Trace is a golang package that implements a distributed tracing capability inspired by Google's Dapper. Traces may be optionally recorded to a database backend or local file. HTTP middleware is provided to facilitate easy tracing of requests and cascading of trace contexts during request fanout scenarios.
To install the package:
$ go get github.com/SpirentOrion/trace
Each trace is logically comprised of one or more spans in a tree-like structure. You are free to determine the granularity of traces and spans. Typical usages include:
Traces are identified by a probabilistically unique 64-bit integer. All spans within in a trace share the same trace id. Identifiers are randomly generated within this number space and thus do not require use of a centralized allocator.
Spans are also identified by their own unique 64-bit integer values. Each span records its trace id, the id of its parent span, its location, its start time, its finish time, and other data that you may provide.
With this structure it is possible to build a causal record of activity within trace. For any trace, activity began with the first span -- the span with a parent id of 0. Any spans within the same trace that have a parent id matching the first span's id were caused by the first span. And so on.
If processing of a span requires fanout to other services or processes the trace context may be propagated using HTTP request headers or other appropriate mechanisms. Each span's location indicates where the activity actually occurred, in terms of both process and hostname. When activity spans multiple hosts, start and finish times are based on the host's clock and aren't necessarily synchronized.
Currently, two recording backend packages are provided:
A simple example is provided with trace recording via the YAML recorder:
cd $GOPATH/src/github.com/SpirentOrion/trace/example go run main.go
$ curl -i http://127.0.0.1/foo/bar $ cat example.yaml
Note that the YAML recorder only records finished spans. Each span is rendered as a separate document in the YAML stream.