appdash (view on Sourcegraph)
Appdash is an application tracing system for Go, based on Google's Dapper and Twitter's Zipkin.
Appdash allows you to trace the end-to-end handling of requests and operations in your application (for perf and debugging). It displays timings and application-specific metadata for each step, and it displays a tree and timeline for each request and its children.
To use appdash, you must instrument your application with calls to an appdash recorder. You can record any type of event or operation. Recorders and schemas for HTTP (client and server) and SQL are provided, and you can write your own.
To install appdash, run:
go get -u sourcegraph.com/sourcegraph/appdash/cmd/...
A standalone example using Negroni and Gorilla packages is available in the examples/cmd/webapp
folder.
A demo / pure net/http
application (which is slightly more verbose) is also available at cmd/appdash/example_app.go
, and it can be ran easily using appdash demo
on the command line.
Questions or comments? Join us on #sourcegraph in the Gophers slack!
Appdash uses vfsgen to package HTML templates with the appdash binary for
distribution. This means that if you want to modify the template data in traceapp/tmpl
you can first build using the dev
build tag, which makes the template data be reloaded from disk live.
After you're finished making changes to the templates, always run go generate sourcegraph.com/sourcegraph/appdash/traceapp/tmpl
so that the data_vfsdata.go
file is updated for normal Appdash users that aren't interested in modifying the template data.
Appdash follows the design and naming conventions of Google's Dapper. You should read that paper if you are curious about why certain architectural choices were made.
There are 4 main components/concepts in appdash:
- Spans: A span refers to an operation and all of its children. For example, an HTTP handler handles a request by calling other components in your system, which in turn make various API and DB calls. The HTTP handler's span includes all downstream operations and their descendents; likewise, each downstream operation is its own span and has its own descendents. In this way, appdash constructs a tree of all of the operations that occur during the handling of the HTTP request.
- Event: Your application records the various operations it performs (in the course of handling a request) as Events. Events can be arbitrary messages or metadata, or they can be structured event types defined by a Go type (such as an HTTP ServerEvent or an SQLEvent).
- Recorder: Your application uses a Recorder to send events to a Collector (see below). Each Recorder is associated with a particular span in the tree of operations that are handling a particular request, and all events sent via a Recorder are automatically associated with that context.
- Collector: A Collector receives Annotations (which are the encoded form of Events) sent by a Recorder. Typically, your application's Recorder talks to a local Collector (created with NewRemoteCollector. This local Collector forwards data to a remote appdash server (created with NewServer that combines traces from all of the services that compose your application. The appdash server in turn runs a Collector that listens on the network for this data, and it then stores what it receives.
Appdash has clients available for Go, Python (see python/
subdir) and Ruby (see https://github.com/bsm/appdash-rb).
Appdash supports the OpenTracing API. Please see the
opentracing
subdir for the Go implementation, or see the GoDoc
for API documentation.
appdash was influenced by, and uses code from, Coda Hale's lunk.