https://build.spring.io/browse/INT-FATS5IC-430
Looks like there is a race condition when our polling rate around
data base is too fast and we have access to the DB files even during
application context close.
* Stop polling channel adapter in the tests explicitly after test methods
* Increase some tests performance decreasing timeouts to wait
* Upgrade to Reactor-3.1.5, AssertJ-3.9.1, Derby-10.14.1.0 and some
Gradle plugins
* Upgrade to `reactor-netty-0.7.5`
* Remove workaround from the `StompServerIntegrationTests`
* Upgrade to Spring Data Kay SR5
* Upgrade to Spring Security 5.0.3
* Increase timeouts and performance in the `StreamTransformerParserTests`
JIRA: https://jira.spring.io/browse/INT-4420
Use `hamcrest-core` and `hamcrest-library` in place of `hamcrest-all` to match
behavior of other spring projects and avoid multiple JARs containing
the same classes.
https://build.spring.io/browse/INT-MASTER-952
The SSL error might be like `Connection reset by peer` as well.
Also move `client` and `server` `stop()` to the finally, since it looks
like in case of handshake error we leave dangling opened socket
If there is `json`-aware `contentType` header in the message to process
and payload type is `String` or `byte[]` and it isn't equal to the
expected exclusive method argument type (even generic from the `Message<>`),
the conversion should happen only if SpelInvoker is no method.
The regular `InvocableHandlerMethod` takes care about payload conversion
via configured `MessageConverter` in the `PayloadArgumentResolver`
* Fix `MessagingMethodInvokerHelper` to call JSON conversion only when
`this.handlerMethod.spelOnly`
NOTE: We can't do generics conversion right now because it is going
to be a breaking change around
`org.springframework.integration.support.json.JsonObjectMapper` property
in the `MessagingMethodInvokerHelper`.
We can do that only in `5.1` and rely there only on the functionality
from the Jackson Object mapper
* Convert from JSON only in the `invokeExpression()`
* Optimize message recreation only if target param type is `Message`.
Otherwise override just `payload` property of the `ParametersWrapper`
5.0 introduced the `TcpNetConnectionSupport` user hook to create connection objects.
If a user-supplied instance threw an exception, the server socket would remain open
but the server `accept()` thread is gone.
- wrap the connection initialization code in try/catch
- close the server socket if necessary on an exception
JIRA: https://jira.spring.io/browse/INT-4413
When we override entity in the `IntegrationFlowContext.registry`,
we may get dangling beans in the application context,
when the structure of a new `IntegrationFlow` is different.
* Fix `IntegrationFlowContext` to disallow to override the flow
registration under the same name
* Add JavaDocs to the `IntegrationFlowRegistrationBuilder` methods
* Add `toString()` to the `IntegrationFlowRegistration` and
`StandardIntegrationFlow` to make the logging of these components
friendlier
JIRA: https://jira.spring.io/browse/INT-4411
The `RouterSpec.RouterMappingProvider` relies on the
`ContextRefreshedEvent` which happens only during application start up.
When we register `IntegrationFlow` at runtime, this event doesn't
happen and therefore sub-flow mappings don't populated.
* Fix `RouterSpec.RouterMappingProvider` to parse sub-flow mappings in
the `onInit()`
* Reorder components registration for the router in the
`IntegrationFlowDefinition` to let lately sub-flows to start earlier,
then lifecycles in the main flow
JIRA: https://jira.spring.io/browse/INT-4407
To make a `Log4j2LevelAdjuster` more builder friendly, add a `merge`
option to the overloaded `classes()` and `categories()` methods.
This way the existing (or previously configured) `categories` and
`classes` are merged with the provided values and default
`org.springframework.integration` category can be reuse.
Note: this behavior can be reconsidered as default one in the `5.1`
JIRA: https://jira.spring.io/browse/INT-4409
The `SoapMessage` may come as a `payload` and be already with the
`soapAction` header.
Right now it is overridden independently of the value from the headers.
* Do not override `soapAction` in the `SoapMessage` if header doesn't
have value
JIRA: https://jira.spring.io/browse/INT-4410
The `CollectionArgumentResolver` has been introduced especially for the
cases to work with `MessageGroupProcessor` (an aggregator)
when the payload is a `Collection<Message<?>>`.
This use-case doesn't apply for the general collection parameter use-case.
* Register `CollectionArgumentResolver` only when `listCapable` option.
For all other collection-based use-cases fallback to the standard
`PayloadArgumentResolver` with an appropriate configured `MessageConverter`
Note: this is for backward compatibility.
