Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

This proposal intends to enhance the AbstractConfig base class to automatically resolve variables of the form specified in KIP-297. This simple change will make it very easy for all the components client applications, Kafka Connect, and Kafka Streams to use shared code to easily incorporate externalized secrets and other variable replacements within their configurations. This KIP will enable all the components broker, connect, producer, consumer, admin client, and so forth.  to automatically resolve the external configurations. This proposal does not add other ConfigProvider implementations or change the behavior of existing methods. By default this feature will be enabled only for the Connect component and other components can enable the feature as needed.

Public Interfaces

This proposal will add 2 new interfaces for the AbstractConfig class:
AbstractConfig

...

The new constructor is similar to an existing constructor, with a flag enableExternalConfigResolution to enable/disable automatic resolution of indirect variable. The originals configurations will contain both the config provider configs as well as configuration properties. The constructor will first instantiate the ConfigProviders using the config provider configs, then it will find all variables in the values of the originals configurations, attempt to resolve the variables using the named ConfigProviders, and then do the normal parsing and validation of the configurations. If ConfigProvider is not provided in the originals, the constructor will skip the variable substitution step and will simply validate and parse the supplied configuration. With this approach both the configs and config providers can be defined in the same config property file.  We will deprecate the old constructor. By default the enableExternalConfigResolution will be set to false.

Variables are defined by KIP-297 and have the form "${providerName:[path:]key}", where "providerName" is the name of a ConfigProvider, "path" is an optional string, and "key" is a required string. Per KIP-297, this variable is resolved by passing the "key" and optional "path" to a ConfigProvider with the specified name, and the result from the ConfigProvider is then used in place of the variable. Variables that cannot be resolved by the AbstractConfig constructor will be left unchanged in the configuration.

In order The second constructor provides a way to support components which do not have configProviders defined in the config file but read it from a separate config provider file, we add another constructor with .  This constructor takes an additional field to specify the config providers Map<?, ?> configProviders. This constructor will first instantiate the ConfigProviders using the map of config provider and then resolve the variable as previous constructor. Any providers in originals map will be ignored.  By default the enableExternalConfigResolution will be set to false.

Configuration Changes

This KIP proposes to reserve all config names starting with "config.providers" to resolve the config values. The config "config.providers"  will be reserved to list the ConfigProviders, "config.providers.providerName<providerName>.class" will be reserved to specify the class to be used to instantiate the providers and  "config.providers.providerName<providerName>.variableName<variableName>to specify any additional configs required by providers where <providerName> is the name of the new config provider and <variableName> variables required by the config provider. The new constructor will look for "config.providers" in the originals fields and instantiate them using the specified configProvider class. It will then use these instantiated config providers to resolve any indirect values.   

Proposed Changes 

Steps to enable the automatic resolution of externalized config feature

This KIP will enable all the components broker, connect, producer, consumer, admin client, and so forth.  to automatically resolve the external configurations. As this feature is built upon KIP-297, each component will have to make the below changes in order to enable this feature


  1. Identify the configs which need to be stored in external systems.
  2. Implement a ConfigProvider which will fetch the config value from external system
  3. Replace the configs values with variables as defined by KIP-297
  4. Enable automatic config resolution by setting the  enableExternalConfigResolution to true in the subclass of AbstractConfig.

As Connect component already supports variables in connector configs via KIP 297 it will use this feature to resolve variables in the worker configurations via the existing 'StandaloneConfig' and 'DistributedConfig' subclasses of AbstractConfig.

Dynamic configurations for Brokers

The indirect dynamic configurations in the brokers should be updated using the AdminClient and kafka.sh script as described by KIP-226. The dynamic config updates need to happen atomically in order to achieve this the user will have to first update the external configs after all the external configs are updated, he will send a AdminClient request to the broker to update those config from config provider. Upon receiving the alterConfig request, the broker will refetch the updated configs from the ConfigProvider. The actual resolution of indirect values will be done by the KafkaConfig which extends from AbstractConfig, it will invoke the ConfigProvider to provide the actual values for the updated configs.

...

Consider the dynamic configurations ssl.keystore.location, ssl.keystore.password and ssl.keystore.type are stored in vault and their values are resolved using a VaultConfigProvider. 

...
ssl.keystore.location=${vault:/path/to/variables.properties:ssl.keystore.location}

ssl.keystore.password=${vault:/path/to/variables.properties:ssl.keystore.password}

ssl.keystore.type=${vault:/path/to/variables.properties:ssl.keystore.type}
 
config.providers=vault
config.providers.file.class=org.apache.kafka.connect.configs.VaultConfigProvider

  • The user will update the ssl.keystore.location, ssl.keystore.password, ssl.keystore.type in the vault.
  • After the updates are complete he will send a adminClient request to the broker to notify the configs are updated.
  • Broker receives a alteredConfig request.
  • The broker will invoke the get function in VaultConfigProvider
  • The VaultConfigProvider will fetch the actual values for ssl.keystore.location, ssl.keystore.password, ssl.keystore.type from the vault.
  • The broker will then validate the configs and apply the changes.

Proposed Changes 

Example

The following is an example config file that defines the configs and config providers in the same file and another notional ConfigProvider implementation named "vault":
Example of ConfigValues

...

  • For this KIP we will be enabling the auto resolution only for Connect component and passing false for the other components.  
  • This KIP proposes to reserve all config names starting with "config.providers" to resolve the config values. The config "config.providers"  will be reserved to list the ConfigProviders, "config.providers.providerName<providerName>.class" will be reserved to specify the class to be used to instantiate the providers and  "config.providers.providerName<providerName>.variableName<variableName>to specify any additional configs required by providers.  If any current configs match this custom configs format and auto resolution is enabled, we will try to resolve the value to a actual value by fetching it from the config provider, If a matching key is found in the config provider it will replace its value and could result in a failure. In order to prevent such side effects it would be recommended to make sure that existing configs matching the format are updated before enabling auto resolution. 

...

  1. Assume the ConfigProvider instances are also configured in the same configuration provided to the AbstractConfig constructor. This was rejected because it did not provide enough flexibility. For example, in Kafka Connect, the worker configurations can define ConfigProviders, whereas the connector configurations do not. Assuming the ConfigProviders are defined in the connector configuration would change the behavior of Connect.
  2. Place the new logic in a separate class or as static methods. This is a viable option, but the current design was thought to be a bit more usable. 
  3. In order to support dynamic configs add a new DynamicConfiguration class that allows a component to obtain an initial configuration instance (subclass of AbstractConfig) and to add a listener that will be called when any of the configuration properties resolved from ConfigProviders change in the underlying provider. This was rejected as it does not handle atomic updates efficiently which is required by broker configs. 
  4. In order to support atomic dynamic configs enhance ConfigProvider to handle batch updates of configs by adding few seconds of delay before the fetching the actual values  and modify ConfigChangeCallback to pass all the config changes at once in one giant map. This is not a viable option as related configs may be obtained from different ConfigProviders and adding delay in the config provider won't guarantee atomic updates.
  5. By default enable automatic resolution of externalized configs in AbstractConfig constructor. This approach was rejected as it gives more flexibility to enable/disable the feature by the subclass of AbstractConfig for all the components