias: MY_relational-database.12
      variables:
        KUBERNETES_CPU_REQUEST: "2"
        KUBERNETES_CPU_LIMIT: "4"
        KUBERNETES_MEMORY_REQUEST: "1Gi"
        KUBERNETES_MEMORY_LIMIT: "2Gi"
        KUBERNETES_EPHEMERAL_STORAGE_REQUEST: "1Gi"
        KUBERNETES_EPHEMERAL_STORAGE_LIMIT: "2Gi"

These specific settings take precedence over the general settings for the job. The values are still restricted to the maximum overwrite setting for that resource.

Overwrite the Kubernetes default service account

To overwrite the Kubernetes service account for each CI/CD job in the .gitlab-ci.yml file, set the variable KUBERNETES_SERVICE_ACCOUNT_OVERWRITE.

You can use this variable to specify a service account attached to the namespace, which you may need for complex RBAC configurations.

variables:
  KUBERNETES_SERVICE_ACCOUNT_OVERWRITE: ci-service-account

To ensure only designated service accounts are used during CI runs, define a regular expression for either:

• The service_account_overwrite_allowed setting.
• The KUBERNETES_SERVICE_ACCOUNT_OVERWRITE_ALLOWED environment variable.

If you don’t set either, the overwrite is disabled.

Set the RuntimeClass

Use runtime_class_name to set the RuntimeClass for each job container.

If you specify a RuntimeClass name and it’s not configured in the cluster or the feature is not supported, the executor fails to create jobs.

concurrent = 1
check_interval = 30
  [[runners]]
    name = "myRunner"
    url = "gitlab.example.com"
    executor = "kubernetes"
    [runners.kubernetes]
      runtime_class_name = "myclass"

Change the base directory for build logs and scripts

You can change the directory where emptyDir volumes are mounted to the pod for build logs and scripts. You can use the directory to:

• Run job pods with a modified image.
• Run as an unprivileged user.
• Customize SecurityContext settings.

To change the directory:

• For build logs, set logs_base_dir.
• For build scripts, set scripts_base_dir.

The expected value is a string that represents a base directory without the trailing slash (for example, /tmp or /mydir/example). The directory must already exist.

This value is prepended to the generated path for build logs and scripts. For example:

  [[runners]]
    name = "myRunner"
    url = "gitlab.example.com"
    executor = "kubernetes"
    [runners.kubernetes]
      logs_base_dir = "/tmp"
      scripts_base_dir = "/tmp"

This configuration would result in an emptyDir volume mounted in:

• /tmp/logs-${CI_PROJECT_ID}-${CI_JOB_ID} for build logs instead of the default /logs-${CI_PROJECT_ID}-${CI_JOB_ID}.
• /tmp/scripts-${CI_PROJECT_ID}-${CI_JOB_ID} for build scripts.

User namespaces

In Kubernetes 1.30 and later, you can isolate the user running in the container from the one on the host with user namespaces. A process running as root in the container can run as a different unprivileged user on the host.

With user namespaces, you can have more control over which images are used to run your CI/CD jobs. Operations that require additional settings (such as running as root) can also function without opening up additional attack surface on the host.

To use this feature, ensure your cluster has been properly configured. The following example adds pod_spec for the hostUsers key and disables both privileged pods and privilege escalation:

  [[runners]]
    environment = ["FF_USE_ADVANCED_POD_SPEC_CONFIGURATION=true"]
    builds_dir = "/tmp/builds"
  [runners.kubernetes]
    logs_base_dir = "/tmp"
    scripts_base_dir = "/tmp"
    privileged = false
    allowPrivilegeEscalation = false
  [[runners.kubernetes.pod_spec]]
    name = "hostUsers"
    patch = '''
      [{"op": "add", "path": "/hostUsers", "value": false}]
    '''
    patch_type = "json"

With user namespaces, you cannot use the default path for the build directory (builds_dir), the build logs (logs_base_dir), or the build scripts (scripts_base_dir). Otherwise, even the container’s root user does not have permission to mount volumes or create directories in the root of the container’s file system.

Instead, you can change the base directory for build logs and scripts. You can also change the build directory by setting [[runners]].builds_dir.

Operating system, architecture, and Windows kernel version

GitLab Runner with the Kubernetes executor can run builds on different operating systems if the configured cluster has nodes running those operating systems.

The system determines the helper image’s operating system, architecture, and Windows kernel version (if applicable). It then uses those parameters for other aspects of the build, for example the containers or images to use.

The following diagram explains how the system detects these details:

The following are the only parameters that influence the operating system, architecture, and Windows kernel version selection of the build.

