You are browsing documentation for an older version. See the latest documentation here.

Distributed Configuration for Kong using decK

decK can operate on a subset of configuration instead of taking care of managing the entire configuration of Kong.

This can be very useful in a variety of scenarios. For example:

• Your organization uses a single Kong installation whose configuration is built from multiple teams. For example, one team manages the inventory API, while another team manages the users API and both APIs are exposed via Kong. In such a case, you want team inventory to manage it’s configuration without worrying about the configuration of the other team. The two teams should be able to push out configuration changes as they see fit and when they see fit.
• Kong’s configuration is very large, and you would like to manage it using different files, where you different parts take care of different problems.
• You want to declaratively manage all of the configuration for Kong except consumers and their credentials. The consumers are being managed by another service or you are dynamically creating them or the number of consumer is just so large that it makes no sense to manage them in a declarative fashion.

In such cases, you can use decK’s select-tag feature to export, sync, reset only a sub-set of configuration.

Tags

Tags, introduced in Kong 1.1, provide a way to associate metadata with entities in Kong. You can also filter entities by tags on the list endpoints in Kong.

Using this feature, decK associates tags with entities and can manage a group of entities which share a common tag(s).

When multiple tags are specified in decK, decK ANDs those tags together, meaning only entities containing all the tags will be managed by decK. You can specify a combination of up to 5 tags, but it is recommended to use fewer or only one tag, for performance reasons in Kong’s codebase.

Dump

You can export a subset of entities in Kong by specifying common tags using the --select-tag flag.

For example, the following command generates a kong.yaml file with all entities which have both of the specified tags:

deck dump --select-tag foo-tag --select-tag bar-tag

If you observe the file generated by decK, you will see the following section:

_info:
  select_tags:
  - foo-tag
  - bar-tag

This sub-section tells decK to filter out entities containing select-tags during a sync operation.

Sync

You don’t need to specify --select-tag in sync and diff commands. The commands will use the tags present in the state file and perform the diff accordingly.

Since the state files contain the tagging information, different teams can make updates to the part of configuration in Kong without worrying about configuration of other teams. You no longer need to maintain Kong’s configuration in a single repository, where multiple teams need to co-ordinate.

The --select-tag flag is present on those two commands for use cases where the file cannot have select_tags defined inside it. It is strongly advised that you do not supply select-tags to sync and diff commands via flags. This is because the tag information should be part of the declarative configuration file itself in order to provide a practical declarative file. The tagging information and entity definitions should be present in one place, else an error in supplying the wrong tag via the CLI can break the configuration.

Reset

You can delete only a subset of entities sharing a tag using the --select-tag flag on the reset command.

Initial setup problem

When you initially get started with a distributed configuration management, you will likely run into a problem where the related entities you would like to manage don’t share a single database.

To get around this problem, you can use one of the following approaches:

• Go through each entity in Kong, and patch those entities with the common tag(s) you’d like, then use decK’s dump command to export by different tags.
• Export the entire configuration of Kong, and divide up the configuration into different files. Then, add the select_tags info to the file. This will require re-creation of the database now, since decK will not detect any of the entities present (as they are missing the common tag).