Connecting a Kubernetes cluster with GitLab

Tier: Free, Premium, Ultimate
Offering: GitLab.com, GitLab Self-Managed, GitLab Dedicated

Flux recommended as GitOps solution in GitLab 15.10.

You can connect your Kubernetes cluster with GitLab to deploy, manage, and monitor your cloud-native solutions.

To connect a Kubernetes cluster to GitLab, you must first install an agent in your cluster.

The agent runs in the cluster, and you can use it to:

Communicate with a cluster, which is behind a firewall or NAT.
Access API endpoints in a cluster in real time.
Push information about events happening in the cluster.
Enable a cache of Kubernetes objects, which are kept up-to-date with very low latency.

For more details about the agent's purpose and architecture, see the architecture documentation.

You must deploy a separate agent to every cluster you want to connect to GitLab. The agent was designed with strong multi-tenancy support. To simplify maintenance and operations you should run only one agent per cluster.

An agent is always registered in a GitLab project. After an agent is registered and installed, the agent connection to the cluster can be shared with other projects, groups, and users. This approach means you can manage and configure your agent instances from GitLab itself, and you can scale a single installation to multiple tenants.

Supported Kubernetes versions for GitLab features

GitLab supports the following Kubernetes versions. If you want to run GitLab in a Kubernetes cluster, you might need a different version of Kubernetes:

For the Helm Chart.
For GitLab Operator.

You can upgrade your Kubernetes version to a supported version at any time:

1.35 (support ends when GitLab version 19.10 is released or when 1.38 becomes supported)
1.34 (support ends when GitLab version 19.7 is released or when 1.37 becomes supported)
1.33 (support ends when GitLab version 19.2 is released or when 1.36 becomes supported)

GitLab aims to support a new minor Kubernetes version three months after its initial release. GitLab supports at least three production-ready Kubernetes minor versions at any given time.

When a new version of Kubernetes is released:

This page updates with the results of early smoke tests within approximately four weeks.
If the release of a new version support is delayed, this page updates with the expected GitLab support version within approximately eight weeks.

When installing the agent, use a Helm version compatible with your Kubernetes version. Other versions of Helm might not work. For a list of compatible versions, see the Helm version support policy.

Support for deprecated APIs can be removed from the GitLab codebase when GitLab no longer supports the Kubernetes version that only supports the deprecated API.

Some GitLab features might work on versions not listed here. This epic tracks support for Kubernetes versions.

Kubernetes deployment workflows

You can choose from two primary workflows. The GitOps workflow is recommended.

GitOps workflow

GitLab recommends using Flux for GitOps. To get started, see Tutorial: Set up Flux for GitOps.

GitLab CI/CD workflow

In a CI/CD workflow, you configure GitLab CI/CD to use the Kubernetes API to query and update your cluster.

This workflow is considered push-based, because GitLab pushes requests from GitLab CI/CD to your cluster.

Use this workflow:

When you have pipeline-driven processes.
When you need to migrate to the agent, but the GitOps workflow doesn't support your use case.

This workflow has a weaker security model. You should not use a CI/CD workflow for production deployments.

Agent connection technical details

The agent opens a bidirectional channel to KAS for communication. This channel is used for all communication between the agent and KAS:

Each agent can maintain up to 500 logical gRPC streams, including active and idle streams.
The number of TCP connections used by the gRPC streams is determined by gRPC itself.
Each connection has a maximum lifetime of two hours, with a one-hour grace period.
- A proxy in front of KAS might influence the maximum lifetime of connections. On GitLab.com, this is two hours. The grace period is 50% of the maximum lifetime.

For detailed information about channel routing, see Routing KAS requests in the agent.

Receptive agents

Tier: Ultimate
Offering: GitLab Self-Managed

Introduced in GitLab 17.4.

Receptive agents allow GitLab to integrate with Kubernetes clusters that cannot establish a network connection to the GitLab instance, but can be connected to by GitLab. For example, this can occur when:

GitLab runs in a private network or behind a firewall, and is only accessible through VPN.
The Kubernetes cluster is hosted by a cloud provider, but is exposed to the internet or is reachable from the private network.

When this feature is enabled, GitLab connects to the agent with the provided URL. You can use agents and receptive agents simultaneously.

Kubernetes integration glossary

This glossary provides definitions for terms related to the GitLab Kubernetes integration.

Term	Definition	Scope
GitLab agent for Kubernetes	The overall offering, including related features and the underlying components `agentk` and `kas`.	GitLab, Kubernetes, Flux
`agentk`	The cluster-side component that maintains a secure connection to GitLab for Kubernetes management and deployment automation.	GitLab
GitLab agent server for Kubernetes (`kas`)	The GitLab-side component of GitLab that handles operations and logic for the Kubernetes agent integration. Manages the connection and communication between GitLab and Kubernetes clusters.	GitLab
Pull-based deployment	A deployment method where Flux checks for changes in a Git repository and automatically applies these changes to the cluster.	GitLab, Kubernetes
Push-based deployment	A deployment method where updates are sent from GitLab CI/CD pipelines to the Kubernetes cluster.	GitLab
Flux	An open-source GitOps tool that integrates with the agent for pull-based deployments.	GitOps, Kubernetes
GitOps	A set of practices that involve using Git for version control and collaboration in the management and automation of cloud and Kubernetes resources.	DevOps, Kubernetes
Kubernetes namespace	A logical partition in a Kubernetes cluster that divides cluster resources between multiple users or environments.	Kubernetes