The `ArrayList` cannot be used in the test-case for checking
its size because the property is not `volatile`: even if
content of the list is what we expect (test logs confirms that),
the non-`volatile` property doesn't give us an actual value
* Use `CopyOnWriteArrayList` for this concurrent test to
be sure that internal `array.length` reflects the reality
after each interaction with the list
In preparation for https://github.com/spring-cloud/spring-cloud-stream-binder-rabbit/issues/303
Now that `CorrelationData` has a `Future<?>`, users might simply add
correlation data in a header and not receive confirm/return messages.
- No longer require channels for returns and confirms
- don't build the confirm message if there are no channels
- reduce the log level for no channels to DEBUG
- complete the user's future when a message is returned (async GW)
* Fix typo.
* Add gauges for queue channel size
The `QueueChannel` provides a current size and remaining capacity metrics
* Add Micrometer gauges into `QueueChannel` to expose the current values
of the size and remaining capacity
**Cherry-pick to 5.3.x, 5.2.x & 5.1.x**
* * Revert `@SuppressWarnings("unchecked")` for test
* Document new gauges for queue channel
* * Fix IntegrationManagementConfigurer for NPE on `metricsCaptor`
* Fix wording in meter descriptions
Co-authored-by: Michel Jung <michel.jung89@gmail.com>
Co-authored-by: Michel Jung <michel.jung89@gmail.com>
* Restore `Mono<RSocketRequester> getRSocketRequester()` on
`ClientRSocketConnector` and deprecate it in favor of newly introduced
`RSocketRequester getRequester()`: there is no need in wrapping a
`RSocketRequester` to deferred `Mono` since internal logic of the
`RSocketRequester` on client side is based on lazy-load API of the
`RSocketClient`
Since `RSocketRequester` is now lazy load on client side
there is no need to wrap it into a `Mono`.
* Change `getRSocketRequester()` to a plain getter
If there is a requirement to force connect to the server for receiving
requests from there, the `ClientRSocketConnector.connect()` should be used
Resolves https://github.com/spring-projects/spring-integration/issues/3340
- add common super-interface for MQTT components
- add `getConnectionInfo()` so users can examine server URIs etc
* Reinstate per-adapter URIs - support multiple
* Restore single URL per adapter.
* Code cleanup for previous commit.
* GH-3334: Add "embedded reaper" into CorrelationMH
Fixes https://github.com/spring-projects/spring-integration/issues/3334
* Add `expireTimeout` property into `AbstractCorrelatingMessageHandler`
to call newly introduced `purgeOrphanedGroups()` API for removing old
groups from the store
* Add `expireDuration` to perform `purgeOrphanedGroups()` task periodically
* * Add Java DSL and XML support for `expireTimeout` and `expireDuration` options
* Document the new feature
* * Fix language in docs
Co-authored-by: Gary Russell <grussell@vmware.com>
Co-authored-by: Gary Russell <grussell@vmware.com>
* GH-3344: Treat kotlin.Unit return as null in MMIH
Fixes https://github.com/spring-projects/spring-integration/issues/3344
When function lambda doesn't return anything (e.g. a `void` method call is the last one),
Kotlin produces a `kotlin.Unit` instance as a return value which is not null and produced
as a reply message payload.
* Fix `MessagingMethodInvokerHelper` to treat a `kotlin.Unit` as `null` for reply
making Kotlin lambdas working the same way as Java lambdas when we don't return anything
from from there
**Cherry-pick to `5.3.x`**
* * Introduce `ClassUtils.isKotlinUnit(Class)` API;
use it in the `MessagingMethodInvokerHelper` instead of
`.getName().equals()`
* * Fix since on new `isKotlinUnit()` API
* Fix HTTP & WebFlux for a new CORS logic when `*` pattern
is used together with `allowCredentials`
* Fix RSocket module for deprecated API in SF
* The deprecated RMI API is left for the upcoming changes
Thanks to the pair of `CLIENT_ID`/`CREATED_DATE`,
it is possible to know if the lock is expired, already held or can be reacquire.
I think that the pre delete row is unnecessary and redundant with these two columns.
The rows doesn't need to be purged because the number of row in the table is finite and equal to the number of locks.
At worst, the table will be able to be purged by `deleteExpired()`.
I propose to overwrite these two columns at acquire lock time instead of pre delete the row.
So `delete` then `update` or `insert` become `update` or `insert`.
One client held a lock.
