Wingtips is a distributed tracing solution for Java 7 and greater based on the Google Dapper paper.
This submodule contains a sample application based on Spring Boot 2 with WebFlux (not Web MVC) that uses
WingtipsWithZipkinSpringBoot2WebfluxConfiguration
from
wingtips-zipkin2-spring-boot2-webflux
to setup both Wingtips'
WingtipsSpringWebfluxWebFilter
(to automatically start and complete the overall request span for incoming requests)
and Wingtips' Zipkin integration (to send completed Wingtips spans to a
Zipkin server), all configured from the sample's application.properties
. As always with
WingtipsSpringWebfluxWebFilter
, if the incoming request contains tracing headers then they will be used as the
parent span, otherwise a new trace will be started.
This sample also shows the other half of the equation
(the Propagating Distributed Traces Across Network or Application Boundaries
section of the main Wingtips readme). Specifically it shows how to use the interceptor from the
wingtips-spring-webflux
module (WingtipsSpringWebfluxExchangeFilterFunction
) to
make Spring's reactive HTTP client WebClient
automatically surround HTTP client requests in a separate subspan and
propagate the tracing info in the downstream call request headers.
There are also a few examples showing how to make tracing state hop threads when you do asynchronous processing.
- Build the sample by running the
./buildSample.sh
script. - Launch the sample by running the
./runSample.sh
script. It will bind to port 8080 by default.- You can override the default port by passing in a system property to the run script,
e.g. to bind to port 8181:
./runSample.sh -Dserver.port=8181
- You can override the default port by passing in a system property to the run script,
e.g. to bind to port 8181:
Part of what this sample application does is show how to integrate a Wingtips-enabled Spring Boot server with Zipkin. In order to see the results of this integration you'll need a locally-running Zipkin server. See the Zipkin quickstart page to learn how to download and launch the Zipkin server - it should only take a few seconds.
Once your local Zipkin server is running and you've hit a few of the sample application's endpoints (and/or executed
VerifySampleEndpointsComponentTest
) you can open http://localhost:9411 in a browser and
search for traces.
If you don't launch a locally-running Zipkin server then the sample application will still function normally - you just won't be able to visualize the span data.
All examples here assume the sample app is running on port 8080, so you would hit each path by going to
http://localhost:8080/[endpoint-path]
. It's recommended that you use a REST client like
Postman for making the requests so you can easily specify HTTP method, payloads,
headers, etc, and fully inspect the response.
Also note that all the following things to try are verified in a component test: VerifySampleEndpointsComponentTest
.
If you prefer to experiment via code you can run, debug, and otherwise explore that test.
As you are doing the following you should check the logs that are output by the sample application and notice what is included in the log messages. In particular notice how you can search for a specific trace ID that came back in the response headers and find all the relevant log message for that request in the logs. Additionally if you launch a local Zipkin server you can open http://localhost:9411 in a browser and search for traces.
