Hoplite

Hoplite is a Kotlin library for loading configuration files into typesafe classes in a boilerplate-free way. Define your config using Kotlin data classes, and at startup Hoplite will read from one or more config files, mapping the values in those files into your config classes. Any missing values, or values that cannot be converted into the required type will cause the config to fail with detailed error messages.

Features

• Multiple formats: Write your configuration in several formats: Yaml, JSON, Toml, Hocon, or Java .properties files or even mix and match formats in the same system.
• Property Sources: Per-system overrides are possible from JVM system properties, environment variables, JNDI or a per-user local config file.
• Batteries included: Support for many standard types such as primitives, enums, dates, collection types, inline classes, uuids, nullable types, as well as popular Kotlin third party library types such as NonEmptyList, Option and TupleX from Arrow.
• Custom Data Types: The Decoder interface makes it easy to add support for your custom domain types or standard library types not covered out of the box.
• Cascading: Config files can be stacked. Start with a default file and then layer new configurations on top. When resolving config, lookup of values falls through to the first file that contains a definition. Can be used to have a default config file and then an environment specific file.
• Beautiful errors: Fail fast at runtime, with beautiful errors showing exactly what went wrong and where.
• Preprocessors: Support for several preprocessors that will replace placeholders with values resolved from external configs, such as AWS Secrets Manager, Azure KeyVault and so on.
• Reloadable config: Trigger config reloads on a fixed interval or in response to external events such as consul value changes.

Changelog

See the list of changes in each release here.

Getting Started

Add Hoplite to your build:

implementation 'com.sksamuel.hoplite:hoplite-core:<version>'

You will also need to include a module for the format(s) you to use.

Next define the data classes that are going to contain the config. You should create a top level class which can be named simply Config, or ProjectNameConfig. This class then defines a field for each config value you need. It can include nested data classes for grouping together related configs.

For example, if we had a project that needed database config, config for an embedded HTTP server, and a field which contained which environment we were running in (staging, QA, production etc), then we may define our classes like this:

data class Database(val host: String, val port: Int, val user: String, val pass: String)
data class Server(val port: Int, val redirectUrl: String)
data class Config(val env: String, val database: Database, val server: Server)

For our staging environment, we may create a YAML (or Json, etc) file called application-staging.yaml. The name doesn't matter, you can use any convention you wish.

env: staging

database:
  host: staging.wibble.com
  port: 3306
  user: theboss
  pass: 0123abcd

server:
  port: 8080
  redirectUrl: /404.html

Finally, to build an instance of Config from this file, and assuming the config file was on the classpath, we can simply execute:

val config = ConfigLoaderBuilder.default()
               .addResourceSource("/application-staging.yml")
               .build()
               .loadConfigOrThrow<Config>()

If the values in the config file are compatible, then an instance of Config will be returned. Otherwise, an exception will be thrown containing details of the errors.

Config Loader

As you have seen from the getting started guide, ConfigLoader is the entry point to using Hoplite. We create an instance of this loader class through the ConfigLoaderBuilder builder. To this builder we add sources, configuration, enable reports, add preprocessors and more.

To create a default builder, use ConfigLoaderBuilder.default() and after adding your sources, call build. Here is an example:

ConfigLoaderBuilder.default()
  .addResourceSource("/application-prod.yml")
  .addResourceSource("/reference.json")
  .build()
  .loadConfigOrThrow<MyConfig>()

The default method on ConfigLoaderBuilder sets up recommended defaults. If you wish to start with a completely empty config builder, then use ConfigLoaderBuilder.empty().

There are two ways to retrieve a populated data class from config. The first is to throw an exception if the config could not be resolved. We do this via the loadConfigOrThrow<T> function. Another is to return a ConfigResult validation monad via the loadConfig<T> function if you want to handle errors manually.

For most cases, when you are resolving config at application startup, the exception based approach is better. This is because you typically want any errors in config to abort application bootstrapping, dumping errors immediately to the console.

Beautiful Errors

When an error does occur, if you choose to throw an exception, the errors will be formatted in a human-readable way along with as much location information as possible. No more trying to track down a NumberFormatException in a 400 line config file.

Here is an example of the error formatting for a test file used by the unit tests. Notice that the errors indicate which file the value was pulled from.

