Develop Reference

Spring Webflux doOnCancel is not invoked when an app is deployed to Kubernetes

I have a controller with an endpoint that provide a flux like this reported below.
When the app is deployed to kubernetes, methods doOnCancel and doOnTerminate are not invoked.
Locally instead, it works like a charm (when the tab of the browser is closed as instance).
#Slf4j
#RestController
public class TestController {
...
#GetMapping(value = "/test", produces = MediaType.TEXT_EVENT_STREAM_VALUE)
public Flux<String> testStream() {
log.info("Requested test streaming");
return mySink.asFlux()
.startWith("INIT TEST")
.doOnCancel(() -> log.info("On cancel"))
.doOnTerminate(() -> log.info("On terminate"));
}
...
}
2022-08-06 18:25:42.115 INFO 3685 --- [ main] com.wuase.sinkdemo.SinkDemoApplication : Starting SinkDemoApplication using Java 1.8.0_252 on aniello-pc with PID 3685 (/home/pc/eclipse-workspace/sink-demo/target/classes started by pc in /home/pc/eclipse-workspace/sink-demo)
2022-08-06 18:25:42.124 INFO 3685 --- [ main] com.wuase.sinkdemo.SinkDemoApplication : No active profile set, falling back to 1 default profile: "default"
2022-08-06 18:25:44.985 INFO 3685 --- [ main] o.s.b.web.embedded.netty.NettyWebServer : Netty started on port 8080
2022-08-06 18:25:45.018 INFO 3685 --- [ main] com.wuase.sinkdemo.SinkDemoApplication : Started SinkDemoApplication in 3.737 seconds (JVM running for 5.36)
2022-08-06 18:26:09.706 INFO 3685 --- [or-http-epoll-3] com.wuase.sinkdemo.TestController : Requested test streaming
2022-08-06 18:26:14.799 INFO 3685 --- [or-http-epoll-3] com.wuase.sinkdemo.TestController : On cancel
Has anyone encountred the same problem?
Any idea about that?

camel kafka route does not stay up

I am trying to use kafka with camel and set up the following route:
public class WorkflowEventConsumerRoute extends RouteBuilder {
private static final String KAFKA_ENDPOINT =
"kafka:payments-bus?brokers=localhost:9092";
...
#Override
public void configure() {
from(KAFKA_ENDPOINT)
.routeId(format(KAFKA_CONSUMER))
.to("mock:end);
}
}
When I start my spring boot application I can see the route gets started but immediately after this it shuts down without any reasons given in the logs:
2018-12-21 12:06:45.012 INFO 12184 --- [ main] o.a.kafka.common.utils.AppInfoParser : Kafka version : 2.0.1
2018-12-21 12:06:45.013 INFO 12184 --- [ main] o.a.kafka.common.utils.AppInfoParser : Kafka commitId : fa14705e51bd2ce5
2018-12-21 12:06:45.014 INFO 12184 --- [ main] o.a.camel.spring.SpringCamelContext : Route: kafka-consumer started and consuming from: kafka://payments-bus?brokers=localhost%3A9092
2018-12-21 12:06:45.015 INFO 12184 --- [r[payments-bus]] o.a.camel.component.kafka.KafkaConsumer : Subscribing payments-bus-Thread 0 to topic payments-bus
2018-12-21 12:06:45.015 INFO 12184 --- [ main] o.a.camel.spring.SpringCamelContext : Total 1 routes, of which 1 are started
2018-12-21 12:06:45.015 INFO 12184 --- [ main] o.a.camel.spring.SpringCamelContext : Apache Camel 2.23.0 (CamelContext: camel-1) started in 0.234 seconds
2018-12-21 12:06:45.019 INFO 12184 --- [ main] a.c.n.t.p.workflow.WorkflowApplication : Started WorkflowApplication in 3.815 seconds (JVM running for 4.513)
2018-12-21 12:06:45.024 INFO 12184 --- [ Thread-10] o.a.camel.spring.SpringCamelContext : Apache Camel 2.23.0 (CamelContext: camel-1) is shutting down
On the other hand if create an unit test and point to the same kafka endpoint I am able to read the kafka topic content using the org.apache.camel.ConsumerTemplate instance provided by the CamelTestSupport
Ultimately if I replace the kafka endpoint in my route with an activemq one the route starts OK and the application stays up.
Obviously I am missing something but I cannot figure out what.
Thank you in advance for your help.

Do your spring-boot app have a -web-starter or not. If not then you should turn on the camel run-controller to keep the boot application running.
In the application.properties add
camel.springboot.main-run-controller = true

What causes Spring scheduler to execute before the application and servlet starts up?