- Another client will be able to hold the same lock only if:
1. the row doesn't exist. Done when the first client will call `unlock()`.
2. the lock expire in case of hardware shutdown. Done by "`OR CREATED_DATE<?`" in `updateQuery`.
- The same client will be able to reacquire it only if:
1. the row doesn't exist.
2. the lock is already held by him. Done by "`CLIENT_ID=? OR`" in `updateQuery`.
**Cost with PostgreSQL:**
```
EXPLAIN DELETE FROM INT_LOCK WHERE REGION='DEFAULT' AND LOCK_KEY='FOO' AND CREATED_DATE<'2020-07-14';
Delete on int_lock (cost=0.14..8.17 rows=1 width=6)
-> Index Scan using int_lock_pk on int_lock (cost=0.14..8.17 rows=1 width=6)
Index Cond: (((region)::text = 'DEFAULT'::text) AND (lock_key = 'FOO'::bpchar))
Filter: (created_date < '2020-07-14 00:00:00'::timestamp without time zone)
EXPLAIN UPDATE INT_LOCK SET CREATED_DATE='2020-07-15' WHERE REGION='DEFAULT' AND LOCK_KEY='FOO' AND CLIENT_ID=NULL;
Update on int_lock (cost=0.00..11.40 rows=1 width=520)
-> Result (cost=0.00..11.40 rows=1 width=520)
One-Time Filter: false
-> Seq Scan on int_lock (cost=0.00..11.40 rows=1 width=520)
```
```
EXPLAIN UPDATE INT_LOCK SET CLIENT_ID=NULL, CREATED_DATE='2020-07-15' WHERE REGION='DEFAULT' AND LOCK_KEY='FOO' AND (CLIENT_ID=NULL OR CREATED_DATE<'2020-07-14')
Update on int_lock (cost=0.14..8.17 rows=1 width=372)
-> Index Scan using int_lock_pk on int_lock (cost=0.14..8.17 rows=1 width=372)
Index Cond: (((region)::text = 'DEFAULT'::text) AND (lock_key = 'FOO'::bpchar))
Filter: (created_date < '2020-07-14 00:00:00'::timestamp without time zone)
```
Related to https://stackoverflow.com/questions/62761903/spring-integration-reactive-streams-support-exception-in-creating-a-reactive
The `MessagingGatewaySupport` creates an internal endpoint
for consuming messages from the provided `replyChannel`.
The endpoint type depends on the channel type.
The `ReactiveStreamsConsumer` was missing a `BeanFactory` injection
causing an error when `ReactiveStreamsConsumer` tries to extract a default `ErrorHandler`
from `BeanFactory`
* Refactor `MessagingGatewaySupport` to inject a `BeanFactory` to all
the possible correlator endpoints.
Also always call `afterPropertiesSet()` for all of them
**Cherry-pick to 5.3.x & 5.2.x**
Close the pool so that any sessions returned after the factory is
`destroy()`ed are closed.
* Call `removeAllIdleItems()` in `close()`.
* Close sessions in `SftpStreamingMessageSourceTests`.
**cherry-pick to all supported branches**
Fixes https://github.com/spring-projects/spring-integration/issues/3272
* Introduce a `RenewableLockRegistry` since not all `LockRegistry` implementations
provide a way to renew the lease for the lock
* Implement `RenewableLockRegistry` in the `JdbcLockRegistry`
* Test and document the feature
Related to https://build.spring.io/browse/INT-MASTERSPRING40-1129
When there is no other candidates in the cluster, the current `yield()`
will give a leadership to the current active initiator back
Therefore a `revoke()` might not be called at all
* Fir the `RedisLockRegistryLeaderInitiatorTests` to check revoking
after `stop()` instead of `yield()`
* GH-3320: Refine lifecycle control in StdIntFlow
Fixes https://github.com/spring-projects/spring-integration/issues/3320
Turns out that lifecycle control for the whole bunch of components
in one `IntegrationFlow` is useful in fields.
* Change the logic in the `StandardIntegrationFlow` to let to call
`start()` and `stop()` independently how the flow was registered in
the application context.
This way it can be autowired as a `Lifecycle` to let end-user to
avoid the search for proper component in the flow to stop or start
manually - all the components registered with the flow are going
to be stopped or started respectively
* * Add `this.` to class property usage to satisfy Checkstyle
Co-authored-by: Artem Bilan <abilan@vmware.com>