Error loading config because:

    - Could not instantiate 'com.sksamuel.hoplite.json.Foo' because:

        - 'bar': Required type Boolean could not be decoded from a Long (classpath:/error1.json:2:19)

        - 'baz': Missing from config

        - 'hostname': Type defined as not-null but null was loaded from config (classpath:/error1.json:6:18)

        - 'season': Required a value for the Enum type com.sksamuel.hoplite.json.Season but given value was Fun (/home/user/default.json:8:18)

        - 'users': Defined as a List but a Boolean cannot be converted to a collection (classpath:/error1.json:3:19)

        - 'interval': Required type java.time.Duration could not be decoded from a String (classpath:/error1.json:7:26)

        - 'nested': - Could not instantiate 'com.sksamuel.hoplite.json.Wibble' because:

            - 'a': Required type java.time.LocalDateTime could not be decoded from a String (classpath:/error1.json:10:17)

            - 'b': Unable to locate a decoder for java.time.LocalTime

Supported Formats

Hoplite supports config files in several formats. You can mix and match formats if you really want to. For each format you wish to use, you must include the appropriate hoplite module on your classpath. The format that hoplite uses to parse a file is determined by the file extension.

If you wish to add another format you can extend Parser and provide an instance of that implementation to the ConfigLoaderBuilder via addParser.

That same function can be used to map non-default file extensions to an existing parser. For example, if you wish to have your config in files called application.data but in yaml format, then you can register .data with the Yaml parser like this:

ConfigLoaderBuilder.default().addParser("data", YamlParser).build()

Property Sources

The PropertySource interface is how Hoplite reads configuration values.

Hoplite supports several built in property source implementations, and you can write your own if required.

The EnvironmentVariableOverridePropertySource, SystemPropertiesPropertySource and UserSettingsPropertySource sources are automatically registered, with precedence in that order. Other property sources can be passed to the config loader builder as required.

EnvironmentVariablesPropertySource

The EnvironmentVariablesPropertySource reads config from environment variables. It does not map cases. So, HOSTNAME does not provide a value for a field with the name hostname.

For nested config, use a period to separate keys, for example topic.name would override name located in a topic parent. Alternatively, in some environments a . is not supported in ENV names, so you can also use double underscore __. Eg topic__name would be translated to topic.name.

Optionally you can also create a EnvironmentVariablesPropertySource with allowUppercaseNames set to true to allow for uppercase-only names.

EnvironmentVariableOverridePropertySource

The EnvironmentVariableOverridePropertySource reads config from environment variables like the EnvironmentVariablesPropertySource. However, unlike that latter source, it is registered by default and only looks for env vars with a special config.override. prefix. This prefix is stripped from the variable before being applied. This can be useful to apply changes at runtime without requiring a build.

For example, given a config key of database.host, if an env variable exists with the key config.override.database.host, then the value in the env var would override.

In some environments a . is not supported in ENV names, so you can also use double underscore __. Eg topic__name would be translated to topic.name.

SystemPropertiesPropertySource

The SystemPropertiesPropertySource provides config through system properties that are prefixed with config.override.. For example, starting your JVM with -Dconfig.override.database.name would override a config key of database.name residing in a file.

UserSettingsPropertySource

The UserSettingsPropertySource provides config through a config file defined at ~/.userconfig.[ext] where ext is one of the supported formats.

InputStreamPropertySource

The InputStreamPropertySource provides config from an input stream. This source requires a parameter that indicates what the format is. For example, InputStreamPropertySource(input, "yml")

ConfigFilePropertySource

Config from files or resources are retrieved via instances of ConfigFilePropertySource. This property source is added automatically when we pass strings to the loadConfigOrThrow or loadConfig functions.

There are convenience methods on ConfigLoaderBuilder to construct ConfigFilePropertySources from resources on the classpath or files.

For example, the following are equivalent:

ConfigLoader().loadConfigOrThrow<MyConfig>("config.json")

and

ConfigLoaderBuilder.default()
   .addResourceSource("/config.json")
   .build()
   .loadConfigOrThrow<MyConfig>()

The advantage of the second approach is that we can specify a file can be optional, for example:

ConfigLoaderBuilder.default()
  .addResourceSource("/missing.yml", optional = true)
  .addResourceSource("/config.json")
  .build()
  .loadConfigOrThrow<MyConfig>()

JsonPropertySource

To use a JSON string as a property source, we can use the JsonPropertySource implementation. For example,

ConfigLoaderBuilder.default()
   .addSource(JsonPropertySource(""" { "database": "localhost", "port": 1234 } """))
   .build()
   .loadConfigOrThrow<MyConfig>()

YamlPropertySource

To use a Yaml string as a property source, we can use the YamlPropertySource implementation.

