This commit updates the `ScheduledTaskObservationDocumentation` to
better align the contributed KeyValues with OpenTelemetry conventions
for observations of code executions.
Instead of a "target.type" key with the bean class simple name, this
is now contributing the canonical class name of the bean under the
"code.namespace" key.
The "method.name" key is renamed to "code.function" and its values
remain unchanged.
Closes gh-30721
This commit enhances the `ScheduledAnnotationBeanPostProcessor` to
instrument `@Scheduled` methods declared on beans. This will create
`"tasks.scheduled.execution"` observations for each execution of a
scheduled method. This supports both blocking and reactive variants.
By default, observations are no-ops; developers must configure the
current `ObservationRegistry` on the `ScheduledTaskRegistrar` by using a
`SchedulingConfigurer`.
Closes gh-29883
This commit overhauls the TestExecutionListener for Micrometer's
ObservationRegistry that was introduced in the previous commit.
Specifically, this commit:
- Renames the listener to MicrometerObservationRegistryTestExecutionListener
since the use of a ThreadLocal is an implementation detail that may
change over time.
- Makes the listener package-private instead of public in order to
allow the team greater flexibility in evolving this feature.
- Eagerly loads the ObservationThreadLocalAccessor class and verifies
that it has a getObservationRegistry() method to ensure that the
listener is properly skipped when SpringFactoriesLoader attempts to
load it, if Micrometer 1.10.8+ is not on the classpath.
- Switches the listener's automatic registration order to 2500 in order
to register it after the DependencyInjectionTestExecutionListener.
- Only tracks the previous ObservationRegistry in beforeTestMethod() if
the test's ApplicationContext contains an ObservationRegistry bean.
- Properly removes the TestContext attribute for the previous
ObservationRegistry in afterTestMethod().
- Introduces DEBUG logging for diagnostics.
- Adds an entry in the Javadoc for TestExecutionListener as well as in
the Testing chapter in the reference manual.
Closes gh-30658
This commit adds a note in the reference documentation stating that
`ErrorHandler` infrastructure is not involved when reactive methods send
an error signal: the exception is sent as a message in the pipeline and
is not thrown from the task `Runnable`.
See gh-23533
This commit adds support for `@Scheduled` annotation on reactive
methods and Kotlin suspending functions.
Reactive methods are methods that return a `Publisher` or a subclass
of `Publisher`. The `ReactiveAdapterRegistry` is used to support many
implementations, such as `Flux`, `Mono`, `Flow`, `Single`, etc.
Methods should not take any argument and published values will be
ignored, as they are already with synchronous support.
This is implemented in `ScheduledAnnotationReactiveSupport`, which
"converts" Publishers to `Runnable`. This strategy keeps track of
active Subscriptions in the `ScheduledAnnotationBeanPostProcessor`,
in order to cancel them all in case of shutdown.
The existing scheduling support for tasks is reused, aligning the
triggering behavior with the existing support: cron, fixedDelay and
fixedRate are all supported strategies.
If the `Publisher` errors, the exception is logged at warn level and
otherwise ignored. As a result new `Runnable` instances will be
created for each execution and scheduling will continue.
The only difference with synchronous support is that error signals
will not be thrown by those `Runnable` tasks and will not be made
available to the `org.springframework.util.ErrorHandler` contract.
This is due to the asynchronous and lazy nature of Publishers.
Closes gh-23533
Closes gh-28515
Instead of having antora run always, which happens regardless of
whether anything has changed, we'll have it invoked as part of the CI
build, and we'll have to run it locally ourselves when necessary.
See gh-30481
Prior to this commit, the Observation instrumentation for Reactive
server applications was implemented with a `WebFilter`. This allowed to
record observations and set up a tracing context for the controller
handlers.
The limitation of this approach is that all processing happening at a
lower level is not aware of any observation. Here, the
`HttpWebHandlerAdapter` handles several interesting aspects:
* logging of HTTP requests and responses at the TRACE level
* logging of client disconnect errors
* handling of unresolved errors
With the current instrumentation, these logging statements will miss the
tracing context information. As a result, this commit deprecates the
`ServerHttpObservationFilter` in favor of a more direct instrumentation
of the `HttpWebHandlerAdapter`. This enables a more precise
instrumentattion and allows to set up the current observation earlier in
the reactor context: log statements will now contain the relevant
information.
Fixes gh-30013
This commit improves how the build deals with javadoc invalid references
in two ways.
Link/see references that are temporarily invalid during javadoc
generation of individual modules are better masked by using the option
`Xdoclint:syntax` instead of `Xdoclint:none` (warnings were still
visible in some cases, e.g. when individually building the javadoc for
a specific module).
Global javadoc-building task `api` now combines `syntax` and `reference`
`Xdoclint` groups, allowing to raise truly invalid references even when
all the modules have been aggregated.
This commit also fixes the 20+ errors which appeared following the later
change in doclet configuration.
Closes gh-30428
Brian Clozel
rstoyanchev
Sam Brannen
Anton Gubenko
Stephane Nicoll
Juergen Hoeller
Sébastien Deleuze
Johnny Lim
Simon Baslé
K
Jan Mewes
Anyuruf