Get ahead
VMware offers training and certification to turbo-charge your progress.
Learn moreSpring Boot 2.4.0.M2 has just been released, and it brings with it some interesting changes to the way that application.properties
and application.yml
files are loaded.
If your applications use the fairly typical setup of only using a single application.properties
or application.yml
file, then you’ll probably not notice any difference. If, however, you have applications that use more complex setups (such as profile specific properties), you might want to read on to learn about what we’ve changed and why.
With recent versions of Spring Boot we’ve been working hard to improve our Kubernetes support. One thing that we wanted to add in Spring Boot 2.3, but couldn’t, was support for volume mounted configuration.
Volume mounted configuration is a popular feature of Kubernetes where a ConfigMap
directive is used to surface configuration directly on the file system. You can either mount a complete YAML file that contains an amalgamation of multiple keys and values, or you can use a simpler directory tree format where the filename is the key and the contents of the file is the value.
We wanted to offer support for both, and in a way that felt natural to use alongside our existing application.properties
and application.yml
support. To do that we needed to touch the dreaded ConfigFileApplicationListener
class.
A few years ago some amusing clips from the video game Trap Adventure 2 started doing the rounds. They’re a pretty good analogy for what can happen in software. Sometimes you just find yourself with areas of code that are very hard to change. In Spring Boot, ConfigFileApplicationListener
ended up becoming one of these "trap adventures".
It’s not that code is badly written, or is missing tests. It’s just that as we added features to it, we ended up tying ourselves in knots.
The two main problems that we have with the code are to do with profile specific documents (mainly in YAML). Namely:
You can enable additional profiles from a profile specific document.
It’s hard to know the order that documents will be added.
Take the following example:
spring.profiles: !dev spring.profiles.include: local security.user.password: userc
Here we have a multi-document YAML file (a single file that’s comprised of three logical documents separated by ---
).
If you run with --spring.profiles.active=prod
what’s the value of security.user.password
? Is the runlocal
property set? Are you sure? Is the middle document even included since the profile isn’t activated at the time it’s processed?
We regularly had issues raised about this file processing logic, but whenever we tried to fix them something would break for someone else. We ultimately decided the only way forward was to rethink the entire thing.
So in Spring Boot 2.4 we’re planning to make two significant changes to the way the properties and YAML files are loaded:
Documents will be loaded in the order that they’re defined.
Profiles can no longer be activated from profile specific documents.
Starting with Spring Boot 2.4, there will be a simple rule that can be applied when loading properties and YAML files. Properties declared lower in the file will override those higher up.
This follows the same ordering rules that plain old .properties
files already use. Think of each line putting an entry in a Map
. Any existing entry is replaced when a new value with the same key is put into the Map
.
Following these rules, given a multi-document YAML file, the lower document will override values in the higher one:
test: "overridden-value"
With Spring Boot 2.4, we’ve decided to bring YAML-like multi-document support to Java properties files. Multi-document properties files use a comment (#
) followed by the familiar three dash notation to split documents (we chose to use a comment so that existing IDE tooling wouldn’t break).
For example, the properties equivalent of the YAML snippet above would be:
test=value #--- test=overridden-value
The above example is a bit artificial since it wouldn’t really make sense to always override a value. A more common setup would be to declare that the second document is only active with a specific profile.
In Spring Boot 2.3, you’d use the spring.profiles
key to do this. With Spring Boot 2.4, we’ve decided to change the property to spring.config.activate.on-profile
.
For example, if we only want to override test
when the dev
profile is active, we can use the following:
test=value #--- spring.config.activate.on-profile=dev test=overridden-value
You can still use spring.profiles.active
properties to activate or include profiles from an application.properties
or application.yaml
file.
For example, the following is perfectly valid:
test=value spring.profiles.active=local #--- spring.config.activate.on-profile=dev test=overridden value
One thing that you’re no longer allowed to do is use that property in combination with spring.config.activate.on-profile
. For example, the following file would now throw an exception:
test=value #--- spring.config.activate.on-profile=dev spring.profiles.active=local # will fail test=overridden value
We hope that this new restriction will ultimately make your application.properties
and application.yml
files easier to reason about and understand. We also hope that it’ll make Spring Boot itself easier to manage and maintain. We are, however, aware of a least one valid use-case where people want to expand a profile into multiple sub-profiles. In order to support that, we’re adding a feature called "profile groups".