ConfigLoaderBuilder.default()
   .addSource(YamlPropertySource(
     """
        database: "localhost"
        port: 1234
     """))
   .build()
   .loadConfigOrThrow<MyConfig>()

TomlPropertySource

To use a Toml string as a property source, we can use the TomlPropertySource implementation.

ConfigLoaderBuilder.default()
  .addSource(TomlPropertySource(
    """
        database = "localhost"
        port = 1234
     """))
  .build()
  .loadConfigOrThrow<MyConfig>()

PropsPropertySource

To use a java.util.Properties object as property source, we can use the PropsPropertySource implementation.

ConfigLoaderBuilder.default()
  .addSource(PropsPropertySource(myProps))
  .build()
  .loadConfigOrThrow<MyConfig>()

Cascading Config

Hoplite has the concept of cascading or layered or fallback config. This means you can pass more than one config file to the ConfigLoader. When the config is resolved into Kotlin classes, a lookup will cascade or fall through one file to another in the order they were passed to the loader, until the first file that defines that key.

For example, if you had the following two files in yaml:

application.yaml:

elasticsearch:
  port: 9200
  clusterName: product-search

application-prod.yaml:

elasticsearch:
  host: prd-elasticsearch.scv
  port: 8200

And both were passed to the ConfigLoader like this: ConfigLoader().loadConfigOrThrow<Config>("/application-prod.yaml", "/application.yaml"), then lookups will be attempted in the order the files were declared. So in this case, the config would be resolved like this:

elasticsearch.port = 8200 // the value in application-prod.yaml takes priority
elasticsearch.host = prd-elasticsearch.scv // only defined in application-prod.yaml
elasitcsearch.clusterName = product-search // only defined in application.yaml

Let's see a more complicated example. In JSON this time.

default.json

{
  "a": "alice",
  "b": {
    "c": true,
    "d": 123
  },
  "e": [
    {
      "x": 1,
      "y": true
    },
    {
      "x": 2,
      "y": false
    }
  ],
  "f": "Fall"
}

prod.json

{
  "a": "bob",
  "b": {
    "d": 999
  },
  "e": [
    {
      "y": true
    }
  ]
}

And we will parse the above config files into these data classes:

enum class Season { Fall, Winter, Spring, Summer }
data class Foo(val c: Boolean, val d: Int)
data class Bar(val x: Int?, val y: Boolean)
data class Config(val a: String, val b: Foo, val e: List<Bar>, val f: Season)

val config = ConfigLoader.load("prod.json", "default.json")
println(config)

The resolution rules are as follows:

• "a" is present in both files and so is resolved from the first file - which was "prod.json"
• "b" is present in both files and therefore resolved from the file as well
• "c" is a nested value of "b" and is not present in the first file so is resolved from the second file "default.json"
• "d" is a nested value of "b" present in both files and therefore resolved from the first file
• "e" is present in both files and so the entire list is resolved from the first file. This means that the list only contains a single element, and x is null despite being present in the list in the first file. List's cannot be merged.
• "f" is only present in the second file and so is resolved from the second file.

Strict Mode

Hoplite can be configured to throw an error if a config value is not used. This is useful to detect stale configs.

To enable this setting, use .strict() on the config builder. For example:

ConfigLoaderBuilder.default()
  .addResourceSource("/config-prd.yml", true)
  .addResourceSource("/config.yml")
  .strict()
  .build()
  .loadConfig<MyConfig>()

An example of this output is:

Error loading config because:

    Config value 'drop_drop' at (classpath:/snake_case.yml:0:10) was unused

    Config value 'double_trouble' at (/home/sam/.userconfig.yml:2:16) was unused

Aliases

If you wish to refactor your config classes and rename a field, but you don't want to have to update all your config files, you can add a migration path by allowing a field to use more than one name. To do this we use the @ConfigAlias annotation.

For example, with this config file:

database:
  host: String

We can marshall this into the following data classes.

data class Database(val host: String)
data class MyConfig(val database: Database)

or

data class Database(@ConfigAlias("host") val hostname: String)
data class MyConfig(val database: Database)

Param Mappers

Hoplite provides an interface ParameterMapper which allows the parameter name to be modified before it is looked up inside a config source. This allows hoplite to find config keys which don't match the exact name. The main use case for this is to allow snake_case or kebab-case names to be used as config keys.

