DB-less and Declarative Configuration

Traditionally, Kong Gateway has always required a database, to store its configured entities such as routes, services and plugins. Kong uses its configuration file, kong.conf, to specify the use of the database and its various settings.

Kong Gateway can be run without a database using only in-memory storage for entities. We call this DB-less mode. When running Kong Gateway DB-less, the configuration of entities is done in a second configuration file, in YAML or JSON, using declarative configuration.

  2. flowchart TD
  4. A( Kong Gateway instance)
  5. B( Kong Gateway instance)
  6. C( Kong Gateway instance)
  8. A2(fa:fa-file kong1.yml)
  9. B2(fa:fa-file kong2.yml)
  10. C2(fa:fa-file kong3.yml)
  12. A2 --> A
  13. B2 --> B
  14. C2 --> C
  16. style A stroke:none,fill:#0E44A2,color:#fff
  17. style B stroke:none,fill:#0E44A2,color:#fff
  18. style C stroke:none,fill:#0E44A2,color:#fff

Figure 1: In DB-less mode, configuration is applied via YAML files. Kong Gateway nodes aren’t connected to a database, or to each other.

The combination of DB-less mode and declarative configuration has a number of benefits:

  • Reduced number of dependencies: No need to manage a database installation if the entire setup for your use-cases fits in memory.
  • Automation in CI/CD scenarios: Configuration for entities can be kept in a single source of truth managed via a Git repository.
  • Enables more deployment options for Kong Gateway.

Note: decK also manages configuration declaratively, but it requires a database to perform any of its sync, dump, or similar operations. Therefore, decK can’t be used in DB-less mode.

Declarative configuration

The key idea in declarative configuration is the notion that it is declarative, as opposed to an imperative style of configuration. “Imperative” means that a configuration is given as a series of orders: “do this, then do that”. “Declarative” means that the configuration is given all at once: “I declare this to be the state of the world”.

The Kong Admin API is an example of an imperative configuration tool. The final state of the configuration is attained through a sequence of API calls: one call to create a Service, another call to create a Route, another call to add a Plugin, and so on.

Performing the configuration incrementally like this has the undesirable side-effect that intermediate states happen. In the above example, there is a window of time in between creating a Route and adding the Plugin in which the Route did not have the Plugin applied.

A declarative configuration file, on the other hand, contains the settings for all desired entities in a single file. Once that file is loaded into Kong Gateway, it replaces the entire configuration. When incremental changes are desired, they are made to the declarative configuration file, which is then reloaded in its entirety. At all times, the configuration described in the file loaded into Kong is the configured state of the system.

Set up Kong in DB-less mode

To use Kong Gateway in DB-less mode, set the database directive of kong.conf to off. As usual, you can do this by editing kong.conf and setting database=off or via environment variables. You can then start Kong as usual:

  1. export KONG_DATABASE=off
  2. kong start -c kong.conf

Once Kong starts, access the / endpoint of the Admin API to verify that it is running without a database. It will return the entire Kong configuration; verify that database is set to off.

Command:

  1. curl -i -X GET http://localhost:8001

Sample response:

  1. HTTP/1.1 200 OK
  2. Access-Control-Allow-Origin: *
  3. Connection: keep-alive
  4. Content-Length: 6342
  5. Content-Type: application/json; charset=utf-8
  6. Date: Wed, 27 Mar 2019 15:24:58 GMT
  7. Server: kong/3.8.0
  8. {
  9.     "configuration:" {
  10.        ...
  11.        "database": "off",
  12.        ...
  13.     },
  14.     ...
  15.     "version": "3.8.0"
  16. }

Kong Gateway is running, but no declarative configuration has been loaded yet. This means that the configuration of this node is empty. There are no routes, services, or entities of any kind.

Command:

  1. curl -i -X GET http://localhost:8001/routes

Sample response:

  1. HTTP/1.1 200 OK
  2. Access-Control-Allow-Origin: *
  3. Connection: keep-alive
  4. Content-Length: 23
  5. Content-Type: application/json; charset=utf-8
  6. Date: Wed, 27 Mar 2019 15:30:02 GMT
  7. Server: kong/3.8.0
  9. {
  10.     "data": [],
  11.     "next": null
  12. }

Creating a declarative configuration file

To load entities into DB-less Kong, you need a declarative configuration file. The following command creates a skeleton file to get you started:

  1. kong config -c kong.conf init

This command creates a kong.yml file in the current directory, containing examples of the syntax for declaring entities and their relationships. All examples in the generated file are commented-out by default. You can experiment by uncommenting the examples (removing the # markers) and modifying their values.

Declarative configuration format

The Kong Gateway declarative configuration format consists of lists of entities and their attributes. This is a small yet complete example that illustrates a number of features:

  1. _format_version: "3.0"
  2. _transform: true
  4. services:
  5. - name: my-service
  6.   url: https://example.com
  7.   plugins:
  8.   - name: key-auth
  9.   routes:
  10.   - name: my-route
  11.     paths:
  12.     - /
  14. consumers:
  15. - username: my-user
  16.   keyauth_credentials:
  17.   - key: my-key

See the declarative configuration schema for all configuration options.