I have a simple Spring-Boot web application with some scheduling tasks:
#SpringBootApplication
#EnableScheduling
public class Application extends SpringBootServletInitializer {
#Override
protected SpringApplicationBuilder configure(final SpringApplicationBuilder app) {
return app.sources(Application.class);
}
public static void main(String[] args) {
SpringApplication.run(Application.class, args);
}
}
And a service executing two scheduling tasks:
#Service
public class SchedulingService {
#Scheduled(fixedRate = 15000)
private void first() {
// log first
}
#Scheduled(fixedRate = 6000)
public void second() {
// log second
}
}
Is there any explanation revealing why one of the scheduling tasks has been executed before the application and servlet starts up? This happens repeatedly and the log order is always the same, so I suspect the multithreading doesn't take place - there is nearly one-second difference between the first and the second scheduling task.
Here is the log with highlighted parts of the scheduling task log:
2018-08-18 20:47:53.085 INFO 251168 --- [ost-startStop-1] o.s.b.a.w.s.WelcomePageHandlerMapping : Adding welcome page template: index
2018-08-18 20:47:53.300 INFO 251168 --- [ost-startStop-1] o.s.j.e.a.AnnotationMBeanExporter : Registering beans for JMX exposure on startup
2018-08-18 20:47:53.314 INFO 251168 --- [ost-startStop-1] s.a.ScheduledAnnotationBeanPostProcessor : No TaskScheduler/ScheduledExecutorService bean found for scheduled processing
2018-08-18 20:47:53.321 HERE LOGGED FIRST
2018-08-18 20:47:53.344 INFO 251168 --- [ost-startStop-1] my.appname.Application : Started Application in 5.565 seconds (JVM running for 16.93)
2018-08-18 20:47:53.396 INFO 251168 --- [ main] org.apache.coyote.ajp.AjpNioProtocol : Starting ProtocolHandler ["ajp-nio-8009"]
2018-08-18 20:47:53.400 INFO 251168 --- [ main] org.apache.catalina.startup.Catalina : Server startup in 15970 ms
2018-08-18 20:47:53.477 INFO 251168 --- [nio-8080-exec-1] o.a.c.c.C.[.[localhost].[/Rattle] : Initializing Spring FrameworkServlet 'dispatcherServlet'
2018-08-18 20:47:53.477 INFO 251168 --- [nio-8080-exec-1] o.s.web.servlet.DispatcherServlet : FrameworkServlet 'dispatcherServlet': initialization started
2018-08-18 20:47:53.520 INFO 251168 --- [nio-8080-exec-1] o.s.web.servlet.DispatcherServlet : FrameworkServlet 'dispatcherServlet': initialization completed in 43 ms
2018-08-18 20:47:54.103 HERE LOGGED SECOND
2018-08-18 20:47:59.335 HERE LOGGED SECOND
2018-08-18 20:48:05.334 HERE LOGGED SECOND
2018-08-18 20:48:08.334 HERE LOGGED FIRST

I just created an empty SpringBoot project, added your class and tried to reproduce the same issue.
On SpringBoot 2.0.4.RELEASE I see that both tasks run at the same time and same thread:
2018-08-18 21:16:54.145 INFO 10239 --- [pool-1-thread-1] com.test.SchedulingService : LOG FIRST
2018-08-18 21:16:54.145 INFO 10239 --- [pool-1-thread-1] com.test.SchedulingService : LOG SECOND
Then, I added a Thread.sleep(100) on the both tasks as I cannot predict the execution order.
2018-08-18 21:21:14.775 INFO 10274 --- [pool-1-thread-1] com.test.SchedulingService : LOG FIRST
2018-08-18 21:21:14.878 INFO 10274 --- [pool-1-thread-1] com.test.SchedulingService : LOG SECOND
The delay of aprox 100ms between the log entries, confirm that they run on the same thread.
You might think that setting the initialDelay = 0 on the #Scheduled might help, but it will not; everything will still be on the main thread.
The solution I found is defining a custom bean that creates a customTaskScheduler:
#Bean()
public ThreadPoolTaskScheduler taskScheduler(){
ThreadPoolTaskScheduler taskScheduler = new ThreadPoolTaskScheduler();
taskScheduler.setPoolSize(2);
taskScheduler.setThreadNamePrefix("Async-");
return taskScheduler;
}
Now, the logs show that both tasks are executed at the same time and by different threads:
2018-08-18 21:30:26.482 INFO 10383 --- [ main] o.s.b.w.embedded.tomcat.TomcatWebServer : Tomcat started on port(s): 8080 (http) with context path ''
2018-08-18 21:30:26.486 INFO 10383 --- [ main] com.test.ProjectApplication : Started ProjectApplication in 1.767 seconds (JVM running for 2.137)
2018-08-18 21:30:26.555 INFO 10383 --- [ Async-2] com.test.SchedulingService : LOG SECOND
2018-08-18 21:30:26.555 INFO 10383 --- [ Async-1] com.test.SchedulingService : LOG FIRST
I defined a thread pool of size = 2. What will happen if I have 3 tasks to run? One of the tasks will need to wait to complete the execution of the previous 2, get the thread released and check the execution queue again.
2018-08-18 21:33:48.895 INFO 10412 --- [ main] o.s.b.w.embedded.tomcat.TomcatWebServer : Tomcat started on port(s): 8080 (http) with context path ''
2018-08-18 21:33:48.899 INFO 10412 --- [ main] com.test.ProjectApplication : Started ProjectApplication in 1.888 seconds (JVM running for 2.258)
2018-08-18 21:33:48.960 INFO 10412 --- [ Async-1] com.test.SchedulingService : LOG FIRST
2018-08-18 21:33:48.960 INFO 10412 --- [ Async-2] com.test.SchedulingService : LOG SECOND
2018-08-18 21:33:49.065 INFO 10412 --- [ Async-2] com.test.SchedulingService : LOG THIRD

