We need to tread a very fine line. Ideally the process of refreshing
the Environment builds it back exactly as happened on startup.
In particular we probably don't want to have things like
application.properties already there when we start the process.
This change replaces the more liberal copy of all property sources
only a select few (actually only one), paremeterized as a constant.
If there were *no* bootstrapProperties (i.e. properties added by
PropertySourceLocators) on startup, and then in a refresh event
some new sources were picked up, they would have been added to the
end, instead of the beginning of the list.
Fixes https://github.com/spring-cloud/spring-cloud-config/issues/815
When the Environment is resetted the EnvironmentChangedEvent always
contains an empty keySet, since the live view of the cleared map was
used. This commit fixes this.
Instead it depends on ContextRefresher with actually does the work.
This allows applications to use the listener even if they don't depend
on actuator.
fixes gh-171
The fix for gh-153 was sensible, but has now exposed another issue
which is that the bootstrap initializers, whilst they are now only
created once, actually only get applied to the parent context
in a hierarchy.
Most of the time this doesn't matter because the child contexts all
benefit from environment changes and additional beans added
to the parent. If the children are supposed to actually have
different environments, though, there is a problem. So this change
locates the existing bootstrap context and applies it to each
new context created separately.
Unfortunately we have to use reflection to do that for now because
there is no way to discover the parent with a public API before
the context is refreshed.
See gh-190
For RefreshScope this stops Spring from moaning about the @Bean
method with a WARN log (which is benign but hard to get rid of
and confusing for users).
See gh-401
LoggingSystemShutdownListener runs immediately after
BootstrapApplicationListener and contains the code which used to be
in that to clean up the logging system (and make it go dark). That
is still a little bit of a hack to work around some limitations in
the static logging libraries (via the Spring Boot Logging System
and LoggingApplicationListener).
If the LoggingApplicationListener is then *not* applied in the
ContextRefresher, then we don't need the LoggingSystemShutdownListener
either. In fact we can narrow down the ContextRefresher to be only
interested in the listeners that affect the Environment (of which
we know of only 2). It means that any listeners that are added that
affect the Environment will not be applied, but there are plenty of
more "official" channels for modifying the Environment
(EnvironmentPostProcessor and PropertySourceLocator), so users can
easily migrate to a better implementation.
Completes the work started in 7f687b1, where the call to
LoggingSystem.beforeInitialization() was ommitted which results
in JUL logging never being available in the application.
Partial fix for #131 (but a /refresh is still broken)
Given the following in one property source
```
mine[0].someValue=Foo
mine[0].someKey={cipher}XXXX
mine[1].someValue=Bar
mine[1].someKey={cipher}XXXX
yours[0].anotherKey=Baz
```
only `mine[*].someKey` were added to "decrypted" property source. Due to house spring boot manages indexed properties, they can not be merged across property source. This change includes all indexed properties in the "decrypted" property source if any one of them is encrypted.
In the above example, `yours[*].*` is included even though unrelated.
fixes gh-466
Anticipating a change in Spring Boot 1.4.1 to support child context
creation without (needless) re-initialization of the logging system,
we need to carefully manage the lifecycle, in particular calling
cleanUp() when we know there are changes in the pipeline.
Fixes gh-125
