Previously, the heap dump endpoint test asserted that the temporary
heap dump file had been deleted as soon as the client received a
response. This led to intermittent test failures as the input
stream is closed after its contents have been sent to the client,
creating a race condition between the client receiving the response
and then asserting that the file had been deleted and the server
close the input stream and deleting the temporary file.
This commit updates the test so that, after receiving the response, it
will wait for up to 5 seconds for the server to have deleted the
temporary heap dump file.
If an annotation attribute is linked to in javadoc before the
javadoc processor encounters a usage of the annotation, the javadoc
tool fails with a class cast exception. This is a known issue [1]
but it has been closed as won't fix so we need to work around it.
Sadly, the only reasonable way to do so appears to be to remove
the links to the annotation attributes and only link to the annotation
itself.
[1] https://bugs.openjdk.java.net/browse/JDK-8170447
This commit migrates the Actuator onto the new endpoint infrastruture.
In addition to the existing support for accessing the endpoints via
JMX and HTTP using Spring MVC, support for access via HTTP using
Jersey and WebFlux has been added. This includes using a separate
management port where we now spin up an additional, appropriately
configured servlet or reactive web server to expose the management
context on a different HTTP port to the main application.
Closes gh-2921
Closes gh-5389
Closes gh-9796
Jolokia is a 100% web concern and does not fit in the Endpoint
infrastructure. This commit removes `JolokiaMvcEndpoint` and exposes
the servlet directly instead while still being part of the
management context. As such, the Jolokia servlet is exposed beneath
the management context path and will move to a separate port when
the management port is not the same as the main server port.
Closes gh-9843
Previously, when read too quickly, the current note may lead a user to
think that RANDOM_PORT or DEFINED_PORT directly affect transaction
management. This is not the case.
This commit attempts to clarify that the difference in transaction
behaviour is due to transactions being bound to a particular thread
and separate threads being used for the client and server when
tests are making HTTP requests.
Closes gh-9965
* gh-9914:
Avoid NPE in AnnotationsPropertySource if getAnnotations returns null
Polish "Avoid NPE in PropertyMappingContextCustomizer"
Avoid NPE in PropertyMappingContextCustomizer
Previously when Handler was creating a URL from a context URL and a
spec, any occurrances of /../ or /./ in the spec would be left as-is.
This differed from the JDK's Handler implementation which normalizes
the URL by modifying the path to remove any occurrences of /../ or
/./
This commit updates our Handler implementation to align it with the
JDK's. Tests have been added to assert that, given the same inputs,
the two Handler classes produce the same output.
Closes gh-9917
Stephane Nicoll
Oleg Zhurakousky
Andy Wilkinson
Benoit Lacelle
pulsation
Johnny Lim
Dennis Kieselhorst
Christian Bühler