Java distributed tracing implementation compatible with Zipkin.
Zipkin is based on Dapper.
dapper (dutch) = brave (english)... so that's where the name comes from.
I advise you to read the Dapper paper, but in short:
What we want to achieve is understand system behavior and performance of complex distributed systems. We want to do this with minimal impact on existing code by introducing some small common libraries that are reusable and don't interfere with the existing business logic or architecture. Besides not impacting business logic or architecute we off course also want it to have a neglectable impact on performance.
You should use brave instead of Zipkin if:
- You can't use Finagle.
- You don't want to add Scala as a dependency to your Java project.
- You want out of the box integration support for RESTEasy, Jersey, Apache HttpClient.
Brave uses the Zipkin thrift generated classes as part of its api so it is easy to use existing Zipkin components with Brave (zipkin-collector, zipkin-query, zipkin-ui, cassandra store,...).
I'm very thankful to Twitter for open sourcing Zipkin! Is is by seeing their Zipkin video and presentation that I got to know Zipkin/Dapper and that I saw the potential and the simplicity of the solution.
- span: A single client/server request/response. Can have an optional parent span id and is part of a trace.
- trace: A tree of spans.
So as you can see a single span is submitted twice:
- from the client side, the initiator, with cs (client send) and cr (client received) annotations
- from the server side with sr (server received) and ss (server send) annotations.
The above image shows how I intend to use Brave in production and also how it integrates with the Zipkin back-end components.
The applications and services have Brave integration at client and server side, for example through the RestEasy support module.
They submit spans to the ZipkinSpanCollector
which submits them to Flume.
I introduced Flume instead of Scribe as Flume is still actively maintained, easier to deploy, has good documentation and extensions are written in Java.
The ZipkinSpanCollectorSink
, part of flume-zipkin-collector-sink
module submits spans to the Zipkin collector service.
The ZipkinMetricsSink
, part of flume-zipkin-metrics-sink
module submits custom annotations with duration (to measure certain sections
of the code) to the Metrics library. Metrics builds histograms for the metrics and sends the data
to a back-end system for storage and visualisation. Metrics supports multiple back-ends but the sink implementation today supports
graphite.
If you use the ZooKeeperSamplingTraceFilter
from brave-tracefilters
module you can enable/disable tracing or adjust
sample rate using ZooKeeper as also indicated on the drawing.
The brave-resteasy-example is a good starting point to get you up to speed on how you can implement brave in your own apps.
Maven artifacts for each release are published to Maven central.
For an overview of the available releases see Github releases. As of release 2.0 we try to stick to Semantic Versioning.