Looking for an example that shows integrating spring cloud sleuth with spring boot amqp (rabbit) publisher and subscriber.
I do see the following messages in the log
2016-10-21 08:35:15.708 INFO [producer,9148f56490e5742f,943ed050691842ab,false] 30928 --- [nio-8080-exec-1] a.b.c.controllers.MessagingController : Received Request to pulish with Activity OrderShipped
2016-10-21 08:35:15.730 INFO [producer,9148f56490e5742f,943ed050691842ab,false] 30928 --- [nio-8080-exec-1] a.b.c.service.ProducerService : Message published
When I look at messages on Queue, I don't see traceId or any other details added to the header. Should I use MessagePostProcessor to add these to the header?
Also what should be done on the receiving service?
We don't instrument Spring AMQP out of the box. You can however use Spring Integration or Spring Cloud Stream that we do support and then everything will work out of the box. If you need to use Spring AMQP for some reason you'll have to instrument the code yourself (and sends us a PR ;) ).
Using Spring AMQP you can set MessagePostProcessor on the RabbitTemplateusing the setBeforePublishPostProcessors method.
We implemented the org.springframework.amqp.core.MessagePostProcessor and Overrided the postProcessMessage method this way:
#Override
public org.springframework.amqp.core.Message postProcessMessage(org.springframework.amqp.core.Message message)
throws AmqpException {
MessagingMessageConverter converter = new MessagingMessageConverter();
MessageBuilder<?> mb = MessageBuilder.fromMessage((Message<?>) converter.fromMessage(message));
inject(tracer.getCurrentSpan(), mb);
return converter.toMessage(mb.build(), message.getMessageProperties());
}
The inject method can now set all the required headers on the message, and it will be passed to the rabbitMq with the changes.
You have a great example of how to implement such inject method in org.springframework.cloud.sleuth.instrument.messaging.MessagingSpanInjector
We are using v1.1.1 of spring-cloud-sleuth-stream so my example is based on this version, in next release(1.2) it will be easier.
Related
I have an angular application using a tracing library to trace each operation (user bouton click).
This application after SPA is loaded sends a list of traces in the request body to the backend microservice to log them.
In the backend microservice, I am using the spring boot 2.3.7, the spring cloud Hoxton.SR9 and logback 1.2.3.
This is the method:
#PostMapping("/traces")
public ResponseEntity<List<Trace>> addTrace(#RequestBody List<Trace> traces) {
traces.forEach(trace -> {
RtLog log = RtLog.builder.parentSpanId(trace.getParentId()).traceId(trace.getTraceId()).spanId(trace.getSpanId)).operationName(trace.getOperationName())
.businessJourney(trace.getBaggages().getBusinessJourney())
.componentId(trace.getBaggages().getComponentId()).componentType(trace.getBaggages().getComponentType())
.sessionId(trace.getBaggages().getSessionId()).userId(trace.getBaggages().getUserId())
.duration(trace.getDuration())
.time(!Null.isNullOrEmpty(datetime) ? datetime.format(DateTimeFormatter.ofPattern(TIMESTAMP_PATTERN))
: EMPTY)
.tags(trace.getTags()).build();
logger.info("spa_log", StructuredArguments.fields(log));
An example of traces send by the application:
The problem is when logging the information, the spring sleuth adds on the log another traceId and spanId that I don't need, and it causes conflict with the correct Ids.
So how can I override the traceId and spanId or is it possible to disable the tracing in this method?
There is the traceContext or brave Tracing but I found that I can builder a new trace but I didn't find a way to change the current trace.
I have two Java processes - which get spawned from the same Jar using different run configurations
Process A - Client UI component , Developed Using Spring bean xml based approach. No Spring Boot is there.
Process B - A new Springboot Based component , hosts REST End points.
Now from Process A , on various button click how can I call the REST end points on Process B using Feign Client.
Note - Since Process A is Spring XML based , right at the moment we can not convert that to Spring boot. Hence #EnableFeignClients can not be used to initialise the Feign Clients
So Two questions
1) If the above is possible how to do it ?
2) Till Process A is moved to Spring boot - is Feign still an easier option than spring REST template ?
Feign is a Java to HTTP client binder inspired by Retrofit, JAXRS-2.0, and WebSockets and you can easily use feign without spring boot. And Yes, feign still better option to use because Feign Simplify the HTTP API Clients using declarative way as Spring REST does.
1) Define http methods and endpoints in interface.
#Headers({"Content-Type: application/json"})
public interface NotificationClient {
#RequestLine("POST")
String notify(URI uri, #HeaderMap Map<String, Object> headers, NotificationBody body);
}
2) Create Feign client using Feign.builder() method.