• The helper_image_autoset_arch_and_os configuration
• The kubernetes.io/os, kubernetes.io/arch, and node.kubernetes.io/windows-build label selectors from:
  • node_selector configuration
  • node_selector overwrites

Other parameters don’t have any influence on the above outlined selection process. However, other parameters, for example, affinity configuration, can be used to further limit the nodes on which builds will be scheduled on.

Nodes

Specify the node to execute builds

Use the node_selector option to specify which node in a Kubernetes cluster can be used to execute the builds. It is a key=value pair in string=string format (string:string in the case of environment variables).

Runner uses the information provided to determine the operating system and architecture for the build. This ensures that the correct helper image is used. The default operating system and architecture is linux/amd64.

You can use specific labels to schedule nodes with different operating systems and architectures.

Example for linux/arm64

  [[runners]]
    name = "myRunner"
    url = "gitlab.example.com"
    executor = "kubernetes"

    [runners.kubernetes.node_selector]
      "kubernetes.io/arch" = "arm64"
      "kubernetes.io/os" = "linux"

Example for windows/amd64

Kubernetes for Windows has certain limitations. If you are using process isolation, you must also provide the specific Windows build version with the node.kubernetes.io/windows-build label.

  [[runners]]
    name = "myRunner"
    url = "gitlab.example.com"
    executor = "kubernetes"

    # The FF_USE_POWERSHELL_PATH_RESOLVER feature flag has to be enabled for PowerShell
    # to resolve paths for Windows correctly when Runner is operating in a Linux environment
    # but targeting Windows nodes.
    environment = ["FF_USE_POWERSHELL_PATH_RESOLVER=true"]

    [runners.kubernetes.node_selector]
      "kubernetes.io/arch" = "amd64"
      "kubernetes.io/os" = "windows"
      "node.kubernetes.io/windows-build" = "10.0.20348"

Overwrite the node selector

To overwrite the node selector:

1. In the config.toml or Helm values.yaml file, enable overwriting of the node selector:

   runners:
    ...
    config: |
      [[runners]]
        [runners.kubernetes]
          node_selector_overwrite_allowed = ".*"
   

2. In the .gitlab-ci.yml file, define the variable to overwrite the node selector:

   variables:
     KUBERNETES_NODE_SELECTOR_* = ''
   

In the following example, to overwrite the Kubernetes node architecture, the settings are configured in the config.toml and .gitlab-ci.yml file:

concurrent = 1
check_interval = 1
log_level = "debug"
shutdown_timeout = 0

listen_address = ':9252'

[session_server]
  session_timeout = 1800

[[runners]]
  name = ""
  url = "https://gitlab.com/"
  id = 0
  token = "__REDACTED__"
  token_obtained_at = "0001-01-01T00:00:00Z"
  token_expires_at = "0001-01-01T00:00:00Z"
  executor = "kubernetes"
  shell = "bash"
  [runners.kubernetes]
    host = ""
    bearer_token_overwrite_allowed = false
    image = "alpine"
    namespace = ""
    namespace_overwrite_allowed = ""
    pod_labels_overwrite_allowed = ""
    service_account_overwrite_allowed = ""
    pod_annotations_overwrite_allowed = ""
    node_selector_overwrite_allowed = "kubernetes.io/arch=.*" # <--- allows overwrite of the architecture

job:
  image: IMAGE_NAME
  variables:
    KUBERNETES_NODE_SELECTOR_ARCH: 'kubernetes.io/arch=amd64'  # <--- select the architecture

Define a list of node affinities

Define a list of node affinities to add to a pod specification at build time.

concurrent = 1
[[runners]]
  name = "myRunner"
  url = "gitlab.example.com"
  executor = "kubernetes"
  [runners.kubernetes]
    [runners.kubernetes.affinity]
      [runners.kubernetes.affinity.node_affinity]
        [[runners.kubernetes.affinity.node_affinity.preferred_during_scheduling_ignored_during_execution]]
          weight = 100
          [runners.kubernetes.affinity.node_affinity.preferred_during_scheduling_ignored_during_execution.preference]
            [[runners.kubernetes.affinity.node_affinity.preferred_during_scheduling_ignored_during_execution.preference.match_expressions]]
              key = "cpu_speed"
              operator = "In"
              values = ["fast"]
            [[runners.kubernetes.affinity.node_affinity.preferred_during_scheduling_ignored_during_execution.preference.match_expressions]]
              key = "mem_speed"
              operator = "In"
              values = ["fast"]
        [[runners.kubernetes.affinity.node_affinity.preferred_during_scheduling_ignored_during_execution]]
          weight = 50
          [runners.kubernetes.affinity.node_affinity.preferred_during_scheduling_ignored_during_execution.preference]
            [[runners.kubernetes.affinity.node_affinity.preferred_during_scheduling_ignored_during_execution.preference.match_expressions]]
              key = "core_count"
              operator = "In"
              values = ["high", "32"]
            [[runners.kubernetes.affinity.node_affinity.preferred_during_scheduling_ignored_during_execution.preference.match_fields]]
              key = "cpu_type"
              operator = "In"
              values = ["arm64"]
      [runners.kubernetes.affinity.node_affinity.required_during_scheduling_ignored_during_execution]
        [[runners.kubernetes.affinity.node_affinity.required_during_scheduling_ignored_during_execution.node_selector_terms]]
          [[runners.kubernetes.affinity.node_affinity.required_during_scheduling_ignored_during_execution.node_selector_terms.match_expressions]]
            key = "kubernetes.io/e2e-az-name"
            operator = "In"
            values = [
              "e2e-az1",
              "e2e-az2"
            ]