For example, given the following config class:

data class Database(val instanceHostName: String)

Then we can of course define our config file (using YML as an example):

database:
    instanceHostName: server1.prd

But because Hoplite registers KebabCaseParamMapper and SnakeCaseParamMapper automatically, we can just as easily use:

database:
  instance-host-name: server1.prd

or

database:
  instance_host_name: server1.prd

Decoders

Hoplite converts the raw value in config files to JDK types using instances of the Decoder interface. There are built in decoders for all the standard day to day types, such as primitives, dates, lists, sets, maps, enums, arrow types and so on. The full list is below:

Duration formats

Duration types support unit strings in the following format (lower case only), with an optional space between the unit value and the unit type.

• ns, nano, nanos, nanosecond, nanoseconds
• us, micro, micros, microsecond, microseconds
• ms, milli, millis, millisecond, milliseconds
• s, second, seconds
• m, minute, minutes
• h, hour, hours
• d, day, days

For example, 10s, 3 days, or 12 hours.

Preprocessors

Hoplite supports what it calls preprocessors. These are just functions that are applied to every value as they are read from the underlying config file. The preprocessor is able to transform the value (or return the input - aka identity function) depending on the logic of that preprocessor.

For example, a preprocessor may choose to perform environment variable substitution, configure default values, perform database lookups, or whatever other custom action you need when the config is being resolved.

You can add custom pre-processors in addition to the built in ones, by using the function withPreprocessor on the ConfigLoader class, and passing in an instance of the Preprocessor interface. A typical use case of a custom preprocessor is to lookup some values in a database, or from a third party secrets store such as Vault or Amazon Parameter Store.

One way this can be implemented is to have a prefix, and then use a preprocessor to look for the prefix in strings, and if the prefix is present, use the rest of the string as a key to the service. The PrefixProcessor abstract class implements this by handling the node traversal, while leaving the specific processing as an exercise for the reader.

For example

database:
  user: root
  password: vault:/my/key/path

Note: You can repeatedly apply preprocessors by setting the property withPreprocessingIterations on the ConfigLoaderBuilder to a value greater than 1. This causes looped application of all preprocessors. This can be useful if you wish to have one preprocessor resolve a value that then needs to be resolved by another preprocessor.

Built-in Preprocessors

These built-in preprocessors are registered automatically.

Optional Preprocessors

These preprocessors must be added to the ConfigBuilder before they take effect, and require extra modules to be added to the build.

Random Preprocessor

The random preprocessor replaces placeholder strings with random values.

For example:

my.number=${random.int}
my.bignumber=${random.long}
my.uuid=${random.uuid}
my.number.less.than.ten=${random.int(10)}
my.number.in.range=${random.int[1024,65536]}

Masked values

It is quite common to output the resolved config at startup for reference when debugging. In this case, the default toString generated by Kotlin's data classes is very useful. However configuration typically includes sensitive information such as passwords or keys which normally you would not want to appear in logs.

To avoid sensitive fields appearing in the log output, Hoplite provides a built in type called Masked which is a wrapper around a String. By declaring a field to have this type, the value will still be loaded from configuration files, but will not be included in the generated toString.

For example, you may define a config class like this:

data class Database(val host: String, val user: String, val password: Masked)

And corresponding json config:

{
  "host": "localhost",
  "user": "root",
  "password": "letmein"
}

And then the output of the Database config class via toString would be Database(host=localhost, user=root, password=****)

Note: The masking effect only happens if you use toString. If you marshall your config to a String using a reflection based tool like Jackson, it will still be able to see the underlying value. In these cases, you would need to register a custom serializer. For the Jackson project, a HopliteModule object is available in the hoplite-json module. Register this with your Jackson mapper, like mapper.registerModule(HopliteModule) and then Masked values will be ouputted into Json as "****"

Inline Classes

Some developers, this writer included, like to have strong types wrapping simple values. For example, a Port object rather than an Int. This helps to alleviate Stringy typed development. Kotlin has support for what it calls inline classes which fulfil this need.

Hoplite directly supports inline classes. When using inline classes, you don't need to nest config keys.