Feign.builder()
.encoder(new JacksonEncoder())
.decoder(customDecoder())
.target(Target.EmptyTarget.create(NotificationClient.class));
There are various decoders available in feign to simplify your tasks.
You are able to just initialise Feign in any code (without spring) just like in the readme example:
public static void main(String... args) {
GitHub github = Feign.builder()
.decoder(new GsonDecoder())
.target(GitHub.class, "https://api.github.com");
...
}
Please take a look at the getting started guide: feign on github
I got a Spring Boot application making use of Spring Sleuth for tracing inter-service calls. Within that application a ScheduledExecutorService exists that performs http requests in a loop (pseudo-code below):
class HttpCaller implements Runnable {
public void run() {
performHttpCall();
// "loop"
executor.submit(this::run);
}
}
// start it once
scheduler.submit(new HttpCaller());
If I now have a look at the traces produced by Sleuth and stored in Zipkin I can see that all http calls are associated to a single Trace. Most likely because the trace context is handed over during the call to ScheduledExecutorService::submit.
How can I clear the current trace before starting the next iteration so that each http call will result in a new detached/orphaned trace?
If you're using Sleuth 2.0 you can call on the Tracer a method to create a new trace. In the older version of sleuth I guess what I'd do is to use an executor that is NOT a bean. That way you would lose the trace and it would get restarted at some point (by rest template or sth like that).
I am trying to integrate the Brave MySql Instrumentation into my Spring Boot 2.x service to automatically let its interceptor enrich my traces with spans concerning MySql-Queries.
The current Gradle-Dependencies are the following
compile 'io.zipkin.zipkin2:zipkin:2.4.5'
compile('io.zipkin.reporter2:zipkin-sender-okhttp3:2.3.1')
compile('io.zipkin.brave:brave-instrumentation-mysql:4.14.3')
compile('org.springframework.cloud:spring-cloud-starter-zipkin:2.0.0.M5')
I already configured Sleuth successfully to send traces concerning HTTP-Request to my Zipkin-Server and now I wanted to add some spans for each MySql-Query the service does.
The TracingConfiguration it this:
#Configuration
public class TracingConfiguration {
/** Configuration for how to send spans to Zipkin */
#Bean
Sender sender() {
return OkHttpSender.create("https://myzipkinserver.com/api/v2/spans");
}
/** Configuration for how to buffer spans into messages for Zipkin */
#Bean AsyncReporter<Span> spanReporter() {
return AsyncReporter.create(sender());
}
#Bean Tracing tracing(Reporter<Span> spanListener) {
return Tracing.newBuilder()
.spanReporter(spanReporter())
.build();
}
}
The Query-Interceptor works properly, but my problem now is that the spans are not added to the existing trace but each are added to a new one.
I guess its because of the creation of a new sender/reporter in the configuration, but I have not been able to reuse the existing one created by the Spring Boot Autoconfiguration.
That would moreover remove the necessity to redundantly define the Zipkin-Url (because it is already defined for Zipkin in my application.yml).
I already tried autowiring the Zipkin-Reporter to my Bean, but all I got is a SpanReporter - but the Brave-Tracer-Builder requries a Reporter<Span>
Do you have any advice for me how to properly wire things up?
Please use latest snapshots. Sleuth in latest snapshots uses brave internally so integration will be extremely simple.
Below is an interesting example of sending messages over MQTT with the standard outbound-channel-adapter (not the MQTT outbound adapter):
https://github.com/joshlong/spring-integration-mqtt
The authors implement their own message handler, and pass it to the adapter.
Now my question is: Is it possible to implement a custom message handler using the MQTT outbound adapter? Or is it only possible with the general outbound-channel-adapter of Spring Integration?
My objective is to obtain logs and handle callbacks from the Paho client, so I can for example handle connection errors, timeouts, etc...
Spring Integration 4.0 provides the MQTT module with MqttPahoMessageHandler as default implementation of AbstractMqttMessageHandler.
I'd say that you can extend from MqttPahoMessageHandler to achieve your MqttCallback wishes, but yes, you can use that custom MessageHandler implementation only from <int:outbound-channel-adapter ref="">.
The out-of-the-box <int-mqtt:outbound-channel-adapter> is just for population a bean for MqttPahoMessageHandler and you can't change that behaviour.
From other side, when you will start to do Spring Integration from JavaConfig you will get deal just only with classes, so there is no boundaries to restict you with custom tags:
#ServiceActivator(inputChannel = "sendToMqttChannel")
#Bean
public MessageHandler mqttHandler() {
return new MyMqttPahoMessageHandler();
}