Define nodes where pods are scheduled

Use pod affinity and anti-affinity to constrain the nodes your pod is eligible to be scheduled on, based on labels on other pods.

Example configuration in the config.toml:

concurrent = 1
[[runners]]
  name = "myRunner"
  url = "gitlab.example.com"
  executor = "kubernetes"
  [runners.kubernetes]
    [runners.kubernetes.affinity]
      [runners.kubernetes.affinity.pod_affinity]
        [[runners.kubernetes.affinity.pod_affinity.required_during_scheduling_ignored_during_execution]]
          topology_key = "failure-domain.beta.kubernetes.io/zone"
          namespaces = ["namespace_1", "namespace_2"]
          [runners.kubernetes.affinity.pod_affinity.required_during_scheduling_ignored_during_execution.label_selector]
            [[runners.kubernetes.affinity.pod_affinity.required_during_scheduling_ignored_during_execution.label_selector.match_expressions]]
              key = "security"
              operator = "In"
              values = ["S1"]
        [[runners.kubernetes.affinity.pod_affinity.preferred_during_scheduling_ignored_during_execution]]
        weight = 100
        [runners.kubernetes.affinity.pod_affinity.preferred_during_scheduling_ignored_during_execution.pod_affinity_term]
          topology_key = "failure-domain.beta.kubernetes.io/zone"
          [runners.kubernetes.affinity.pod_affinity.preferred_during_scheduling_ignored_during_execution.pod_affinity_term.label_selector]
            [[runners.kubernetes.affinity.pod_affinity.preferred_during_scheduling_ignored_during_execution.pod_affinity_term.label_selector.match_expressions]]
              key = "security_2"
              operator = "In"
              values = ["S2"]
      [runners.kubernetes.affinity.pod_anti_affinity]
        [[runners.kubernetes.affinity.pod_anti_affinity.required_during_scheduling_ignored_during_execution]]
          topology_key = "failure-domain.beta.kubernetes.io/zone"
          namespaces = ["namespace_1", "namespace_2"]
          [runners.kubernetes.affinity.pod_anti_affinity.required_during_scheduling_ignored_during_execution.label_selector]
            [[runners.kubernetes.affinity.pod_anti_affinity.required_during_scheduling_ignored_during_execution.label_selector.match_expressions]]
              key = "security"
              operator = "In"
              values = ["S1"]
          [runners.kubernetes.affinity.pod_anti_affinity.required_during_scheduling_ignored_during_execution.namespace_selector]
            [[runners.kubernetes.affinity.pod_anti_affinity.required_during_scheduling_ignored_during_execution.namespace_selector.match_expressions]]
              key = "security"
              operator = "In"
              values = ["S1"]
        [[runners.kubernetes.affinity.pod_anti_affinity.preferred_during_scheduling_ignored_during_execution]]
        weight = 100
        [runners.kubernetes.affinity.pod_anti_affinity.preferred_during_scheduling_ignored_during_execution.pod_affinity_term]
          topology_key = "failure-domain.beta.kubernetes.io/zone"
          [runners.kubernetes.affinity.pod_anti_affinity.preferred_during_scheduling_ignored_during_execution.pod_affinity_term.label_selector]
            [[runners.kubernetes.affinity.pod_anti_affinity.preferred_during_scheduling_ignored_during_execution.pod_affinity_term.label_selector.match_expressions]]
              key = "security_2"
              operator = "In"
              values = ["S2"]
          [runners.kubernetes.affinity.pod_anti_affinity.preferred_during_scheduling_ignored_during_execution.pod_affinity_term.namespace_selector]
            [[runners.kubernetes.affinity.pod_anti_affinity.preferred_during_scheduling_ignored_during_execution.pod_affinity_term.namespace_selector.match_expressions]]
              key = "security_2"
              operator = "In"
              values = ["S2"]

Networking

Configure a container lifecycle hook

Use container lifecycle hooks to run code configured for a handler when the corresponding lifecycle hook is executed.

You can configure two types of hooks: PreStop and PostStart. Each of them allows only one type of handler to be set.

