zipkin-finagle

Integration between Finagle tracing to Zipkin transports including http, kafka and scribe.

Quick start

Finagle will use a tracer that it detects in the classpath. For example, depending on io.zipkin.finagle2:zipkin-finagle-http_2.12 will send to Zipkin over Http.

You can look at the example for this in use.

Choosing a tracer explicitly

You can also explicitly choose a tracer, which is more reliable, at the tradeoff of more configuration.

Here an example initializing an http client in scala and java

client = Http.client
             .withTracer(new HttpZipkinTracer())
             .withLabel("frontend") // becomes the zipkin service name
             .newService("remotehost:8080");

client = Http$.MODULE$.client()
              .withTracer(new HttpZipkinTracer())
              .withLabel("frontend") // becomes the zipkin service name
              .newService("remotehost:8080");

Note on service names

Sometimes labels used in Finagle do not match the intent of the Zipkin service name. If your traces or dependency graph looks incorrect, set the flag zipkin.localServiceName. This will ignore any of the labels set by servers. In clients, the ServiceName recorded by Finagle will be moved to span.remoteEndpoint.serviceName in Zipkin under the assumption that it was intended to name the remote host.

Configuration

Global Flags

zipkin-finagle configuration is via global flags.

Global flags can either be set by system property, or commandline argument (ex if using TwitterServer).

Ex the following are equivalent ways to trace every request:

$ java -Dzipkin.localServiceName=favstar  -Dzipkin.initialSampleRate=1.0 ...
$ java -cp my-twitter-server.jar -zipkin.localServiceName=favstar -zipkin.initialSampleRate=1.0

Here are the flags that apply to all transports:

Flag	Default	Description
zipkin.localServiceName	unknown	Overrides any ServiceName annotation set by Finagle. Controls Span.localEndpoint.serviceName in Zipkin.
zipkin.initialSampleRate	0.001 (0.1%)	Percentage of traces to sample (report to zipkin) in the range [0.0 - 1.0]

Http Configuration

Adding io.zipkin.finagle2:zipkin-finagle-http_2.12 to your classpath will configure Finagle to report trace data to a Zipkin server via HTTP.

Here are the flags that apply to Http:

Flag	Default	Description
zipkin.http.host	localhost:9411	The network location of the Zipkin http service. See http://twitter.github.io/finagle/guide/Names.html
zipkin.http.hostHeader	zipkin	The Host header used when sending spans to Zipkin
zipkin.http.path	/api/v2/spans	The path to the spans endpoint
zipkin.http.compressionEnabled	true	True implies that spans will be gzipped before transport
zipkin.http.tlsEnabled	false	Whether or not the Zipkin host uses TLS
zipkin.http.tlsValidationEnabled	true	Whether or not to enable TLS validation for the Zipkin host when TLS is enabled

Ex. Here's how to configure the Zipkin server with a system property:

$ java -Dzipkin.http.host=192.168.99.100:9411 ...

Kafka Configuration

Adding io.zipkin.finagle2:zipkin-finagle-kafka_2.12 to your classpath will configure Finagle to report trace data to a Kafka topic. The minimum Kafka server version is 0.10

Here are the flags that apply to Kafka:

Flag	Default	Description
zipkin.kafka.bootstrapServers	localhost:9092	Initial set of kafka servers to connect to, rest of cluster will be discovered (comma separated)
zipkin.kafka.topic	zipkin	Kafka topic zipkin traces will be sent to

Ex. Here's how to configure the Kafka server with a system property:

$ java -Dzipkin.kafka.bootstrapServers=192.168.99.100 ...

Scribe Configuration

Adding io.zipkin.finagle2:zipkin-finagle-scribe_2.12 to your classpath will configure Finagle to report trace data to the zipkin category of Scribe.

Here are the flags that apply to Scribe:

Flag	Default	Description
zipkin.scribe.host	localhost:1463	The network location of the Scribe service. See http://twitter.github.io/finagle/guide/Names.html

Ex. Here's how to configure the Scribe host with a system property. In this case pointing directly to a zipkin server:

$ java -Dzipkin.scribe.host=zipkinhost:9410 ...

Programmatic

You can also configure zipkin tracers programmatically. Here's an example:

HttpZipkinTracer.Config config = HttpZipkinTracer.Config.builder()
    // The frontend makes a sampling decision (via Trace.letTracerAndId) and propagates it downstream.
    // This property says sample 100% of traces.
    .initialSampleRate(1.0f)
    // All servers need to point to the same zipkin transport
    .host("127.0.0.1:9411").build();

Tracer tracer = HttpZipkinTracer.create(config,
    // print stats about zipkin to the console
    new JavaLoggerStatsReceiver(Logger.getAnonymousLogger()));

Custom

You may need to use alternative reporters like zipkin-aws, control configuration not available in flags, or change metrics reporting configuration. To do that, use ZipkinReporter.newBuilder() and supply the arguments you need.

// Setup a sender without using Finagle flags like so:
Properties overrides = new Properties();
overrides.put(ProducerConfig.MAX_BLOCK_MS_CONFIG, 5000);
sender = KafkaSender.newBuilder()
  .bootstrapServers("host1:9092,host2:9092")
  .overrides(overrides)
  .encoding(Encoding.PROTO3)
  .build();

zipkinStats = stats.scope("zipkin");
spanReporter = AsyncReporter.builder(sender)
  .metrics(new ReporterMetricsAdapter(zipkinStats)) // routes reporter metrics to finagle stats
  .build()

// Now, use it here, but don't forget to close the sender!
tracer = ZipkinTracer.newBuilder(spanReporter).stats(zipkinStats).build();

Metrics

The following metrics are reported under a transport-specific category to the configured StatsReceiver. For example, if using kafka, they are reported relative to "zipkin.kafka".

Metric	Description
spans	the count of spans recorded by the tracer
span_bytes	the count of encoded span bytes recorded by the tracer
spans_dropped	the count of spans dropped for any reason. For example, failure queueing or sending.
messages	the count of messages sent to zipkin. Ex POST requests or Kafka messages.
message_bytes	the count of encoded message bytes sent. This includes encoding overhead and excludes compression.
messages_dropped/exception_class_name+	count of messages dropped broken down by cause.
span_queue_size	last count of spans in the pending queue
span_queue_bytes	last count of encoded span bytes in the pending queue

Artifacts

All artifacts publish to the group ID "io.zipkin.finagle2". We use a common release version for all components.

Library Releases

Releases are at Sonatype and Maven Central

Library Snapshots

Snapshots are uploaded to Sonatype after commits to master.