The only mandatory piece of metadata is _format_version: "3.0", which specifies the version number of the declarative configuration syntax format. This also matches the minimum version of Kong required to parse the file.

The _transform metadata is an optional boolean (defaults to true), which controls whether schema transformations will occur while importing. The rule of thumb for using this field is: if you are importing plain-text credentials (i.e. passwords), you likely want to set it to true, so that Kong will encrypt/hash them before storing them in the database. If you are importing already hashed/encrypted credentials, set _transform to false so that the encryption does not happen twice.

At the top level, you can specify any Kong entity, be it a core entity such as services and consumers as in the above example, or custom entities created by plugins, such as keyauth_credentials. This makes the declarative configuration format inherently extensible, and it is the reason why kong config commands that process declarative configuration require kong.conf to be available, so that the plugins directive is taken into account.

When entities have a relationship, such as a route that points to a service, this relationship can be specified via nesting.

Only one-to-one relationships can be specified by nesting: a plugin that is applied to a service can have its relationship depicted via nesting, as in the example above. Relationships involving more than two entities, such as a plugin that is applied to both a service and a consumer, must be done via a top-level entry. In that case, the entities can be identified by their primary keys or identifying names (the same identifiers that can be used to refer to them in the Admin API). This is an example of a plugin applied to a service and a consumer:

  1. plugins:
  2. - name: syslog
  3.   consumer: my-user
  4.   service: my-service

Check the file

Once you are done editing the file, you can check the syntax for any errors before attempting to load it into Kong Gateway:

  1. kong config -c kong.conf parse kong.yml

Response:

  1. parse successful

Load the file

There are two ways to load a declarative configuration file into Kong: using kong.conf or the Admin API.

To load a declarative configuration file at Kong start-up, use the declarative_config directive in kong.conf (or, as usual to all kong.conf entries, the equivalent KONG_DECLARATIVE_CONFIG environment variable):

  1. export KONG_DATABASE=off \
  2. export KONG_DECLARATIVE_CONFIG=kong.yml \
  3. kong start -c kong.conf

You can also load a declarative configuration file into a running Kong node with the Admin API, using the /config endpoint. The following example loads kong.yml:

  1. curl -i -X POST http://localhost:8001/config \
  2.   --form config=@kong.yml

The /config endpoint replaces the entire set of entities in memory with the ones specified in the given file.

Another way you can start Kong in DB-less mode is by including the entire declarative configuration in a string using the KONG_DECLARATIVE_CONFIG_STRING environment variable:

  1. export KONG_DATABASE=off
  2. export KONG_DECLARATIVE_CONFIG_STRING='{"_format_version":"1.1", "services":[{"host":"httpbin.org","port":443,"protocol":"https", "routes":[{"paths":["/"]}]}],"plugins":[{"name":"rate-limiting", "config":{"policy":"local","limit_by":"ip","minute":3}}]}'
  3. kong start

DB-less with Helm (KIC disabled)

When deploying Kong Gateway on Kubernetes in DB-less mode (env.database: "off") and without the Ingress Controller (ingressController.enabled: false), you have to provide a declarative configuration for Kong Gateway to run. You can provide an existing ConfigMap (dblessConfig.configMap) or place the whole configuration into a values.yaml (dblessConfig.config) parameter. See the example configuration in the default values.yaml for more detail.

Use --set-file dblessConfig.config=/path/to/declarative-config.yaml in Helm commands to substitute in a complete declarative config file.

Externally supplied ConfigMaps are not hashed or tracked in deployment annotations. Subsequent ConfigMap updates require user-initiated deployment rollouts to apply the new configuration. Run kubectl rollout restart deploy after updating externally supplied ConfigMap content.

Using Kong in DB-less mode

There are a number of things to be aware of when using Kong in DB-less mode.

Memory cache requirements

The entire configuration of entities must fit inside the Kong cache. Make sure that the in-memory cache is configured appropriately: see the mem_cache_size directive in kong.conf.

No central database coordination

Since there is no central database, multiple Kong nodes have no central coordination point and no cluster propagation of data: nodes are completely independent of each other.

This means that the declarative configuration should be loaded into each node independently. Using the /config endpoint does not affect other Kong nodes, since they have no knowledge of each other.

Read-only Admin API

Since the only way to configure entities is via declarative configuration, the endpoints for CRUD operations on entities are effectively read-only in the Admin API when running Kong in DB-less mode. GET operations for inspecting entities work as usual, but attempts to POST, PATCH PUT or DELETE in endpoints such as /services or /plugins will return HTTP 405 Not Allowed.

This restriction is limited to what would be otherwise database operations. In particular, using POST to set the health state of targets is still enabled, since this is a node-specific in-memory operation.

Kong Manager compatibility

Kong Manager cannot guarantee compatibility with Kong Gateway operating in DB-less mode. You cannot create, update, or delete entities with Kong Manager when Kong Gateway is running in this mode. Entity counters in the “Summary” section on the global and workspace overview pages will not function correctly as well.

Plugin compatibility

Not all Kong plugins are compatible with DB-less mode. By design, some plugins require central database coordination or dynamic creation of entities.

For current plugin compatibility, see Plugin compatibility.

See also