Example configuration in the config.toml file:

[[runners]]
  name = "kubernetes"
  url = "https://gitlab.example.com/"
  executor = "kubernetes"
  token = "yrnZW46BrtBFqM7xDzE7dddd"
  [runners.kubernetes]
    image = "alpine:3.11"
    privileged = true
    namespace = "default"
    [runners.kubernetes.container_lifecycle.post_start.exec]
      command = ["touch", "/builds/postStart.txt"]
    [runners.kubernetes.container_lifecycle.pre_stop.http_get]
      port = 8080
      host = "localhost"
      path = "/test"
      [[runners.kubernetes.container_lifecycle.pre_stop.http_get.http_headers]]
        name = "header_name_1"
        value = "header_value_1"
      [[runners.kubernetes.container_lifecycle.pre_stop.http_get.http_headers]]
        name = "header_name_2"
        value = "header_value_2"

Use the following settings to configure each lifecycle hook:

KubernetesLifecycleExecAction

KubernetesLifecycleHTTPGet

KubernetesLifecycleHTTPGetHeader

KubernetesLifecycleTcpSocket

Configure pod DNS settings

Use the following options to configure the DNS settings of the pods.

Example configuration in the config.toml file:

concurrent = 1
check_interval = 30
[[runners]]
  name = "myRunner"
  url = "https://gitlab.example.com"
  token = "__REDACTED__"
  executor = "kubernetes"
  [runners.kubernetes]
    image = "alpine:latest"
    [runners.kubernetes.dns_config]
      nameservers = [
        "1.2.3.4",
      ]
      searches = [
        "ns1.svc.cluster-domain.example",
        "my.dns.search.suffix",
      ]

      [[runners.kubernetes.dns_config.options]]
        name = "ndots"
        value = "2"

      [[runners.kubernetes.dns_config.options]]
        name = "edns0"

KubernetesDNSConfigOption

Default list of dropped capabilities

GitLab Runner drops the following capabilities by default.

User-defined cap_add has priority over the default list of dropped capabilities. If you want to add the capability that is dropped by default, add it to cap_add.

• NET_RAW

Add extra host aliases

This feature is available in Kubernetes 1.7 and higher.

Configure a host aliases to instruct Kubernetes to add entries to /etc/hosts file in the container.

Use the following options:

Example configuration in the config.toml file:

concurrent = 4

[[runners]]
  # usual configuration
  executor = "kubernetes"
  [runners.kubernetes]
    [[runners.kubernetes.host_aliases]]
      ip = "127.0.0.1"
      hostnames = ["web1", "web2"]
    [[runners.kubernetes.host_aliases]]
      ip = "192.168.1.1"
      hostnames = ["web14", "web15"]

You can also configure host aliases by using the command-line parameter --kubernetes-host_aliases with JSON input. For example:

gitlab-runner register --kubernetes-host_aliases '[{"ip":"192.168.1.100","hostnames":["myservice.local"]},{"ip":"192.168.1.101","hostnames":["otherservice.local"]}]'

Volumes

Using the cache with the Kubernetes executor

When the cache is used with the Kubernetes executor, a volume called /cache is mounted on the pod. During job execution, if cached data is needed, the runner checks if cached data is available. Cached data is available if a compressed file is available on the cache volume.

To set the cache volume, use the cache_dir setting in the config.toml file.

• If available, the compressed file is extracted into the build folder and can then be used in the job.
• If not available, the cached data is downloaded from the configured storage and saved into the cache dir as a compressed file. The compressed file is then extracted into the build folder.

Configure volume types

You can mount the following volume types:

• hostPath
• persistentVolumeClaim
• configMap
• secret
• emptyDir
• csi

Example of a configuration with multiple volume types:

concurrent = 4

[[runners]]
  # usual configuration
  executor = "kubernetes"
  [runners.kubernetes]
    [[runners.kubernetes.volumes.host_path]]
      name = "hostpath-1"
      mount_path = "/path/to/mount/point"
      read_only = true
      host_path = "/path/on/host"
    [[runners.kubernetes.volumes.host_path]]
      name = "hostpath-2"
      mount_path = "/path/to/mount/point_2"
      read_only = true
    [[runners.kubernetes.volumes.pvc]]
      name = "pvc-1"
      mount_path = "/path/to/mount/point1"
    [[runners.kubernetes.volumes.config_map]]
      name = "config-map-1"
      mount_path = "/path/to/directory"
      [runners.kubernetes.volumes.config_map.items]
        "key_1" = "relative/path/to/key_1_file"
        "key_2" = "key_2"
    [[runners.kubernetes.volumes.secret]]
      name = "secrets"
      mount_path = "/path/to/directory1"
      read_only = true
      [runners.kubernetes.volumes.secret.items]
        "secret_1" = "relative/path/to/secret_1_file"
    [[runners.kubernetes.volumes.empty_dir]]
      name = "empty-dir"
      mount_path = "/path/to/empty_dir"
      medium = "Memory"
    [[runners.kubernetes.volumes.csi]]
      name = "csi-volume"
      mount_path = "/path/to/csi/volume"
      driver = "my-csi-driver"
      [runners.kubernetes.volumes.csi.volume_attributes]
        size = "2Gi"

hostPath volume

Configure the hostPath volume to instruct Kubernetes to mount a specified host path in the container.

