A library for sending events to Honeycomb.io, wrapping libhoney-java 1.0.6.
Include the following in your project.clj
:
; This is libhoney-java 1.0.6 build number 1 on CircleCI
; The most recent version can be found via the Clojars badge above
[conormcd/clj-honeycomb "1.0.6.1"]
Then initialize the library somewhere:
(require '[clj-honeycomb.core :as honeycomb])
(honeycomb/init {:data-set "The name of your Honeycomb data set"
:write-key "Your Honeycomb team write key (API key)"})
Now send an event:
(honeycomb/send {:foo "bar"})
You can also wrap a chunk of code and trigger an event when the code completes. It also times the code and instruments any exceptions thrown from the body.
(honeycomb/with-event {:foo "foo"} {}
(honeycomb/add-to-event {:bar "bar"})
... code ...
(honeycomb/add-to-event :baz "baz")
... code ...)
; The above sends an event like this:
; {"foo" "foo"
; "bar" "bar"
; "baz" "baz"
; "durationMs" 1234.567}
You can add fields to every event by setting them on the client at initialization time. You can delay computation of the value of those fields until the time the event is sent by using atoms, delays or promises.
(def my-dynamic-field (atom 1))
(honeycomb/init {:global-fields {:global-dynamic my-dynamic-field
:global-static "static"}})
(honeycomb/send {:event-specific "event"})
(swap! my-dynamic-field inc)
(honeycomb/send {:event-specific "event"})
; The above results in the following two events being sent:
; {"event-specific" "event" "global-dynamic" 1 "global-static" "static"}
; {"event-specific" "event" "global-dynamic" 2 "global-static" "static"}
You can make more sophisticated dynamic fields by implementing a
ValueSupplier.
There is a convenience function clj-honeycomb.fields/->ValueSupplier
to
transform a Clojure function into one.
(honeycomb/init {:global-fields {:vs (->ValueSupplier rand-int 100)}})
(honeycomb/send {:event "event"})
; The above produces:
; {"event" "event" "vs" 15}
Honeycomb provides useful
information on why and
when to sample. Sample rates can be set globally at initialization time with
the :sample-rate
key or individually on each send
by passing a
sample-rate
in the options map which is the third argument to send
. If you
implement your own sampling, you must pass {:sample-rate ... :pre-sampled true}
to each call to send
.
Events can be manipulated before they're sampled (with a clj-honeycomb pre-processor function) and after they're sampled (with a libhoney-java EventPostProcessor).
To pre-process an event before it's handed off to libhoney-java, add a
function like (fn [event-data options] ...)
to the :event-pre-processor
optional argument to init
which returns a tuple [event-data options]
. The
arguments and return value are identical to the arguments to
clj-honeycomb.core/send
and may be manipulated in any way so long as the
returned value is a valid set of arguments for send
.
Example:
(defn- event-pre-processor
"Add an extra field to the event data which is a count of the number of
fields being sent."
[event-data options]
[(merge event-data
{:num-items (inc (count event-data))})
options])
(honeycomb/init {...
:event-pre-processor event-pre-processor
...})
To post-process an event, add an
EventPostProcessor
to the :event-post-processor
optional argument to init
. The process
method on that object will be called with a single, mutable
EventData
object. This is called after sampling has taken place, so it will only be run
on events which will be sent to Honeycomb.io. See the documentation for the
EventPostProcessor class to understand the constraints on modifying the event.
TL;DR: Wrap something with with-event
and make sure it includes :traceId
in the event data in order to create tracing spans.
Honeycomb provides powerful tracing capabilities and this library attempts to
create tracing spans from every use of with-event
. The following event data
fields control the creation of spans:
Field | Required? | Default | Description |
---|---|---|---|
:traceId |
Yes | The ID of the trace this belongs to. This is the ID that will tie multiple spans together. | |
:id |
No | (honeycomb/generate-span-id) |
A unique ID for the span. If not provided, one will be randomly
generated for you and added to the event if (and only if) a
traceId has been added to the event before. |
:durationMs |
Yes | Automatically calculated by with-event . |
The duration of the span, in milliseconds. |
:name |
No | The name of the function/method/API handler generating the event. This can be blank. | |
:parentId |
No | The ID of this span's parent span. This is set automatically for you
if you nest calls to with-event . You can set this manually
if you need to "waterfall" spans which don't enclose each other. |
|
:serviceName |
No | The name of the service generating this span. This can be blank. |
Here is the simplest example of creating a tracing span:
(honeycomb/with-event {:traceId (honeycomb/generate-trace-id)} {}
... Do stuff here ...)
You can create multiple spans that are part of the same trace. This will create two spans within a trace, but won't establish any relationship between them:
(honeycomb/with-trace-id (honeycomb/generate-trace-id)
(honeycomb/with-event {:name "X"} {}
... Do stuff ...)
... Do more stuff ...
(honeycomb/with-event {:name "Y"} {}
... Do even more stuff ...))
This produces a trace that looks like this:
+-----------------+ +-------+
| X | | Y |
+-----------------+ +-------+
It's usually useful to establish a relationship between spans in a trace. There are two patterns:
- While doing X we did Y.
- Because we did X we subsequently did Y.
The first of those is most easily expressed by nesting with-event
calls:
(honeycomb/with-event {:name "X" :traceId (honeycomb/generate-trace-id)} {}
... Some X ...