I had same issue, but I was able to get around it by specifying the cron parameter instead of using fixedRate or fixedRateString like so #Scheduled(cron = "0 0 4 ? * SUN,THU")
More details on using the cron expression in spring-boot can be found here https://docs.spring.io/spring-framework/docs/current/javadoc-api/org/springframework/scheduling/support/CronExpression.html

Hazelcast web clustered session service: Retrying the connection

hazelcast unable to connect, the message receive is as follows
2018-03-03 10:27:51,074 INFO c.h.i.DefaultAddressPicker [LOCAL] [dev] [3.6] Picked Address[127.0.0.1]:5703, using socket ServerSocket[addr=/0:0:0:0:0:0:0:0,localport=5703], bind any local is true
2018-03-03 10:28:01.078 INFO 19478 --- [.ensureInstance] c.hazelcast.web.ClusteredSessionService : Retrying the connection!!
2018-03-03 10:28:01,078 INFO c.h.w.ClusteredSessionService Retrying the connection!!
2018-03-03 10:28:01.079 INFO 19478 --- [.ensureInstance] com.hazelcast.config.XmlConfigLocator : Loading 'hazelcast-default.xml' from classpath.
2018-03-03 10:28:01,079 INFO c.h.c.XmlConfigLocator Loading 'hazelcast-default.xml' from classpath.
2018-03-03 10:28:01.085 INFO 19478 --- [.ensureInstance] c.hazelcast.web.HazelcastInstanceLoader : Creating a new HazelcastInstance for session replication
2018-03-03 10:28:01,085 INFO c.h.w.HazelcastInstanceLoader Creating a new HazelcastInstance for session replication
2018-03-03 10:28:01.086 INFO 19478 --- [.ensureInstance] c.h.instance.DefaultAddressPicker : [LOCAL] [dev] [3.6] Picked Address[127.0.0.1]:5703, using socket ServerSocket[addr=/0:0:0:0:0:0:0:0,localport=5703], bind any local is true