Use the following options in the config.toml file:

persistentVolumeClaim volume

Configure the persistentVolumeClaim volume to instruct Kubernetes to use a persistentVolumeClaim defined in a Kubernetes cluster and mount it in the container.

Use the following options in the config.toml file:

configMap volume

Configure the configMap volume to instruct Kubernetes to use a configMap defined in a Kubernetes cluster and mount it in the container.

Use the following options in the config.toml:

Each key from the configMap is changed into a file and stored in the mount path. By default:

• All keys are included.
• The configMap key is used as the filename.
• The value is stored in the file contents.

To change the default key and value storage, use the items option . If you use the items option, only specified keys are added to the volumes and all other keys are skipped.

secret volume

Configure a secret volume to instruct Kubernetes to use a secret defined in a Kubernetes cluster and mount it in the container.

Use the following options in the config.toml file:

Each key from selected secret is changed into a file stored in the selected mount path. By default:

• All keys are included.
• The configMap key is used as the filename.
• The value is stored in the file contents.

To change default key and value storage, use the items option. If you use the items option, only specified keys are added to the volumes and all other keys are skipped.

emptyDir volume

Configure an emptyDir volume to instruct Kubernetes to mount an empty directory in the container.

Use the following options in the config.toml file:

csi volume

Configure a Container Storage Interface (csi) volume to instruct Kubernetes to use a custom csi driver to mount an arbitrary storage system in the container.

Use the following options in the config.toml:

Mount volumes on service containers

Volumes defined for the build container are also automatically mounted for all services containers. You can use this functionality as an alternative to services_tmpfs (available only to Docker executor), to mount database storage in RAM to speed up tests.

Example configuration in the config.toml file:

[[runners]]
  # usual configuration
  executor = "kubernetes"
  [runners.kubernetes]
    [[runners.kubernetes.volumes.empty_dir]]
      name = "mysql-tmpfs"
      mount_path = "/var/lib/mysql"
      medium = "Memory"

Custom volume mount

To store the builds directory for the job, define custom volume mounts to the configured builds_dir (/builds by default). If you use PVC volumes, based on the access mode, you might be limited to running jobs on one node.

Example configuration in the config.toml file:

concurrent = 4

[[runners]]
  # usual configuration
  executor = "kubernetes"
  builds_dir = "/builds"
  [runners.kubernetes]
    [[runners.kubernetes.volumes.empty_dir]]
      name = "repo"
      mount_path = "/builds"
      medium = "Memory"

Persistent per-concurrency build volumes

The build directories in Kubernetes CI jobs are ephemeral by default. If you want to persist your Git clone across jobs (to make GIT_STRATEGY=fetch work), you must mount a persistent volume claim for your build folder. Because multiple jobs can run concurrently, you must either use a ReadWriteMany volume, or have one volume for each potential concurrent job on the same runner. The latter is likely to be more performant. Here is an example of such a configuration:

concurrent = 4

[[runners]]
  executor = "kubernetes"
  builds_dir = "/mnt/builds"
  [runners.kubernetes]
    [[runners.kubernetes.volumes.pvc]]
      # CI_CONCURRENT_ID identifies parallel jobs of the same runner.
      name = "build-pvc-$CI_CONCURRENT_ID"
      mount_path = "/mnt/builds"

In this example, create the persistent volume claims named build-pvc-0 to build-pvc-3 yourself. Create as many as the runner’s concurrent setting dictates.

Use a helper image

After you set the security policy, the helper image must conform to the policy. The image does not receive privileges from the root group, so you must ensure that the user ID is part of the root group.

The following example creates a user and group called nonroot and sets the helper image to run as that user.

ARG tag
FROM registry.gitlab.com/gitlab-org/ci-cd/gitlab-runner-ubi-images/gitlab-runner-helper-ocp:${tag}
USER root
RUN groupadd -g 59417 nonroot && \
    useradd -u 59417 nonroot -g nonroot
WORKDIR /home/nonroot
USER 59417:59417

Using Docker in builds

When you use Docker in your builds, there are several considerations you should be aware of.

