You are browsing documentation for an outdated version. See the latest documentation here.
Kubernetes Deployment Options
The Kubernetes Ingress Controller translates Kubernetes resources into Kong Gateway configuration. Kong Gateway uses that configuration to route and control traffic.
The kong-gateway proxy image supports DB-less operation and is recommended for all deployments.
- DB-less installation with the Kong Ingress Controller
- Database-backed installation with or without the Kong Ingress Controller
Migrating to 2.1.x and up
Existing users of the
kong-enterprise-k8s image who want to use 2.1.x or later
should switch to the
If you encounter issues after switching images, please contact Enterprise Support.
DB-less versus database-backed deployments
When using Kong Gateway with Kubernetes Ingress Controller, the source of truth for Kong’s configuration is the Kubernetes configuration in etcd: Kong’s custom Kubernetes resources, ingresses, and services provide the information necessary for the ingress controller to configure Kong. This differs from Kong deployments that do not use an ingress controller, where configuration in the database or DB-less config file is the source of truth.
In traditional deployments, Kong’s database provides a persistent store of configuration available to all Kong nodes to ensure consistent proxy behavior across the cluster that is not affected by node restarts. Because etcd provides this functionality in Kong for Kubernetes deployments, it is not necessary to run an additional database, reducing maintenance and infrastructure requirements.
While running Kong Gateway with Kubernetes Ingress Controller does not require a database, it is fully compatible with PostgreSQL and requires it for some features. etcd still remains the source of truth in database-backed deployments: the controller translates Kubernetes resources from etcd into Kong configuration and inserts them into the database via the Admin API.
Choosing between DB-less or database-backed deployments
In general, DB-less deployments are simpler to maintain and require less
resources to run.
These deployments must set
KONG_DATABASE=off in their environment variables.
Database-backed deployments offer a wider range of features. Review the sections below to determine if your use case requires a feature that is not available in DB-less deployments.
Some Enterprise features are not available in DB-less deployments. Use a database-backed deployment if you want to use:
- Dev Portal
- Teams (RBAC)
The following features have support in DB-less mode, but work differently than in DB-backed modes:
- Kong Manager (read-only)
- Vitals (using Prometheus or InfluxDB strategies)
When Kong Gateway is configured by the ingress controller, some functionality in these features is different from traditional deployments:
- Instead of using Kong Manager, proxy configuration is managed by the controller, and you provide configuration via Kubernetes resources.
- Because the controller creates proxy configuration on behalf of users, you do not need to interact with the Admin API directly. Kong’s own RBAC implementation isn’t required for typical Kong for Kubernetes deployments, as they do not expose the Admin API; only the controller can access it. Kubernetes-level RBAC rules and namespaces should be used to restrict what configuration administrators can create.
- Ingress controller instances create configuration in a single workspace only
defaultby default). To use multiple workspaces, deploy multiple controller instances, setting the
CONTROLLER_KONG_WORKSPACEenvironment variable to the workspace that instance should use. These instances should set
CONTROLLER_INGRESS_CLASSto unique values for each instance to avoid creating duplicate configuration in workspaces. Note that if controller instances are deployed outside the Kong pod the Admin API must be exposed, and users should enable RBAC with workspace admin users for the controllers. Set
CONTROLLER_KONG_ADMIN_TOKENto the RBAC user’s token.
- The controller cannot manage configuration for features that require a database: it cannot create workspaces, Dev Portal content, admins, etc. These features must be configured manually through the Admin API or Kong Manager.
Not all plugins are compatible with DB-less operation. Review the list of supported plugins to see if you require a plugin that needs a database.
Third-party plugins are generally compatible with DB-less as long as they do not create custom entities (i.e. they do not add new entities that users can create and modify through the Admin API).
DB-less configuration must be supplied as a complete unit: it is not possible to add or modify entities individually through the Admin API, or provide partial configuration that is added to existing configuration. As such, all configuration must be sourced from Kubernetes resources so that the ingress controller can render it into a complete configuration.
On database-backed deployments, users can create or modify configuration
through the Admin API. The ingress controller uses a tag (set by the
CONTROLLER_KONG_ADMIN_FILTER_TAG environment variable) to identify
configuration that it manages. While the controller will revert changes to
configuration with its tag, other configuration is left as-is.
Although database-backed deployments can use controller-generated and manually-added configuration simultaneously, Kong’s recommended best practice is to manage as much configuration through Kubernetes resources as possible. Using both controller-managed and manual configuration can result in conflicts between the two, and conflicts will prevent the controller from applying its configuration. To minimize this risk:
- Use the admission webhook to reject Kubernetes resources that conflict with other configuration, or are otherwise invalid.
- Manually create configuration in a workspace that is not managed by the controller. This avoids most conflicts, but not all: routes may still conflict depending on your route validation setting.
Large numbers of consumers (and associated credentials) are the exception to this rule: if your consumer count is in the tens of thousands, we recommend that you create them and their credentials through the Admin API to reduce etcd load.
Migrating between deployment types
Because etcd is the source of truth for Kong’s configuration, the ingress controller can re-create Kong’s proxy configuration even if the underlying datastore changes.
While most Kubernetes resources can be left unchanged when migrating between deployment types, users must remove any KongPlugin resources that use unavailable plugins when migrating from a database-backed deployment to a DB-less deployment. No changes to Kubernetes resources are required if migrating in the opposite direction.