A collection of characters, stories, and other elements
Every developer knows they should instrument their app to identify perf bottlenecks, but it’s hard to actually get around to doing it — especially when you’re focused on shipping the latest and greatest features of your site.
Today we’re announcing Appdash, an open source multi-language distributed performance and tracing suite that makes tracing the performance of web apps a breeze. Appdash is used to monitor Sourcegraph, which semantically indexes and serves the code of over a million repositories. It is currently usable from applications written in Go and Python. We hope to add support for more languages in the future with help from the community. (Submit an issue if you’d like to add support for your language of choice!)
There are two primary parts to Appdash:
This post will walk you through an example app that uses Appdash and cover exactly how you can integrate Appdash into your Go web app to start monitoring real-time perf today.
If you don’t already have Go, install that first. Then install Appdash by running:
go get -u sourcegraph.com/sourcegraph/appdash/...
This will also install the example web app in the Appdash repository that demonstrates the basic features of Appdash. Let’s run that first:
$ go get -u sourcegraph.com/sourcegraph/appdash/... $ webapp 2015/04/27 20:40:56 Appdash web UI running on HTTP :8700 [negroni] listening on :8699
Now, point your browser at localhost:8699. This loads the main page of the sample web app, which issues three API requests on the backend. You can view the trace of this page load by clicking on the link in the interface, which opens up the Appdash UI for the trace.
The best way to learn how to use something is by example. So let’s take a look at the source code for this example Go app. Note that this app uses Negroni and the Gorilla Toolkit, but these are not requirements to use Appdash.
For our example purposes, the app has two routes:
The API endpoint code pauses for 200ms, to simulate slowness that in a real application might be due to hitting the database or some external service. When a user visits the root route (/), the backend makes three outbound API requests before responding to the user.
Appdash is heavily influenced by the Dapper data model and has 4 main concepts:
Appdash supports storing data in different underlying Stores. The example code uses an appdash.MemoryStore wrapped in an appdash.RecentStore. This means that the data is stored in memory for 20 seconds before being evicted and discarded (which is useful for applications with storage limitations). You can also store data in a SQL database. It’s easy to add support for other databases, as well. You just need to implement the PersistentStore interface.
There are two ways to run the Appdash web UI. You can either use appdash serve on the command line or embed the UI directly into the application that is being monitored, in which case Appdash will run in-process and listen on a separate port. The example app does it the second way.
In a production environment, you could either use a centralized Appdash server or simply block the Appdash port from external access via your firewall. To embed the web UI into our app, we use the appdash/traceapp package.
To integrate Appdash into your app, you’ll need to create a Collector to collect Annotations (the encoded form of Events) for a given Span provided by a Recorder. In the example code, the collector runs in process, but it can also be a remote service run via appdash serve.
To create our local collector, we simply give it an appdash.Store.
Lastly, we need to create and configure our appdash/httptrace.Middleware, which hooks into your web app’s HTTP handler and pulls the necessary timing and meta-data information (e.g., HTTP headers, status codes, etc.) and generates Events for the Collector to consume.
The RouteName field is simply a function that returns the name of the URL route for a given HTTP request. In the case of the example app, we simply use the request’s path (r.URL.Path). If you are using a routing library like Gorilla mux, you would probably assign it the name of the mux.Route. The route name will be displayed on the span in the web UI.
The SetContextSpan field lets you store a appdash.SpanID of an HTTP request. In our case, we simply use gorilla/context to associate it locally with the request for future use. We will explain below why you would want to know the span ID.
If your web app makes HTTP requests to other services (these can be external services or an internal API if it’s run as a separate service), Appdash can keep track of a thread of execution across HTTP boundaries. This means if an application endpoint (that generates dynamic HTML) makes a call to an API endpoint over the course of its execution, Appdash will be able to associate the specific API request with the specific application request.
This is done by supplying a special http.Transport to the HTTP client that issues the request inside your app. To do this, wrap the existing http.Transport in an instance of appdash/httptrace.Transport. You can then use the http.Client as you would any other http.Client.
Let’s say that a user of your application has reported to you that it’s responding very slowly to their requests. With thousands of people using the service everyday, browsing through the list of all traces hoping to find the trace corresponding to their request is like trying to find a needle in a haystack.
Appdash solves this problem by giving you access to the appdash.SpanID for a given request in the SetContextSpan function for the httptrace.Middleware described above. Earlier, we used gorilla/context to associate the SpanID with the request. And we can render the span IDright into our very simple web page (perhaps in an HTML comment if you didn’t want it to be visible to all users).
Now, a user who is experiencing performance issues with our site can directly give us the trace ID of a slow request. Alternatively, you could create an automated system to do this whenever a user reports an issue from a slow page.
Appdash is an incredibly versatile and easy-to-deploy performance and debug tracing suite for web applications. It supports Go and Python, and we’d love to add more languages with help from the community. It’s being used today in production applications at Sourcegraph, and we hope you’ll find it useful for your own web app.
So check out the source, or file an issue or feature request. And while you’re at it, check out Sourcegraph, which is the best way to discover and understand code — and which was the first site to ever use Appdash.