Exposed /var/run/docker.sock

There are certain risks if you use the runners.kubernetes.volumes.host_path option to expose the /var/run/docker.sock of your host into your build container. The node’s containers are accessible from the build container, and depending on if you are running builds in the same cluster as your production containers, it might not be wise to do that.

Using docker:dind

If you run the docker:dind, also called the docker-in-docker image, containers must run in privileged mode. This may have potential risks and cause additional issues.

The Docker daemon runs as a separate container in the pod because it is started as a service, typically in the .gitlab-ci.yml. Containers in pods only share volumes assigned to them and an IP address, that they use to communicate with each other with localhost. The docker:dind container does not share /var/run/docker.sock and the docker binary tries to use it by default.

To configure the client use TCP to contact the Docker daemon, in the other container, include the environment variables of the build container:

• DOCKER_HOST=tcp://<hostname>:2375 for no TLS connection.
• DOCKER_HOST=tcp://<hostname>:2376 for TLS connection.

For hostname set the value to:

• localhost for GitLab Runner 12.7 and earlier, and Kubernetes 1.6 and earlier.
• docker for GitLab Runner 12.8 and later, and Kubernetes 1.7 and later.

In Docker 19.03 and later, TLS is enabled by default but you must map certificates to your client. You can enable non-TLS connection for DIND or mount certificates. For more information, see Use Docker In Docker Workflow with Docker executor.

Prevent host kernel exposure

If you use docker:dind or /var/run/docker.sock, the Docker daemon has access to the underlying kernel of the host machine. This means that any limits set in the pod do not work when Docker images are built. The Docker daemon reports the full capacity of the node, regardless of limits imposed on the Docker build containers spawned by Kubernetes.

If you run build containers in privileged mode, or if /var/run/docker.sock is exposed, the host kernel may become exposed to build containers. To minimize exposure, specify a label in the node_selector option. This ensures that the node matches the labels before any containers can be deployed to the node. For example, if you specify the label role=ci, the build containers only run on nodes labeled role=ci, and all other production services run on other nodes.

To further separate build containers, you can use node taints. Taints prevent other pods from scheduling on the same nodes as the build pods, without extra configuration for the other pods.

Use kaniko to build Docker images

You can use kaniko to build Docker images inside a Kubernetes cluster.

Kaniko works without the Docker daemon and builds images without privileged access.

For more information, see Building images with kaniko and GitLab CI/CD.

There is a known issue when using kaniko to build multi-stage Dockerfiles. If a pipeline job includes an after_script section, when the after_script section is executed it fails with the following error message. The job still completes successfully.

OCI runtime exec failed: exec failed: container_linux.go:380: starting container process caused: chdir to cwd
("/workspace") set in config.json failed: no such file or directory: unknown

The section fails because kaniko deletes its container’s WORKDIR when building multi-stage Dockerfiles. This prevents kubectl exec (and the analogous SDK API) from attaching to the container.

Two workarounds exists for this issue:

• Add --ignore-path /workspace to the kaniko executor invocation.
• Add mkdir -p /workspace to the job’s script after the kaniko executor invocation.

For more information, see issue 30769.

Restrict Docker images and services

You can restrict the Docker images that are used to run your jobs. To do this, you specify wildcard patterns. For example, to allow images from your private Docker registry only:

[[runners]]
  (...)
  executor = "kubernetes"
  [runners.kubernetes]
    (...)
    allowed_images = ["my.registry.tld:5000/*:*"]
    allowed_services = ["my.registry.tld:5000/*:*"]

Or, to restrict to a specific list of images from this registry:

[[runners]]
  (...)
  executor = "kubernetes"
  [runners.kubernetes]
    (...)
    allowed_images = ["my.registry.tld:5000/ruby:*", "my.registry.tld:5000/node:*"]
    allowed_services = ["postgres:9.4", "postgres:latest"]

Restrict Docker pull policies

In the .gitlab-ci.yml file, you can specify a pull policy. This policy determines how a CI/CD job should fetch images.

To restrict which pull policies can be used in the .gitlab-ci.yml file, you can use allowed_pull_policies.

For example, to allow only the always and if-not-present pull policies:

[[runners]]
  (...)
  executor = "kubernetes"
  [runners.kubernetes]
    (...)
    allowed_pull_policies = ["always", "if-not-present"]

• If you don’t specify allowed_pull_policies, the default is the value in the pull_policy keyword.
• If you don’t specify pull_policy, the cluster’s image default pull policy is used.
• The existing pull_policy keyword must not include a pull policy that is not specified in allowed_pull_policies. If it does, the job returns an error.

Job execution

GitLab Runner uses kube attach instead of kube exec by default. This should avoid problems like when a job is marked successful midway in environments with an unstable network.

If you experience problems after updating to GitLab Runner 14.0, toggle the feature flag FF_USE_LEGACY_KUBERNETES_EXECUTION_STRATEGY to true and submit an issue.