In the `5.1` we may reconsider to use `MessageConverter` in the
`CollectionArgumentResolver` as well, or just remove it altogether with
an appropriate logic in the `PayloadArgumentResolver`, since this
`CollectionArgumentResolver` solution isn't robust
JIRA: https://jira.spring.io/browse/INT-4402
Make `GlobalChannelInterceptorProcessor` as a `BeanPostProcessor`,
so it can apply global `ChannelInterceptor`s to any initialized channel
beans, even those created at runtime, like in case of dynamic flows with
Java DSL
**Cherry-pick to 4.3.x**
* Add `postProcessDynamicBeans` global property with `false` by default
* Rely on the `postProcessDynamicBeans` property in the
`GlobalChannelInterceptorProcessor.postProcessAfterInitialization()`
* Add `postProcessDynamicBeans=true` to the
`GlobalChannelInterceptorTests-context.xml` to be sure in the test-case
that option works
* Document the option and effect from the global channel interceptors
JIRA: https://jira.spring.io/browse/INT-4402
Previously, Micrometer instrumentation was only applied to components during
ApplicationContext initialization.
`IntegrationManagementConfigurer` is now a `BeanPostProcessor` as well as a
`SmartInitializingSingleton`, but only acts as a BPP after the context is
initialized. This enables Micrometer instrumentation to beans added later,
either via a new `BeanDefinition` or `bf.initializeBean`.
NOTE: destroying and re-creating a bean will use the same `Meters`.
Fixes https://github.com/spring-projects/spring-integration/issues/2358
The WebFlux on the server side can deal with MVC `@Controllers` as well
without any Servlet environment.
* Enable `IntegrationGraphController` when we not only in the
Servlet environment, but also in the WebFlux
* Upgrade to SF-5.0.4 and fix WebFlux tests to meet the latest requirements
(https://jira.spring.io/browse/SPR-16337)
* Move `WEB_FLUX_PRESENT` to the `HttpContextUtils`;
deprecate it in the `WebFluxContextUtils`
* Mention WebFlux ability for the `IntegrationGraphController` in the
`graph.adoc`
JIRA: https://jira.spring.io/browse/INT-4396
When target distributed `Lock` implementation throws an exception, e.g.
in case of no connection to the service, the `LockRegistryLeaderInitiator`
exists the loop and can come back to the elections only after restart
* Catch all the exception on `this.lock.tryLock()` and resubmit
`LeaderSelector` for a new locking cycle if
`LockRegistryLeaderInitiator.isRunning()` and `InterruptedException`
* Remove diagnostics from the `JdbcLockRegistryLeaderInitiatorTests`
since this fix confirms that we just didn't have a reconnect logic before
when this test failed sporadically
**Cherry-pick to 4.3.x**
JIRA: https://jira.spring.io/browse/INT-4394
Looks like Jackson has became much smarter and now it stores
the target type for the object, not returned interface.
We can't use Jackson annotations on the Framework classes and
we can't restrict the `messagingAwareMapper()` `ObjectMapper` just
to use fields for all the object passed through it.
As a compromise solution we shouldn't use
`Collections.unmodifiableList()` in the getter.
The copy of the internal collection is pretty enough to protect
our object from mutation.
**Cherry-pick to 4.3.x (excluding `build.gradle` change)**
https://build.spring.io/browse/INT-FATS5IC-400
When we run several concurrent builds (e.g. CI) and use the same shared
Redis server we may end up with the case when one process reads data
populated by another because we use the same key (groupId in our case)
* Fix `RedisMessageGroupStoreTests` to use `UUID.randomUUID()`
for the `groupId`
**Cherry-pick to 4.3.x**
https://build.spring.io/browse/INT-MJATS41-1242
When different `DefaultLockRepository` instances use the same client id,
there is a possibility that they will update the same row in the table.
This way we have a chance that not only one obtains a lock and try to
add a value to the collection.
* Remove the test-case for the same client id as non-stable and even
dangerous by the the distributed lock purpose
* Use `tryLock(Long.MAX_VALUE)` to really ensure the wait behavior
during the concurrent loop
* Use `20` for thread pool to align with the tasks amount
**Cherry-picked to 4.3.x**
https://build.spring.io/browse/INT-SI43X-230
Looks like Ruby engine takes some time to start and
`Jsr223InboundChannelAdapterTests` is slow as 3 seconds locally as well
* Change the script to JavaScript and now we have 200 millis
**Cherry-pick to 4.3.x**