Flux subscribed through SSE raise a cancel() event

I have a Spring Boot 2.0.0.M7 + Spring Webflux application in which I am using Thymeleaf Reactive.
I noticed that on my microservices, when I call an endpoint returning a flux of data in SSE mode (text/event-stream), a cancel() occurs on this flux even if it has been processed correctly.
For example, here's a simple controller endpoint:
#GetMapping(value = "/posts")
public Flux<String> getCommunityPosts() {
return Flux.just("A", "B", "C").log("POSTS");
}
And here's the subscribed flux logs I get when I request it in SSE mode:
2018-02-13 17:04:09.841 INFO 4281 --- [nio-9090-exec-4] POSTS : | onSubscribe([Synchronous Fuseable] FluxArray.ArraySubscription)
2018-02-13 17:04:09.841 INFO 4281 --- [nio-9090-exec-4] POSTS : | request(1)
2018-02-13 17:04:09.842 INFO 4281 --- [nio-9090-exec-4] POSTS : | onNext(A)
2018-02-13 17:04:09.847 INFO 4281 --- [nio-9090-exec-4] POSTS : | request(1)
2018-02-13 17:04:09.847 INFO 4281 --- [nio-9090-exec-4] POSTS : | onNext(B)
2018-02-13 17:04:09.848 INFO 4281 --- [nio-9090-exec-4] POSTS : | request(1)
2018-02-13 17:04:09.848 INFO 4281 --- [nio-9090-exec-4] POSTS : | onNext(C)
2018-02-13 17:04:09.849 INFO 4281 --- [nio-9090-exec-4] POSTS : | request(1)
2018-02-13 17:04:09.849 INFO 4281 --- [nio-9090-exec-4] POSTS : | onComplete()
2018-02-13 17:04:09.852 INFO 4281 --- [nio-9090-exec-4] POSTS : | cancel()
We can notice the cancel event after the onComplete. I don't have this behaviour when I call the same endpoint through a classic GET request. I suspect this cancel event to make the client side event source (javascript) throw a onError event.
Is it a known/wanted behaviour specific to SSE?
QUESTION UPDATE
I actually use SSE on some of my streams because I sometimes need my event sources to get JSON data instead of HTML already processed by Thymeleaf. Should I do it in another way?
I based my implementation on the last method of this example: https://github.com/danielfernandez/reactive-matchday/blob/master/src/main/java/com/github/danielfernandez/matchday/web/controller/MatchController.java
However, I may have missed providing some information in my previous post. I use Tomcat Server (8.5.23 with M7), and not Netty server. I forced Tomcat use including the following Maven dependency:
<dependency>
<groupId>org.springframework.boot</groupId>
<artifactId>spring-boot-starter-tomcat</artifactId>
</dependency>
Using your code on a sample project, this seems to cause the issue.
When I run the code on a Netty server, I get the same results as you:
2018-02-14 12:30:48.713 INFO 3060 --- [ctor-http-nio-2] reactor.Flux.ConcatMap.1 : onSubscribe(FluxConcatMap.ConcatMapImmediate)
2018-02-14 12:30:48.714 INFO 3060 --- [ctor-http-nio-2] reactor.Flux.ConcatMap.1 : request(1)
2018-02-14 12:30:49.717 INFO 3060 --- [ parallel-2] reactor.Flux.ConcatMap.1 : onNext(a)
2018-02-14 12:30:49.739 INFO 3060 --- [ctor-http-nio-2] reactor.Flux.ConcatMap.1 : request(31)
2018-02-14 12:30:50.731 INFO 3060 --- [ parallel-3] reactor.Flux.ConcatMap.1 : onNext(b)
2018-02-14 12:30:51.733 INFO 3060 --- [ parallel-4] reactor.Flux.ConcatMap.1 : onNext(c)
2018-02-14 12:30:51.735 INFO 3060 --- [ parallel-4] reactor.Flux.ConcatMap.1 : onComplete()
When I run the same code on the Tomcat server, I have the cancel issue:
2018-02-14 12:33:18.294 INFO 3088 --- [nio-8080-exec-3] reactor.Flux.ConcatMap.2 : onSubscribe(FluxConcatMap.ConcatMapImmediate)
2018-02-14 12:33:18.295 INFO 3088 --- [nio-8080-exec-3] reactor.Flux.ConcatMap.2 : request(1)
2018-02-14 12:33:19.295 INFO 3088 --- [ parallel-4] reactor.Flux.ConcatMap.2 : onNext(a)
2018-02-14 12:33:19.297 INFO 3088 --- [ parallel-4] reactor.Flux.ConcatMap.2 : request(1)
2018-02-14 12:33:20.302 INFO 3088 --- [ parallel-5] reactor.Flux.ConcatMap.2 : onNext(b)
2018-02-14 12:33:20.302 INFO 3088 --- [ parallel-5] reactor.Flux.ConcatMap.2 : request(1)
2018-02-14 12:33:21.306 INFO 3088 --- [ parallel-6] reactor.Flux.ConcatMap.2 : onNext(c)
2018-02-14 12:33:21.306 INFO 3088 --- [ parallel-6] reactor.Flux.ConcatMap.2 : request(1)
2018-02-14 12:33:21.307 INFO 3088 --- [ parallel-6] reactor.Flux.ConcatMap.2 : onComplete()
2018-02-14 12:33:21.307 INFO 3088 --- [nio-8080-exec-4] reactor.Flux.ConcatMap.2 : cancel()
Could it be a Tomcat issue or am I doing something wrong?

First, I don't think you should use SSE for finite streams.
When I create a Controller method like:
#GetMapping(path = "/test", produces = MediaType.TEXT_EVENT_STREAM_VALUE)
#ResponseBody
public Flux<String> test() {
return Flux.just("a", "b", "c").delayElements(Duration.ofSeconds(1)).log();
}
and request it from a browser (Chrome or Firefox) with:
<script type="text/javascript">
var testEventSource = new EventSource("/test");
testEventSource.onmessage = function (e) {
console.log(e);
};
</script>
I get the following logs on the server:
| onSubscribe([Fuseable] FluxOnAssembly.OnAssemblySubscriber)
| request(1)
| onNext(a)
| request(31)
| onNext(b)
| onNext(c)
| onComplete()
| onSubscribe([Fuseable] FluxOnAssembly.OnAssemblySubscriber)
| request(1)
| onNext(a)
| request(31)
| onNext(b)
| onNext(c)
| onComplete()
As soon as the Flux is completed, the connection is closed by the server and the browser reconnects automatically. This will replay the same sequence over and over again.
The only way I get a cancel() event on the server is when I close the browser tab during the stream.