Follow issue #27976 for progress on legacy execution strategy removal.

Configure the number of request attempts to the Kubernetes API

By default, the Kubernetes executor retries specific requests to the Kubernetes API after five failed attempts. The delay is controlled by a backoff algorithm with a 500 millisecond floor and a customizable ceiling with default value of two second. To configure the number of retries and backoff ceiling, use the retry_limit and retry_backoff_max options respectively in the config.toml file. The following failures are automatically retried:

• error dialing backend
• TLS handshake timeout
• read: connection timed out
• connect: connection timed out
• Timeout occurred
• http2: client connection lost
• connection refused
• tls: internal error
• io.unexpected EOF
• syscall.ECONNRESET
• syscall.ECONNREFUSED
• syscall.ECONNABORTED
• syscall.EPIPE

To control the amount of retries for each error, use the retry_limits option. The rety_limits specifies the amount of retries for each error separately, and is a map of error messages to the amount of retries. The error message can be a substring of the error message returned by the Kubernetes API. The retry_limits option has precedence over the retry_limit option.

For example, if you need to control the number of TLS related errors in your environment, you can configure the retry_limits option to retry those errors 10 times instead of the default of five times:

[[runners]]
  name = "myRunner"
  url = "https://gitlab.example.com/"
  executor = "kubernetes"
  [runners.kubernetes]
    retry_limit = 5

    [runners.kubernetes.retry_limits]
        "TLS handshake timeout" = 10
        "tls: internal error" = 10

To retry an entirely different error, such as exceeded quota 20 times:

[[runners]]
  name = "myRunner"
  url = "https://gitlab.example.com/"
  executor = "kubernetes"
  [runners.kubernetes]
    retry_limit = 5

    [runners.kubernetes.retry_limits]
        "exceeded quota" = 20

Container entrypoint known issues

The container entry point has the following known issues:

• If an entrypoint is defined in the Dockerfile for an image, it must open a valid shell. Otherwise, the job hangs.

  • To open a shell, the system passes the command as args for the build container.
• File type CI/CD variables are not written to disk when the entrypoint is executed. The file is only accessible in the job during script execution.
• The following CI/CD variables are not accessible in the entrypoint. You can use before_script to make any setup changes before running script commands:
  • CI/CD variables defined in the settings.
  • Masked CI/CD variables.

Before GitLab Runner 17.4:

• The entrypoint logs were not forwarded to the build’s log.
• With the Kubernetes executor with kube exec, GitLab Runner did not wait for the entrypoint to open a shell (see above).

Starting with GitLab Runner 17.4, the entrypoint logs are now forwarded. The system waits for the entrypoint to run and spawn the shell. This has the following implications:

• If FF_KUBERNETES_HONOR_ENTRYPOINT is set, and the image’s entrypoint takes longer than poll_timeout (default: 180s), the build fails. The poll_timeout value (and potentially poll_interval) must be adapted if the entrypoint is expected to run longer.
• When FF_KUBERNETES_HONOR_ENTRYPOINT and FF_USE_LEGACY_KUBERNETES_EXECUTION_STRATEGY are set, the system adds a startup probe to the build container, so that it knows when the entrypoint is spawning the shell. If a custom entrypoint uses the provided args to spawn the expected shell, then the startup probe is resolved automatically. However, if the container image is spawning the shell without using the command passed in through args, the entrypoint must resolve the startup probe itself by creating a file named .gitlab-startup-marker inside the root of the build directory. The startup probe checks every poll_interval for the .gitlab-startup-marker file. If the file is not present within poll_timeout, the pod is considered unhealthy, and the system abort the build.

Restrict access to job variables

When using Kubernetes executor, users with access to the Kubernetes cluster can read variables used in the job. By default, job variables are stored in:

• Pod’s environment section

To restrict access to job variable data, you should use role-based access control (RBAC) so that only GitLab administrators have access to the namespace used by the GitLab Runner.

If you need other users to access the GitLab Runner namespace, set the following verbs to restrict the type of access users have in the GitLab Runner namespace:

• For pods and configmaps:
  • get
  • watch
  • list
• For pods/exec and pods/attach, use create.

Example RBAC definition for authorized users:

kind: Role
apiVersion: rbac.authorization.k8s.io/v1
metadata:
  name: gitlab-runner-authorized-users
rules:
- apiGroups: [""]
  resources: ["configmaps", "pods"]
  verbs: ["get", "watch", "list"]
- apiGroups: [""]
  resources: ["pods/exec", "pods/attach"]
  verbs: ["create"]

Resources check during prepare step

Prerequisites:

• image_pull_secrets or service_account is set.
• resource_availability_check_max_attempts is set to a number greater than zero.
• Kubernetes serviceAccount used with the get and list permissions.