Profile groups are a new feature in Spring Boot 2.4 that allow you to expand a single profile into multiple sub-profiles. For example, suppose you have a complex set of @Configuration
classes that you conditionally enable using @Profile
annotations. You might have a database configuration with @Profile("proddb")
, a message configuration with @Profile("prodmq")
and so on.
Using multiple discrete profiles might make your code easier to reason about, but it’s not ideal for deployment. You don’t want to force users to remember that they must always activate proddb
, prodmq
, prodmetrics
etc at the same time. Instead, you just want them to be able to activate a single prod
profile. Groups allow you to do just that.
To define a group, you can use the spring.profiles.group
property in your application.properties
or application.yml
file. For example:
spring.profiles.group.prod=proddb,prodmq,prodmetrics
Now that we’ve fixed that basic issues with config file processing, we’re finally able to think about new features that we want to offer. The main one that we’re delivering with Spring Boot 2.4 is support for importing additional configuration.
With earlier versions of Spring Boot, it was pretty hard to import additional properties or yaml files beyond application.properties
and application.yml
. You could use the spring.config.additional-location
property, but you needed to set it pretty early and it was quite limited with the types of files that it could deal with.
With the latest milestone, you can use the new spring.config.import
property directly in your application.properties
or application.yml
files. For example, you might want to import a "git ignored" developer.properties
file so that any developer in your team can quickly change properties just for them:
application.name=myapp spring.config.import=developer.properties
You can even combine spring.config.import
declarations with spring.config.activate.on-profile
properties. For example, here we load prod.properties
only if the prod
profile is active:
spring.config.activate.on-profile=prod spring.config.import=prod.properties
Imports can be considered as additional documents inserted just below the document that declares them. They follow the same top-down ordering as regular multi-document files: An import will only be imported once, no matter how many times it is declared.
Import definitions use URL-like syntax as their value. If your location doesn’t have a prefix, it’s considered to be a regular file or folder. If, however, you use the configtree:
prefix, you’re telling Spring Boot that you expect a Kubernetes style volume mounted configuration tree at the location.
For example, you can declare the following in your application.properties
:
spring.config.import=configtree:/etc/config
If you have the following mounted content:
etc/ +- config/ +- my/ | +- application +- test
You end up with my.application
and test
properties in your Spring Environment
. The value for my.application
will be the contents of /etc/config/my/application
, and the value of test
will be the contents of /etc/config/test
.
If you only want volume mounted config trees (or any properties for that matter) to be active on a specific cloud platform, you can use the spring.config.activate.on-cloud-platform
property. This works in a similar way to the spring.config.activate.on-profile
property, but uses CloudPlatform
values, rather than profile names.
If we want to only enable our configtree example above when we’re deployed to Kubernetes, we can do the following:
spring.config.activate.on-cloud-platform=kubernetes spring.config.import=configtree:/etc/config
The location strings specified in spring.config.import
properties are completely pluggable and may be extended by writing a few custom classes. We expect that third-party libraries may start to offer support for custom locations in the future. For example, you could imagine third-party jars to support locations such as archaius://…
, vault://…
or zookeeper://…
.
If you’re interested in adding additional location support, look at the javadoc for ConfigDataLocationResolver
and ConfigDataLoader
in the org.springframework.boot.context.config
package.
If you’re upgrading an existing Spring Boot application, and you don’t feel comfortable using all these new features, you can switch back to the older processor. To do that, you can set spring.config.use-legacy-processing
to true
to your application.properties
or application.yml
file. This should give you application config processing that’s identical to a Spring Boot 2.3 application.
If you do find you need to switch to legacy processing because we’ve missed a specific use-case, please raise an issue on GitHub and we’ll try to address it.
We hope the new config data processing classes are useful, and they don’t cause too much upgrade pain. If you want to read more about them, you can checkout the updated reference documentation.