For example, given the following config classes:

inline class Port(val value: Int)
inline class Hostname(val value: String)
data class Database(val port: Port, val host: Hostname)

And then this config file:

port: 9200
host: localhost

We can parse directly:

val config = ConfigLoader().loadConfigOrThrow<Database>("config.file")
println(config.port) // Port(9200)
println(config.host) // Hostname("localhost")

Sealed Classes

Hoplite will support sealed classes where it is able to match up the available config keys with the parameters of one of the implementations. For example, lets create a config hierarchy as implementations of a sealed class.

sealed class Database {
  data class Elasticsearch(val host: String, val port: Int, val index: String) : Database()
  data class Postgres(val host: String, val port: Int, val schema: String, val table: String) : Database()
}

data class TestConfig(val databases: List<Database>)

For the above definition, if hoplite encountered a host, port, and index then it would be clear that it should instantiate an Elasticsearch instance. Similarly, if the config keys were host, port, schema, and table, then the Postgres implementation should be used. If the keys don't match an implementation, the config loader would fail. If keys match multiple implementations then the first match is taken.

For example, the following yaml config file could be used:

databases:
  - host: localhost
    port: 9200
    index: foo
  - host: localhost
    port: 9300
    index: bar
  - host: localhost
    port: 5234
    schema: public
    table: faz

And the output would be:

TestConfig(
  databases=[
    Elasticsearch(host=localhost, port=9200, index=foo),
    Elasticsearch(host=localhost, port=9300, index=bar),
    Postgres(host=localhost, port=5234, schema=public, table=faz)
  ]
)

Objects in sealed classes

Hoplite additionally supports using objects in sealed classes. For example, lets expand the database definition to include an Embedded object subclass:

sealed class Database {
  data class Elasticsearch(val host: String, val port: Int, val index: String) : Database()
  data class Postgres(val host: String, val port: Int, val schema: String, val table: String) : Database()
  object Embedded : Database()
}

data class TestConfig(val databases: List<Database>)

We can indicate to Hoplite to use the Embedded option in two ways. The first is by referencing the type name:

For example, in yaml:

database: Embedded

Or in Json:

{
  "database": "Embedded"
}

This also works for lists, and we can mix and match:

Yaml:

databases:
  - "Embedded"
  - host: localhost
    port: 9300
    index: bar

Json:

{
  "databases": ["Embedded", { "host": "localhost", "port": 9200, "index": "foo" }]
}

The second method is only for Json by specifying an empty object:

{
  "database": { }
}

When using the second option, there must be only a single object instance in the hierarchy, otherwise a disambiguation error is thrown. If you want to support multiple object instances, then refer to the type by name.

Reloadable Config

Hoplite embraces immutable config, but if you require that config is dynamic, then Hoplite provides a ReloadableConfig wrapper. This functionality is available by adding the module hoplite-watch to your build. The reloader requires a ConfigLoader and then accepts one or more Watchables which cause the config to be reloaded when whatever they are watching is triggered.

To be clear, once Hoplite has parsed a config object, it won't mutate that object. This reloadable wrapper will, in the background, reload the config once a watcher is triggered. Then you can obtain the latest parsed config whenever you wish by using the method getLatest.

If you wish to be notified whenever the config is reloaded, you can call subscribe on the reloader.

A simple example would be to referesh config every 10 seconds:

// create a watchable that will trigger every 10 seconds
val watcher = FixedIntervalWatchable(10.seconds)

// create our config loader which will parse config when invoked
val loader = ConfigLoaderBuilder.default()
  .addSource(PropertySource.resource("/application.yml"))
  .build()

// create the reloader, adding the watcher, the config loader, and specifying the target config class
val reloader = ReloadableConfig(configLoader, TestConfig::class)
  .addWatcher(watcher)

// obtain the latest config whenever we want
reloader.getLatest()

// or subscribe for notifications: (TestConfig) -> Unit
reloader.subscribe { println("New config!: $it") }

You can implement the Watchable interface directly, with whatever triggering logic you wish, or use one of the predefined implementations:

Add on Modules

Hoplite makes available several other modules that add functionality outside of the main core module. They are in seperate modules because they bring in dependencies from those projects and so the modules are optional.

GraalVM native image

GraalVM native image example can be found inside an example-native subdirectory.

License

This software is licensed under the Apache 2 license, quoted below.

Copyright 2019 Stephen Samuel

Licensed under the Apache License, Version 2.0 (the "License"); you may not
use this file except in compliance with the License. You may obtain a copy of
the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS, WITHOUT
WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the
License for the specific language governing permissions and limitations under
the License.