Gnat
A nats.io client for elixir. The goals of the project are resiliency, performance, and ease of use.
Usage
{:ok, gnat} = Gnat.start_link(%{host: '127.0.0.1', port: 4222})
# Or if the server requires TLS you can start a connection with:
# {:ok, gnat} = Gnat.start_link(%{host: '127.0.0.1', port: 4222, tls: true})
{:ok, subscription} = Gnat.sub(gnat, self(), "pawnee.*")
:ok = Gnat.pub(gnat, "pawnee.news", "Leslie Knope recalled from city council (Jammed)")
receive do
{:msg, %{body: body, topic: "pawnee.news", reply_to: nil}} ->
IO.puts(body)
end
Instrumentation
Gnat uses telemetry to make instrumentation data available to clients. If you want to record metrics around the number of messages or latency of message publishes, subscribes, requests, etc you can do the following in your project:
iex(1)> metrics_function = fn(event_name, event_value, event_meta, config) ->
IO.inspect([event_name, event_value, event_meta, config])
:ok
end
#Function<4.128620087/4 in :erl_eval.expr/5>
iex(2)> names = [[:gnat, :pub], [:gnat, :sub], [:gnat, :message_received], [:gnat, :request], [:gnat, :unsub]]
[
[:gnat, :pub],
[:gnat, :sub],
[:gnat, :message_received],
[:gnat, :request],
[:gnat, :unsub]
]
iex(3)> :telemetry.attach_many("my listener", names, metrics_function, %{my_config: true})
:ok
iex(4)> {:ok, gnat} = Gnat.start_link()
{:ok, #PID<0.203.0>}
iex(5)> Gnat.sub(gnat, self(), "topic")
[[:gnat, :sub], 128000, %{topic: "topic"}, %{my_config: true}]
{:ok, 1}
iex(6)> Gnat.pub(gnat, "topic", "ohai")
[[:gnat, :pub], 117000, %{topic: "topic"}, %{my_config: true}]
[[:gnat, :message_received], 1, %{topic: "topic"}, %{my_config: true}]
:ok
The pub
, sub
, request
, and unsub
events all report the latency of those respective calls.
The message_received
event always reports a value of 1
because there isn't a good latency metric to report.
All of the events (except unsub
) include metadata with a :topic
key so you can split your metrics by topic.
Benchmarks
Part of the motivation for building this library is to get better performance.
To this end, there is a bench
branch on this project which includes a server.exs
and client.exs
that can be used for benchmarking various scenarios.
As of this commit, the latest benchmark on a 16-core server shows that you can make 170k+ req/sec or up to 192MB/sec.
The bench/*.exs
files also contain some straight-line single-cpu performance tests.
As of this commit my 2014 macbook pro shows.
ips | average | deviation | median | |
---|---|---|---|---|
parse-128 | 317.75 K | 3.18 μs | ±1661.95% | 3 μs |
pub - 128 | 34.07 K | 29.35 μs | ±50.34% | 26 μs |
req-reply-128 | 3.90 K | 256.58 μs | ±25.75% | 243 μs |
Development
Before running the tests make sure you have a locally running copy of gnatsd
(brew install gnatsd
).
We currently use version 1.3.0
in CI, but anything higher than 0.9.6
should be fine.
The typical mix test
will run all the basic unit tests.
You can also run the multi_server
set of tests that test connectivity to different
gnatsd
configurations. You can run these with mix test --only multi_server
.
The tests will tell you how to start the different configurations.
There are also some property-based tests that generate a lot of test cases.
You can tune how many test cases by setting the environment variable N=200 mix test --only property
(default it 100).
For more details you can look at how Travis runs these things in the CI flow.