- For all of the following things to try, you can specify
X-B3-TraceId
andX-B3-SpanId
headers to cause the server to use those values as parent span information. Try sending requests with and without these headers to see how it affects the resulting server logs. If you are sending your own trace and span IDs you can also optionally send aX-B3-Sampled
header with a value of0
to disable the[DISTRIBUTED_TRACING]
log for that request. Finally, this sample is configured to treatuserid
andaltuserid
headers as "User ID Header Keys" and populate the Span's user ID when one of those headers are found. GET /sample/simple
- A basic blocking endpoint that returns as quickly as possible. The[DISTRIBUTED_TRACING]
log message for this endpoint should have very shortdurationNanos
- note that because the duration is nanosecond precision you can accurately time endpoints that return in well under 1 millisecond. You should see a log message output by the endpoint that is auto-tagged with the correcttraceId
.GET /sample/mono
- AMono
endpoint where theMono
delays for 100 milliseconds before supplying the result. The[DISTRIBUTED_TRACING]
log message for this endpoint should have adurationNanos
around100000000
(100 milliseconds). You should see a log message output by the endpoint that is auto-tagged with the correcttraceId
.GET /sample/flux
- AFlux
endpoint where theFlux
delays for 100 milliseconds total to supply all the results. The[DISTRIBUTED_TRACING]
log message for this endpoint should have adurationNanos
around100000000
(100 milliseconds). You should see a log message output by the endpoint that is auto-tagged with the correcttraceId
.GET /sample/router-function
- ARouterFunction
endpoint where theMono
delays for 100 milliseconds before supplying the result. The[DISTRIBUTED_TRACING]
log message for this endpoint should have adurationNanos
around100000000
(100 milliseconds). You should see a log message output by the endpoint that is auto-tagged with the correcttraceId
.GET /sample/async-error
- AMono
endpoint where theMono
completes with an error. The[DISTRIBUTED_TRACING]
log message for this endpoint should have adurationNanos
around100000000
(100 milliseconds). You should see a log message output by the endpoint that is auto-tagged with the correcttraceId
, along with another auto-tagged one with the error.GET /sample/async-timeout
- AMono
endpoint where theMono
is set to timeout after 100 milliseconds. The[DISTRIBUTED_TRACING]
log message for this endpoint should have adurationNanos
around100000000
(100 milliseconds). You should see a log message output by the endpoint that is auto-tagged with the correcttraceId
, along with another auto-tagged one with the error.GET /sample/async-future
- An async endpoint (usingCompletableFuture
) that waits for 100 milliseconds on another thread before completing the request. The[DISTRIBUTED_TRACING]
log message for this endpoint should have adurationNanos
around100000000
(100 milliseconds). You should see two log messages output by the endpoint that are auto-tagged with the correcttraceId
- one for the endpoint before async processing is started, and another on the async thread before the request is completed.GET /sample/span-info
- AMono
endpoint that waits for 100 milliseconds and then returns a JSON response payload that contains information about both the parent span that came in on the request headers (if any) and the endpoint span. This can be helpful to mentally visualize what's going on when you send in tracing headers on the request (essentially propagating your own tracing info to the sample server). The[DISTRIBUTED_TRACING]
log message for this endpoint should have adurationNanos
around100000000
(100 milliseconds). You should see multiple log messages output by the endpoint that are auto-tagged with the correcttraceId
. If you explore the code for this endpoint you'll see how the correctTracingState
for the request can be extracted from the thread when the endpoint is executed, theServerWebExchange
attributes, and/or theMono
'sContext
.GET /sample/nested-webclient-call
- AMono
endpoint that waits 100 milliseconds and then uses aWebClient
with tracing interceptor to make a HTTP client call to/sample/span-info
that is automatically wrapped in a subspan and automatically propagates the tracing info on the downstream call. The result of/sample/span-info
is returned. The total call time should be around 200 milliseconds (100 for each of the endpoints that are called). There should be three[DISTRIBUTED_TRACING]
log messages - one for the innermost/sample/span-info
endpoint, one for the subspan surrounding theWebClient
HTTP client call, and one for the outermost/sample/nested-webclient-call
endpoint. You should see several log messages auto-tagged with the correcttraceId
across both endpoints. Note how all the different log messages are tagged with the trace ID despite the request hopping threads several times.GET /sample/path-param/{somePathParam}
- Similar to theGET /sample/mono
endpoint, except it has a path parameter in the path. The span name andhttp.route
tag will use the low-cardinality path template, while thehttp.path
andhttp.url
tags will contain the full high-cardinality path.GET /sample/wildcard/**
- Similar to theGET /sample/mono
endpoint, except it has a wildcard in the path. The span name andhttp.route
tag will use the low-cardinality path template, while thehttp.path
andhttp.url
tags will contain the full high-cardinality path.
See the base project README.md and Wingtips repository source code and javadocs for all further information.
Wingtips is released under the Apache License, Version 2.0