Instrumentation library that implements an OpenTracing Tracer for Jaeger (http://jaegertracing.io).
IMPORTANT: The library's import path is based on its original location under github.com/uber
. Do not try to import it as github.com/jaegertracing
, it will not compile. We might revisit this in the next major release.
- โ
import "github.com/uber/jaeger-client-go"
- โ
import "github.com/jaegertracing/jaeger-client-go"
Please see CONTRIBUTING.md.
We recommended using a dependency manager like glide and semantic versioning when including this library into an application. For example, Jaeger backend imports this library like this:
- package: github.com/uber/jaeger-client-go
version: ^2.7.0
If you instead want to use the latest version in master
, you can pull it via go get
.
Note that during go get
you may see build errors due to incompatible dependencies, which is why
we recommend using semantic versions for dependencioes. The error may be fixed by running
make install
(it will install glide
if you don't have it):
go get -u github.com/uber/jaeger-client-go/
cd $GOPATH/src/github.com/uber/jaeger-client-go/
git submodule update --init --recursive
make install
See tracer initialization examples in godoc and config/example_test.go.
The constructor functions for Jaeger Tracer return the tracer itself and an io.Closer
instance.
It is recommended to structure your main()
so that it calls the Close()
function on the closer
before exiting, e.g.
tracer, closer, err := cfg.New(...)
defer closer.Close()
This is especially useful for command-line tools that enable tracing, as well as
for the long-running apps that support graceful shutdown. For example, if your deployment
system sends SIGTERM instead of killing the process and you trap that signal to do a graceful
exit, then having defer closer.Closer()
ensures that all buffered spans are flushed.
The tracer emits a number of different metrics, defined in
metrics.go. The monitoring backend is expected to support
tag-based metric names, e.g. instead of statsd
-style string names
like counters.my-service.jaeger.spans.started.sampled
, the metrics
are defined by a short name and a collection of key/value tags, for
example: name:jaeger.traces, state:started, sampled:y
. See metrics.go
file for the full list and descriptions of emitted metrics.
The monitoring backend is represented by the metrics.Factory
interface from package
"github.com/uber/jaeger-lib/metrics"
. An implementation
of that interface can be passed as an option to either the Configuration object or the Tracer
constructor, for example:
import (
"github.com/uber/jaeger-client-go/config"
"github.com/uber/jaeger-lib/metrics/prometheus"
)
metricsFactory := prometheus.New()
tracer, closer, err := new(config.Configuration).New(
"your-service-name",
config.Metrics(metricsFactory),
)
By default, a no-op metrics.NullFactory
is used.
The tracer can be configured with an optional logger, which will be
used to log communication errors, or log spans if a logging reporter
option is specified in the configuration. The logging API is abstracted
by the Logger interface. A logger instance implementing
this interface can be set on the Config
object before calling the
New
method.
Besides the zap implementation bundled with this package there is also a go-kit one in the jaeger-lib repository.
Since this tracer is fully compliant with OpenTracing API 1.0, all code instrumentation should only use the API itself, as described in the opentracing-go documentation.
A "reporter" is a component receives the finished spans and reports
them to somewhere. Under normal circumstances, the Tracer
should use the default RemoteReporter
, which sends the spans out of
process via configurable "transport". For testing purposes, one can
use an InMemoryReporter
that accumulates spans in a buffer and
allows to retrieve them for later verification. Also available are
NullReporter
, a no-op reporter that does nothing, a LoggingReporter
which logs all finished spans using their String()
method, and a
CompositeReporter
that can be used to combine more than one reporter
into one, e.g. to attach a logging reporter to the main remote reporter.
The remote reporter uses "transports" to actually send the spans out of process. Currently the supported transports include:
- Jaeger Thrift over UDP or HTTP,
- Zipkin Thrift over HTTP.
The tracer does not record all spans, but only those that have the
sampling bit set in the flags
. When a new trace is started and a new
unique ID is generated, a sampling decision is made whether this trace
should be sampled. The sampling decision is propagated to all downstream
calls via the flags
field of the trace context. The following samplers
are available:
RemotelyControlledSampler
uses one of the other simpler samplers and periodically updates it by polling an external server. This allows dynamic control of the sampling strategies.ConstSampler
always makes the same sampling decision for all trace IDs. it can be configured to either sample all traces, or to sample none.ProbabilisticSampler
uses a fixed sampling rate as a probability for a given trace to be sampled. The actual decision is made by comparing the trace ID with a random number multiplied by the sampling rate.RateLimitingSampler
can be used to allow only a certain fixed number of traces to be sampled per second.
The OpenTracing spec allows for baggage, which are key value pairs that are added
to the span context and propagated throughout the trace. An external process can inject baggage
by setting the special HTTP Header jaeger-baggage
on a request:
curl -H "jaeger-baggage: key1=value1, key2=value2" http://myhost.com
Baggage can also be programatically set inside your service:
if span := opentracing.SpanFromContext(ctx); span != nil {
span.SetBaggageItem("key", "value")
}
Another service downstream of that can retrieve the baggage in a similar way:
if span := opentracing.SpanFromContext(ctx); span != nil {
val := span.BaggageItem("key")
println(val)
}
The OpenTracing API defines a sampling.priority
standard tag that
can be used to affect the sampling of a span and its children:
import (
"github.com/opentracing/opentracing-go"
"github.com/opentracing/opentracing-go/ext"
)
span := opentracing.SpanFromContext(ctx)
ext.SamplingPriority.Set(span, 1)
Jaeger Tracer also understands a special HTTP Header jaeger-debug-id
,
which can be set in the incoming request, e.g.
curl -H "jaeger-debug-id: some-correlation-id" http://myhost.com
When Jaeger sees this header in the request that otherwise has no tracing context, it ensures that the new trace started for this request will be sampled in the "debug" mode (meaning it should survive all downsampling that might happen in the collection pipeline), and the root span will have a tag as if this statement was executed:
span.SetTag("jaeger-debug-id", "some-correlation-id")
This allows using Jaeger UI to find the trace by this tag.
Jaeger Tracer supports Zipkin B3 Propagation HTTP headers, which are used by a lot of Zipkin tracers. This means that you can use Jaeger in conjunction with e.g. these OpenZipkin tracers.
However it is not the default propagation format, see here how to set it up.