Zipkin Reporter buffers and sends trace data collected from tracer libraries to a Zipkin compatible backend.
This repository includes a Java reporting library with transport-specific senders. Transport options include HTTP, Apache ActiveMQ, Apache Kafka, gRPC, RabbitMQ and Scribe (Apache Thrift). Requires JRE 6 or later.
These components can be called when spans have been recorded and ready to send to zipkin.
The span encoder is a specialized form of Zipkin's Codec, which only deals with encoding one span.
It is also extensible in case the type of span reported is not zipkin2.Span
After recording an operation into a span, it needs to be reported out of process. There are two builtin reporter implementations in this library, although you are free to create your own.
The simplest mechanism is printing out spans as they are reported.
Reporter.CONSOLE.report(span);AsyncReporter is how you actually get spans to zipkin. By default, it waits up to a second before flushes any pending spans out of process via a BytesMessageSender.
reporter = AsyncReporter.create(URLConnectionSender.create("http://localhost:9411/api/v2/spans"));
// Schedules the span to be sent, and won't block the calling thread on I/O
reporter.report(span);Those using the Brave library can skip
overhead by using AsyncZipkinSpanHandler.
Ex.
sender = URLConnectionSender.create("http://localhost:9411/api/v2/spans");
zipkinSpanHandler = AsyncZipkinSpanHandler.create(sender); // don't forget to close!
tracingBuilder.addSpanHandler(zipkinSpanHandler);See zipkin-reporter-brave for more details.
If you are trying to trace legacy applications, you may be interested in Spring XML Configuration. This allows you to trace legacy Spring 2.5+ applications without any custom code.
By default AsyncReporter starts a thread to flush the queue of reported spans. Spans are encoded before enqueuing so it is easiest to relate the backlog as a function of bytes.
Here are the most important properties to understand when tuning.
| Property | Description |
|---|---|
queuedMaxSpans |
Maximum backlog of spans reported vs sent. Default 10000 |
queuedMaxBytes |
Maximum backlog of span bytes reported vs sent. Corresponds to ReporterMetrics.updateQueuedBytes. Disabled by default |
messageMaxBytes |
Maximum bytes sendable per message including overhead. Default 500,000 bytes (500KB). Defined by BytesMessageSender.messageMaxBytes |
messageTimeout |
Maximum time to wait for messageMaxBytes to accumulate before sending. Default 1 second |
closeTimeout |
Maximum time to block for in-flight spans to send on close. Default 1 second |
When messageTimeout is non-zero, a single thread is responsible for
bundling spans into a message for the sender. If you are using a blocking
sender, a surge of reporting activity could lead to a queue backup. This
will show in metrics as spans dropped. If you get into this position,
switch to an asynchronous sender (like kafka), or increase the concurrency
of your sender.
Spikes of high CPU could be due to encoding many spans, caused indirectly
by a large messageTimeout or messageMaxBytes. Consider lowering the
messageMaxBytes if this occurs, as it will result in less work per
message.
The sender component handles the last step of sending a list of encoded spans onto a transport.
This involves I/O, so you can call sender.send(Collections.emptyList()) to check it works before
using.
BytesMessageSender is used by AsyncReporter, but you can also create your own if you need to.
class CustomReporter implements Flushable {
--snip--
URLConnectionSender sender = URLConnectionSender.json("http://localhost:9411/api/v2/spans");
Callback callback = new IncrementSpanMetricsCallback(metrics);
// Is the connection healthy?
public boolean ok() {
try {
sender.send(Collections.emptyList());
return true;
} catch (Exception e) {
return false;
}
}
public void report(Span span) {
pending.add(SpanBytesEncoder.JSON_V2.encode(span));
}
@Override
public void flush() throws IOException {
if (pending.isEmpty()) return;
List<byte[]> drained = new ArrayList<byte[]>(pending.size());
pending.drainTo(drained);
if (drained.isEmpty()) return;
sender.send(drained);
}By default, senders use json v2 encoding, which is easy to understand, but twice as large as binary encoding for normal spans. If you are running a recent (2.8+) version of Zipkin server, consider Protocol Buffers when you want to optimize for least size.
You can switch to proto3 encoding like so:
sender = KafkaSender.newBuilder()
.encoding(Encoding.PROTO3)
.bootstrapServers("192.168.99.100:9092")
.build()By default, senders use json v2 encoding, which is easy to understand and twice as efficient as the v1 json encoding. However, it relies on recent (1.31+) versions of zipkin server.
You can switch to v1 encoding like so:
reporter = AsyncReporter.builder(URLConnectionSender.create("http://localhost:9411/api/v1/spans"))
.build(SpanBytesEncoder.JSON_V1);All artifacts publish to the group ID "io.zipkin.zipkin.reporter2". We use a common release version for all components.
Releases are uploaded to Sonatype which synchronizes with Maven Central
Snapshots are uploaded to Sonatype after commits to master.
When using multiple reporter components, you'll want to align versions in one place. This allows you to more safely upgrade, with less worry about conflicts.
You can use our Maven instrumentation BOM (Bill of Materials) for this:
Ex. in your dependencies section, import the BOM like this:
<dependencyManagement>
<dependencies>
<dependency>
<groupId>io.zipkin.reporter2</groupId>
<artifactId>zipkin-reporter-bom</artifactId>
<version>${zipkin-reporter.version}</version>
<type>pom</type>
<scope>import</scope>
</dependency>
</dependencies>
</dependencyManagement>Now, you can leave off the version when choosing any supported instrumentation. Also, any indirect use will have versions aligned:
<dependency>
<groupId>io.zipkin.reporter2</groupId>
<artifactId>zipkin-sender-okhttp3</artifactId>
</dependency>With this in place, you can use the property zipkin-reporter.version to override dependency
versions coherently. This is most commonly to test a new feature or fix.
Note: If you override a version, always double check that your version is valid (equal to or later) than what you are updating. This will avoid class conflicts.