(honeycomb/with-event {:name "Y"} {}
... Some Y ...)
... Some more X ...)
This produces a trace that looks like this:
+-----------------+
| X |
+-----------------+
+-------+
| Y |
+-------+
The second case can be accomplished like so:
(honeycomb/with-trace-id (honeycomb/generate-trace-id)
(let [x-id (honeycomb/generate-span-id)]
(honeycomb/with-event {:name "X" :id x-id} {}
... Some X ...)
(honeycomb/with-event {:name "Y" :parentId x-id} {}
... Some Y ....)))
This produces a trace that looks like this:
+-----------------+
| X |
+-----------------+
+-------+
| Y |
+-------+
The above may not make a lot of sense in the context of a single chunk of code, but sequential (as opposed to enclosing) spans are a natural way of expressing chains of calls between services in a distributed system. There is support in the clj-honeycomb Ring middleware to make this feel natural in a Clojure application.
You can turn every request served by a Ring-compatible HTTP server into a
Honeycomb event with clj-honeycomb.middleware.ring/with-honeycomb-event
. By
default the event will contain a selection of items from the request map and
from the response it will contain the status and a small subset of the
headers. You can customize the fields added to the event by passing an options
map to the middleware.
The following defines a custom honeycomb middleware that extracts only some of the request and response data but adds some static and dynamic fields.
(def count-of-thingers
"An atom keeping track of the count of something, to demonstrate dynamic
fields. This will be dereferenced whenever the event fires."
(atom 0))
(def my-custom-honeycomb-middleware
(partial with-honeycomb-event
{:honeycomb-event-data {:static-field "sent with every event"
:num-thingers count-of-thingers}
:extract-request-fields (fn [req]
{:num-headers (count (:headers req))})
:extract-response-fields (fn [res]
{:status (:status res)})}))
; This will produce events that look like this:
; {"durationMs" 83.932
; "num-headers" 12
; "num-thingers" 3
; "static-field" "sent with every event"
; "status" 404}
Tracing support is included in the Ring middleware. If the X-Honeycomb-Trace
header is present in a request then it will be decoded to propagate the trace
ID and to link the event to the parent span in the calling code. The format of
the header is the same as in other Honeycomb libraries and is as follows:
<header> ::= "X-Honeycomb-Trace: " <value>
<value> ::= version ";" data
<version> ::= "1"
<data> ::= <pair> "," <pair>
| <pair>
<pair> ::= "trace_id=" [^,]+
| "parent_id=" [^,]+
| "context=" <base64-encoded-json>
If the version is wrong or if the trace_id
is not present then the
middleware will act as if the header was not present. The parent_id
is
optional but recommended and must refer to the ID of a span associated with
the same trace ID. The context
is also optional and must be a Base64-encoded
JSON string containing data that should be added to the event.
The libhoney-java library sends events to Honeycomb asynchronously on a
background thread. It also batches events. To monitor the progress of the
sending of events you can add a
ResponseObserver
to the client. The easiest way to do that is by adding one or more functions
to a :response-observer
map in the client options passed to either init
or
client
.
(honeycomb/init {:data-set "data-set"
:write-key "write-key"
:response-observer {:on-client-rejected
(fn [client-rejected]
...)
:on-server-accepted
(fn [server-accepted]
...)
:on-server-rejected
(fn [server-rejected]
...)
:on-unknown
(fn [unknown]
...)}})
You may omit any of the functions in the :response-observer
map with no ill
effects. Each of the functions takes a single argument and the types of the
arguments are as follows:
- :on-client-rejected - ClientRejected
- :on-server-accepted - ServerAccepted
- :on-server-rejected - ServerRejected
- :on-unkonwn - Unknown
If you would like to avoid using this library in a stateful manner you can
avoid calling clj-honeycomb.core/init
and accomplish everything with
clj-honeycomb.core/client
and clj-honeycomb.core/send
. The former is used
to create a HoneyClient
which can then be passed as the first argument to send
in order to send
events. The with-event
macro will throw if you attempt to use it without
calling init
first. You are responsible for calling .close
on the client
before disposing of it. It's recommended that you use with-open
or some
state management system like component or mount.
If you're testing code that uses the implicit client created with init
then
you can use clj-honeycomb.testing-utils/validate-events
to make assertions
about the events sent by some code. This also prevents events from being sent
to Honeycomb. The events passed to the second function is a vector of
ResolvedEvent.
(require '[clj-honeycomb.testing-utils :refer (validate-events)])
(validate-events
(fn []
... code that emits events ...
)
(fn [events errors]
... events contains all the events that would have been sent ...
... errors contains any errors emitted by libhoney-java ...))
You can also create a fake HoneyClient
which will record all the events sent
to it.
(require '[clj-honeycomb.core :as honeycomb])
(require '[clj-honeycomb.testing-utils :refer (recording-client)])
(let [events (atom [])]
(with-open [client (recording-client events {})]
(honeycomb/send client {:foo "bar"}))
... events now contains the ResolvedEvent ...)
Automatically generated API documentation is uploaded to GitHub Pages on every release. It can be viewed here:
https://conormcd.github.io/clj-honeycomb/
Since this library wraps libhoney-java
it may also be useful to refer to the
API documentation for that from time to time:
https://honeycombio.github.io/libhoney-java/index.html?overview-summary.html
Copyright 2018-2019 Conor McDermottroe
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this software except in compliance with the License. You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.