This package and its subpackages contain bits of code to use to make your life easier when instrumenting a Go app to send events to Honeycomb. The wrappers here will collect a handful of useful fields about HTTP requests and SQL calls in addition to establishing easy patterns to augment this data as your application runs.
Documentation and examples are available via godoc.
Sign up for a Honeycomb trial to obtain a write key before starting.
For each of the wrappers, documentation is found in godoc rather than the github README files. Each package has a fully functional example in the godoc as well, to show all the pieces fit together.
Add the main package to your $GOPATH in the normal way: go get github.com/honeycombio/beeline-go. You'll add additional subpackages
depending on the specifics of your application.
Regardless of which subpackages are used, there is a small amount of global configuration to add to your application's startup process. At the bare minimum, you must pass in your Team Write Key and identify a dataset name to authorize your code to send events to Honeycomb and tell it where to send events.
import "github.com/honeycombio/beeline-go"
...
func main() {
beeline.Init(beeline.Config{
WriteKey: "abcabc123123defdef456456",
Dataset: "myapp",
})
...After initialization, the next step is to find the http.ListenAndServe call
and add in the
hnynethttp
wrapper. This establishes the most basic instrumentation at the outermost layer
of the call stack on a per-request basis.
import "github.com/honeycombio/beeline-go/wrappers/hnynethttp"
...
http.ListenAndServe(":8080", hnynethttp.WrapHandler(muxer))Now make a few requests in your web app and open up Honeycomb. You should see your new dataset! Open it to see the events you just sent.
Once this middleware wrapper is in place, there is a Honeycomb event in the request context available for use throughout the request's lifecycle. You could stop here and have very basic instrumentation, or continue to get additional context.
- Which endpoints are the slowest?
BREAKDOWN: request.path
CALCULATE: P99(duration_ms)
FILTER: meta.type = http request
ORDER BY: P99(duration_ms) DESC
- Which endpoints are the most frequently hit?
BREAKDOWN: request.path
CALCULATE: COUNT
FILTER: meta.type = http request
- Which users are using the endpoint that I'd like to deprecate? (assuming we add a custom field with user.email)
BREAKDOWN: app.user.email
CALCULATE: COUNT
FILTER: request.url == <endpoint-url>
Depending on which wrappers you use, events will have different fields. This
example gives you a feel for what generated events will look like, though it may
not exactly match what you'll see. All events have at least a timestamp, a
duration, and the meta.type field.
{
"Timestamp": "2018-03-20T00:47:25.339Z",
"duration_ms": 0.809993,
"meta.local_hostname": "cobbler.local",
"meta.type": "http request",
"mux.handler.name": "main.hello",
"mux.handler.pattern": "/hello/",
"mux.handler.type": "http.HandlerFunc",
"request.content_length": 0,
"request.header.user_agent": "curl/7.54.0",
"request.host": "localhost:8080",
"request.method": "GET",
"request.path": "/hello/foo/bar",
"request.proto": "HTTP/1.1",
"request.remote_addr": "127.0.0.1",
"response.status_code": 200,
"trace.trace_id": "5279bdc7-fedc-483b-8e4f-a03b4dbb7f27"
}At any time (once the *http.Request is decorated with a Honeycomb event by the
beeline) you can add additional custom fields to the event associated with this
request.
beeline.AddField(req.Context(), "field_name", value)By default additional fields you add are namespaced under app., so the field
in the example above would appear in your event as app.field_name. This
namespacing groups all your fields together to make them easy to find and
examine.
These additional fields are your opportunity to add important and detailed context to your instrumentation. Put a timer around a section of code, add per- user information, include details about what it took to craft a response, and so on. It is expected that some fields will only be present on some requests. Error handlers are a great example of this; they will obviously only exist when an error has occurred.
It is common practice to add in these fields along the way as they are processed in different levels of middleware. For example, if you have an authentication middleware, it would add a field with the authenticated user's ID and name as soon as it resolves them. Later on in the call stack, you might add additional fields describing what the user is trying to achieve with this specific HTTP request.
After the router has parsed the request, more fields specific to that router are available, such as the specific handler matched or any request parameters that might be attached to the URL. The wrappers for different HTTP routers handle this part differently. The next step is to add middleware or specific handler wrappers in additional places; instructions on how to do this are in each of the subpackages below.
Available HTTP wrappers:
hnynethttp(fornet/http)hnygoji(forgoji/mux)hnygorilla(forgorilla/mux)hnyhttprouter(forhttprouter)
Available DB wrappers:
If your favorite framework isn't listed here, but supports middleware, look at
the hnynethttp wrapper. Chances are, a phrase like "any
middleware in the ecosystem that is also compatible with net/http can be
used"(from go-chi) means that it
expects a function that takes a http.Handler and returns a http.Handler,
which is exactly what the WrapHandler function in hnynethttp does.
Try that out and see how far you can get with it and appropriate custom fields.
If you are using both an HTTP wrapper and a SQL package, make sure you pass the
context from the *http.Request through to the SQL package using the various
Context-enabled function calls. Doing so will tie the SQL calls back to specific
HTTP requests and you'll get extra fields in your request event showing things
like how much time was spent in the DB, as well as request IDs tying the
separate events together so you can see exactly which DB calls were triggered by
a given event.
For very high throughput services, you can send only a portion of the events
flowing through your service by setting the SampleRate during initialization.
This sample rate will send 1/n events, so a sample rate of 5 would send 20% of
all events. For high throughput services, a sample rate of 100 is a good start.
There are two general approaches to finding out what's going wrong when the beeline isn't doing what you expect.
Use the STDOUT flag in configuring the beeline. This will print events to the
terminal instead of sending them to Honeycomb. This lets you quickly see
what's getting sent and modify your code to get what you would like to see.
Use the PrintResponses flag in configuring the beeline. This will print the
responses that come back from Honeycomb to the terminal when sending events.
These responses will have extra detail saying why events are being rejected (or
that they are being accepted) by Honeycomb.
This beeline is also still young, so please reach out to [email protected] or ping us with the chat bubble on https://honeycomb.io for assistance.
The beeline is written with an eye towards a fairly recent set of dependent packages. Some of the dependencies have changed their API interface over the past few years (eg goji, uuid, net/http) so you may need to upgrade to get to a place that works.
- go 1.9+ - the context package moved into the core library and is used extensively by the beeline to make events available to the call stack
- github.com/google/uuid v0.2+ - the signature for NewRandom started returning
UUID, error - github.com/goji/goji v2.0+ - they started using contexts in
net/httpinstead of their own