During processing of a configuration class, the class's complete
hierarchy is processed and during the processing of each class its
member classes are processed. Previously, each pool-specific
inner-class of DataSourceConfiguration extended the abstract outer
class. This meant that when the import from
DataSourceAutoConfiguration.PooledDataSourceConfiguration caused the
first pool-specific inner-class to be processed,
DataSourceConfiguration would be processed as it was the inner-class's
superclass. In turn all of DataSourceConfiguration's member classes
would then be processed. This caused the first import (of
DataSourceConfiguration.Tomcat) to trigger processing of all of the
other pool-specific inner-classes in whatever order they were found
rather than them being processed in the order in which they are
imported by DataSourceAutoConfiguration.PooledDataSourceConfiguration.
Another part of the problem was that none of the pool-specific
inner-classes were conditional on a missing DataSource bean. This
meant that, when multiple pools were on the classpath, each class
after the first would override the previous class's definition of the
DataSource bean.
This commit updates each of the pool-specific inner-classes so that
they no longer extend DataSourceConfiguration. This ensures that
the inner classes are processed in the order defined in the import
on PooledDataSourceConfiguration. Each of the classes has also been
annotated with @ConditionalOnMissingBean(DataSource.class). This
prevents the DataSource bean definition from being overridden and
ensures that the order of precedence for the pool that will be used
is as defined in the import.
Closes gh-13737
Previously, if a jar that used custom compression configuration was
repackaged, a failure may occur if an entry in the repackaged jar had
a different compressed size to the entry in the source jar.
This commit updates JarWriter to clear the input entry's compressed
size (by setting it to -1) so that the repackaged entry's compressed
size does not have to match that of the input entry.
Closes gh-13720
This commit improves the for Actuator Metrics in WebClient.
Unlike the server couterpart of WebFlux, using a `retry` operator on a
`WebClient` pipeline does resubscribes to the whole chain.
The previous implementation recorded start time at the time of pipeline
build phase, but outside of it. This doesn't work since retrying the
same pipeline doesn't update the recorded start time and the duration of
sequential calls are cumulative.
This is now fixed using the Reactor `Context`, since we're now recording
the start time at subscription time and record metrics on `onNext` and
`onError` signals.
Closes gh-12228
