There's a window where the deny-all turbo filter has been removed but
Logback has not yet been configured. If any logging that would have
reached an appender is performed in this window, unwanted status
messages will be logged. This window can be closed by ensuring that
the turbo filter is in place while Logback is being configured.
Closes gh-34505
Previously, the runtime shutdown hook was added as soon as a
shutdown handler was registered. This causes a memory leak in a war
deployment when the application is undeployed as
LoggingApplicationListener always registers a shutdown handler
so the runtime shutdown hook was always registered.
This commit updates the shutdown hook so that the runtime shutdown
hook is only allowed to be added once run() has been called on a
SpringApplication with the shutdown hook enabled. This approach
allows the registerShutdownHook flag on SpringApplication to be a
central point of control for the registration of the runtime shutdown
hook. When that flag is set to false, for example by
SpringBootServletInitializer, the runtime shutdown hook will not
be registered, irrespective of whether other code uses the public
API to add a shutdown handler.
An alternative approach of stopping LoggingApplicationListener from
adding its shutdown handler – for example by adding
logging.register-shutdown-hook=false to the environment – was
considered. This approach was rejected in favor of the centralized
approach described above as it would require every caller that adds
a shutdown handler to deal with the problem.
Closes gh-37096
This commit adapts to API changes in Spring Framework, see
spring-projects/spring-framework#31117
Previously, the "autowired" executable to use for a bean was always
resolved, even if a custom code fragment didn't really need it. This
is key for binding of immutable configuration properties as we use an
instance supplier for it.
This changes means that the workaround added in maintenance releases
can be removed.
See gh-37337
Previously, AOT processing failed on processing an immutable
configuration properties that declare several constructors as the core
framework infrastructure tries to resolve the "autowired" constructor
to use, even if the custom code fragments are never going to use it.
This commit workarounds the problem in maintenance releases until a
proper fix is provided in the core framework. When AOT runs, a
SmartInstantiationAwareBeanPostProcessor is added to the bean factory
to provide the constructor to use. This implementation relies on the
same algorithm that the binder uses at runtime.
Closes gh-37283
Update `PrivateKeyParser` implementations so that algorithm parameters
for PEM encoded 'EC PRIVATE KEY' files are loaded from the incoming
data.
Prior to this commit, the algorithm parameter was hard-coded to
'secp384r1' which could result in a mismatch to the actual file
content.
Fixes gh-34232
This commit configures the virtual threads support in Reactor Core for
Schedulers if:
* the current JDK is 21 or higher
* the current environment enables this globally with the
`"spring.threads.virtual.enabled"` property.
This needs to happen early in the application startup process, as this
feature is detected statically when the first schedulers call is made.
As a result, this is being done with an environment post processor and
not with an auto-configuration class.
Closes gh-36302
This commit makes sure that sub-namespace that are defined in a flat
manner in a configuration properties are considered for runtime hints
Closes gh-36909
This auto-configures a new SimpleAsyncTaskSchedulerBuilder bean in the
context. This bean is configured to use virtual threads, if enabled.
SimpleAsyncTaskSchedulerCustomizers can be used to customize the built
SimpleAsyncTaskScheduler.
If virtual threads are enabled, the application task scheduler is
configured to be a SimpleAsyncTaskScheduler.
Adds a new configuration property spring.task.scheduling.simple
.concurrency-limit
Closes gh-36609
Before that change, we get 32 random bytes, and then used MD5 on them to
get a hex string. This removes the MD5, we now get 128 bits (output size
of MD5) of random bytes directly.
Previously a schema or data locatio that was a directory would
result in an attempt being made to apply the directory listing as
SQL scripts. This would typically result in a hard to diagnose
failure due to the directory listing not being valid SQL.
This commit updates the initializer to ignore locations for
which the Resources is not readable. This works as Framework's
Resource abstraction does not consider directory resources to be
readable.
Closes gh-36386
Andy Wilkinson
Moritz Halbritter
Johnny Lim
Phillip Webb
Stephane Nicoll
hakan-krgn
Brian Clozel
Yanming Zhou
abaudat