GitLab Runner checks if the new service accounts or secrets are available with a 5-second interval between each try.

• In GitLab 15.0 and 15.1, you cannot disable this feature and it defaults to 5 when a negative value is set.
• In GitLab 15.0.1, 15.1.1, 15.2 and later, this feature is disabled by default. To enable this feature, set resource_availability_check_max_attempts to any value other than 0. The value you set defines the amount of times the runner checks for service accounts or secrets.

Overwrite the Kubernetes namespace

Prerequisites:

• In the values.yml file for GitLab Runner Helm charts, rbac.clusterWideAccess is set to true.
• The runner has permissions configured in the core API group.

You can overwrite Kubernetes namespaces to designate a namespace for CI purposes, and deploy a custom set of pods to it. The pods spawned by the runner are in the overwritten namespace to enable access between containers during the CI stages.

To overwrite the Kubernetes namespace for each CI/CD job, set the KUBERNETES_NAMESPACE_OVERWRITE variable in the .gitlab-ci.yml file.

variables:
  KUBERNETES_NAMESPACE_OVERWRITE: ci-${CI_COMMIT_REF_SLUG}

To use only designated namespaces during CI runs, in the config.toml file, define a regular expression for namespace_overwrite_allowed:

[runners.kubernetes]
    ...
    namespace_overwrite_allowed = "ci-.*"

Fault tolerance in Kubernetes clusters

The fault tolerance feature for GitLab Runner helps prevent orphaned resources (such as Kubernetes pods) and jobs stuck in a Running state when a runner manager restarts.

The following steps describe how job execution works under typical conditions. The runner manager:

1. Receives a job from GitLab.
2. Creates the job pod in Kubernetes.
3. Sends commands to be executed in the job pods.
4. Reads the job logs and status, and sends the information back to GitLab. Then, GitLab refreshes the UI.

On Kubernetes, when the node that hosts the runner manager pod is removed when orchestrating CI/CD jobs on the cluster, the runner pods are orphaned. They continue to run until removed by any other mechanism. A new runner manager can’t access the active runner pods and related CI/CD jobs. As a result, the orphaned runner pods continue to consume resources on the cluster. The GitLab UI doesn’t update the job status until the job eventually fails due to the configured timeout.

With the fault tolerance feature, GitLab Runner resumes running jobs from where they stopped.

GitLab Runner saves the Job and the Execution State of each job in the configured store. Before requesting a new job, GitLab checks the store for jobs that are already running.

A fault-tolerant runner manager:

1. Creates the configured store if the executor is a Stateful Executor.

   • If the store is not configured, but the executor is Stateful, a no-op (no operation) store is used. Thanks to no-op stores, you can save on data calls without additional logic branching.

     store := NewNoopStore()
     if isStatefulExecutor := executor.(StatefulExecutor); isStatefulExecutor {
      store, _ = provider.GetStore(runnerConfig)
     }
     

2. Executes jobs as usual, but periodically or on certain events, writes them to the store.

3. Checks for running jobs in the store during restart.

   If a job hasn’t been updated recently, the runner:

   1. Resumes the job.
   2. Disables all trace logging to prevent log duplication, as the runner generates numerous job logs during job execution startup.

   3. Restores the State Metadata on the Kubernetes executor first.

      type executorStateMetadata struct {
        // The secrets created by the executor
       Credentials *api.Secret   `json:"credentials"`
       // The build pod
       Pod         *api.Pod      `json:"pod"`
       // The services
       Services    []api.Service `json:"services"`
       // Offset is the offset off of which to read the job logs from the build pod
       Offset      int64         `json:"offset"`
      }
      
      // Only the name and namespace of these resources is serialized to the store
      /*
      {
       credentials: { name: "", namespace: "" }
       pod: { name: "", namespace: "" }
       services: [{ name: "", namespace: "" }]
       offset: 0
      }
      */
      

   4. Calls the Prepare method on the Executor to set it up for job processing.
   5. Re-enables and resumes trace logging.

   6. Calls the Restore method which is part of the StatefulExecutor interface. This method queries the Kubernetes cluster for full API objects:

      func (s *executor) Restore(ctx context.Context) error {
       s.state.Pod, _ = s.kubeClient.CoreV1().Pods(s.state.Pod.Namespace).Get(s.state.Pod.Name)
       s.state.Credentials, _ = s.kubeClient.CoreV1().Secrets(s.state.Credentials.Namespace).Get(s.state.Credentials.Name)
       // do the same for services
      
       return nil
      }
      

   7. Calls the Resume method on the StatefulExecutor interface instead of the usual Run method. The Resume method re-attaches to the Pod and continues sending the job trace and status to GitLab.

After a restart, the new job execution mechanism functions as follows: