Erlang/Elixir TCP client for Event Store.
This version is tested with EventStore 3.9.3 - 4.1.1, Elixir 1.5, 1.6, 1.7 and Erlang/OTP 19.3, 20.3 and 21.0
Add Extreme as a dependency in your mix.exs
file.
def deps do
[{:extreme, "~> 0.13.2"}]
end
In order to deploy your app using exrm
you should also update your application list to include :extreme
:
def application do
[applications: [:extreme]]
end
Extreme includes all its dependencies so you don't have to name them separately.
After you are done, run mix deps.get
in your shell to fetch and compile Extreme and its dependencies.
Starting from EventStore version 4.0 there are some upgrades to communication protocol. Event number size is changed to 64bits
and there is new messages IdentifyClient
and ClientIdentified
. Since we would like to keep backward compatibility with older v3 protocol,
we introduced new configuration for :extreme
application, where you have to set :protocol_version
equal to 4
if you want to use new protocol, default is 3
.
Below is exact line you have to add in you application config file in order to activate new protocol:
config :extreme, :protocol_version, 4
The best way to understand how adapter should be used is by investigating test/extreme_test.exs
file,
but we'll try to explain some details in here as well.
Extreme is implemented using GenServer and is OTP compatible. If client is disconnected from server we are not trying to reconnect, instead you should rely on your supervisor. For example:
defmodule MyApp.Supervisor do
use Supervisor
def start_link do
Supervisor.start_link __MODULE__, :ok
end
@event_store MyApp.EventStore
def init(:ok) do
event_store_settings = Application.get_env :my_app, :event_store
children = [
worker(Extreme, [event_store_settings, [name: @event_store]]),
# ... other workers / supervisors
]
supervise children, strategy: :one_for_one
end
end
You can manually start adapter as well (as you can see in test file):
{:ok, server} = Application.get_env(:extreme, :event_store) |> Extreme.start_link
From now on, server
pid is used for further communication. Since we are relying on supervisor to reconnect,
it is wise to name server
as we did in example above.
Extreme can connect to single ES node or to cluster specified with node IPs and ports.
Example for connecting to single node:
config :extreme, :event_store,
db_type: :node,
host: "localhost",
port: 1113,
username: "admin",
password: "changeit",
reconnect_delay: 2_000,
connection_name: :my_app,
max_attempts: :infinity
db_type
- defaults to :node, thus it can be omittedhost
- check EXT IP setting of your EventStoreport
- check EXT TCP PORT setting of your EventStorereconnect_delay
- in ms. Defaults to 1_000. If tcp connection fails this is how long it will wait for reconnection.connection_name
- Optional param introduced in EventStore 4. Connection can be identified by this name on ES UImax_attempts
- Defaults to :infinity. Specifies how many times we'll try to connect to EventStore
Example for connecting to cluster:
config :extreme, :event_store,
db_type: :cluster,
gossip_timeout: 300,
mode: :read,
nodes: [
%{host: "10.10.10.29", port: 2113},
%{host: "10.10.10.28", port: 2113},
%{host: "10.10.10.30", port: 2113}
],
connection_name: :my_app,
username: "admin",
password: "changeit"
gossip_timeout
- in ms. Defaults to 1_000. We are iterating throughnodes
list, asking for cluster member details. This setting represents timeout for gossip response before we are asking next node fromnodes
list for cluster details.nodes
- Mandatory for cluster connection. Represents list of nodes in the cluster as we know ithost
- should be EXT IP setting of your EventStore nodeport
- should be EXT HTTP PORT setting of your EventStore node
mode
- Defaults to:write
where Master node is prefered over Slave, otherwise prefer Slave over Master
Example of connection to cluster via DNS lookup
config :extreme, :event_store,
db_type: :cluster_dns,
gossip_timeout: 300,
host: "es-cluster.example.com", # accepts char list too, this whould be multy A record host enrty in your nameserver
port: 2113, # the external gossip port
connection_name: :my_app,
username: "admin",
password: "changeit",
mode: :write,
max_attempts: :infinity
When cluster
mode is used, adapter goes thru nodes
list and tries to gossip with node one after another
until it gets response about nodes. Based on nodes information from that response it ranks their statuses and chooses
the best candidate to connect to. For :write
mode (default) Master
node is prefered over Slave
,
but for :read
mode it is opposite. For the way ranking is done, take a look at lib/cluster_connection.ex
:
defp rank_state("Master", :write), do: 1
defp rank_state("Master", _), do: 2
defp rank_state("PreMaster", :write), do: 2
defp rank_state("PreMaster", _), do: 3
defp rank_state("Slave", :write), do: 3
defp rank_state("Slave", _), do: 1
Note that above will work with same procedure with cluster_dns
mode turned on, since internally it will get ip addresses to which the same connection procedure will be used.
Once client is disconnected from EventStore, supervisor should respawn it and connection starts over again.
EventStore uses ProtoBuf for taking requests and sending responses back.
We are using exprotobuf to deal with them.
List and specification of supported protobuf messages can be found in include/event_store.proto
file.
Instead of wrapping each and every request in elixir function, we are using execute/2
function that takes server pid and request message:
{:ok, response} = Extreme.execute server, write_events()
where write_events
can be helper function like:
alias Extreme.Msg, as: ExMsg
defp write_events(stream \\ "people", events \\ [%PersonCreated{name: "Pera Peric"}, %PersonChangedName{name: "Zika"}]) do
proto_events = Enum.map(events, fn event ->
ExMsg.NewEvent.new(
event_id: Extreme.Tools.gen_uuid(),
event_type: to_string(event.__struct__),
data_content_type: 0,
metadata_content_type: 0,
data: :erlang.term_to_binary(event),
metadata: ""
) end)
ExMsg.WriteEvents.new(
event_stream_id: stream,
expected_version: -2,
events: proto_events,
require_master: false
)
end
This way you can fine tune your requests, i.e. choose your serialization. We are using erlang serialization in this case
data: :erlang.term_to_binary(event)
, but you can do whatever suites you.
For more information about protobuf messages EventStore uses,
take a look at their documentation or for common use cases
you can check test/extreme_test.exs
file.
Extreme.subscribe_to/3
function is used to get notified on new events on particular stream.
This way subscriber, in next example self
, will get message {:on_event, push_message}
when new event is added to stream
people.
def subscribe(server, stream \\ "people"), do: Extreme.subscribe_to(server, self, stream)
def handle_info({:on_event, event}, state) do
Logger.debug "New event added to stream 'people': #{inspect event}"
{:noreply, state}
end
Extreme.read_and_stay_subscribed/7
reads all events that follow a specified event number, and subscribes to future events.
defmodule MyApp.StreamSubscriber
use GenServer
def start_link(extreme, last_processed_event), do: GenServer.start_link __MODULE__, {extreme, last_processed_event}
def init({extreme, last_processed_event}) do
stream = "people"
state = %{ event_store: extreme, stream: stream, last_event: last_processed_event }
GenServer.cast self, :subscribe
{:ok, state}
end
def handle_cast(:subscribe, state) do
# read only unprocessed events and stay subscribed
{:ok, subscription} = Extreme.read_and_stay_subscribed state.event_store, self, state.stream, state.last_event + 1
# we want to monitor when subscription is crashed so we can resubscribe
ref = Process.monitor subscription
{:noreply, %{state|subscription_ref: ref}}
end
def handle_info({:DOWN, ref, :process, _pid, _reason}, %{subscription_ref: ref} = state) do
GenServer.cast self, :subscribe
{:noreply, state}
end
def handle_info({:on_event, push}, state) do
push.event.data
|> :erlang.binary_to_term
|> process_event
event_number = push.link.event_number
:ok = update_last_event state.stream, event_number
{:noreply, %{state|last_event: event_number}}
end
def handle_info(:caught_up, state) do
Logger.debug "We are up to date!"
{:noreply, state}
end
def handle_info(_msg, state), do: {:noreply, state}
defp process_event(event), do: IO.puts("Do something with #{inspect event}")
defp update_last_event(_stream, _event_number), do: IO.puts("Persist last processed event_number for stream")
end
This way unprocessed events will be sent by Extreme, using {:on_event, push}
message.
After all persisted messages are sent, :caught_up message is sent and then new messages will be sent the same way
as they arrive to stream.
If you subscribe to non existing stream you'll receive message {:extreme, severity, problem, stream} where severity can be either :error
(for subscription on hard deleted stream) or :warn
(for subscription on non existing or soft deleted stream). Problem is explanation of problem (i.e. :stream_hard_deleted). So in your receiver you can either have catch all handle_info(_message, _state)
or you can handle such message:
def handle_info({:extreme, _, problem, stream}=message, state) do
Logger.warn "Stream #{stream} issue: #{to_string problem}"
{:noreply, state}
end
Since it is common on read side of system to read events and denormalize them, there is Extreme.Listener macro that hides noise from listener:
defmodule MyApp.MyListener do
use Extreme.Listener
import MyApp.MyProcessor
# returns last processed event by MyListener on stream_name, -1 if none has been processed so far
defp get_last_event(stream_name), do: DB.get_last_event MyListener, stream_name
defp process_push(push, stream_name) do
#for indexed stream we need to follow push.link.event_number, otherwise push.event.event_number
event_number = push.link.event_number
DB.in_transaction fn ->
Logger.info "Do some processing of event #{inspect push.event.event_type}"
:ok = push.event.data
|> :erlang.binary_to_term
|> process_event(push.event.event_type)
DB.ack_event(MyListener, stream_name, event_number)
end
{:ok, event_number}
end
# This override is optional
defp caught_up, do: Logger.debug("We are up to date. YEEEY!!!")
end
defmodule MyApp.MyProcessor do
def process_event(data, "Elixir.MyApp.Events.PersonCreated") do
Logger.debug "Doing something with #{inspect data}"
:ok
end
def process_event(_, _), do: :ok # Just acknowledge events we are not interested in
end
Listener can be started manually but it is most common to place it in supervisor AFTER specifing Extreme:
defmodule MyApp.Supervisor do
use Supervisor
def start_link, do: Supervisor.start_link __MODULE__, :ok
@event_store MyApp.EventStore
def init(:ok) do
event_store_settings = Application.get_env :my_app, :event_store
children = [
worker(Extreme, [event_store_settings, [name: @event_store]]),
worker(MyApp.MyListener, [@event_store, "my_indexed_stream", [name: MyListener]]),
# ... other workers / supervisors
]
supervise children, strategy: :one_for_one
end
end
Subscription can be paused:
{:ok, last_event_number} = MyApp.MyListener.pause MyListener
and resumed
:ok = MyApp.MyListener.resume MyListener
It's not uncommon situation to listen live events and propagate them (for example on web sockets). For that situation there is Extreme.FanoutListener macro that hides noise from listener:
defmodule MyApp.MyFanoutListener do
use Extreme.FanoutListener
import MyApp.MyPusher
defp process_push(push) do
Logger.info "Forward to web socket event #{inspect push.event.event_type}"
:ok = push.event.data
|> :erlang.binary_to_term
|> process_event(push.event.event_type)
end
end
defmodule MyApp.MyPusher do
def process_event(data, "Elixir.MyApp.Events.PersonCreated") do
Logger.debug "Transform and push event with data: #{inspect data}"
:ok
end
def process_event(_, _), do: :ok # Just acknowledge events we are not interested in
end
Listener can be started manually but it is most common to place it in supervisor AFTER specifing Extreme:
defmodule MyApp.Supervisor do
use Supervisor
def start_link, do: Supervisor.start_link __MODULE__, :ok
@event_store MyApp.EventStore
def init(:ok) do
event_store_settings = Application.get_env :my_app, :event_store
children = [
worker(Extreme, [event_store_settings, [name: @event_store]]),
worker(MyApp.MyFanoutListener, [@event_store, "my_indexed_stream", [name: MyFanoutListener]]),
# ... other workers / supervisors
]
supervise children, strategy: :one_for_one
end
end
The Event Store provides an alternate event subscription model, from version 3.2.0, known as competing consumers. Instead of the client holding the state of the subscription, the server remembers it.
The first step in using persistent subscriptions is to create a new subscription. This can be done using the Event Store admin website or in your application code, as shown below. You must provide a unique subscription group name and the stream to receive events from.
alias Extreme.Msg, as: ExMsg
{:ok, _} = Extreme.execute(server, ExMsg.CreatePersistentSubscription.new(
subscription_group_name: "person-subscription",
event_stream_id: "people",
resolve_link_tos: false,
start_from: 0,
message_timeout_milliseconds: 10_000,
record_statistics: false,
live_buffer_size: 500,
read_batch_size: 20,
buffer_size: 500,
max_retry_count: 10,
prefer_round_robin: true,
checkpoint_after_time: 1_000,
checkpoint_max_count: 500,
checkpoint_min_count: 1,
subscriber_max_count: 1
))
Extreme.connect_to_persistent_subscription/5
function is used subscribe to an existing persistent subscription. The subscriber, in this example self
, will receive message {:on_event, push_message}
when each new event is added to stream
people.
{:ok, subscription} = Extreme.connect_to_persistent_subscription(server, self(), group, stream, buffer_size)
You must acknowledge receipt, and successful processing, of each received event. The Event Store will remember the last acknowledged event. The subscription will resume from this position should the subscriber process terminate and reconnect. This simplifies the client logic - the code you must write.
Extreme.PersistentSubscription.ack/3
function is used to acknowledge receipt of an event.
receive do
{:on_event, event, correlation_id} ->
Logger.debug "New event added to stream 'people': #{inspect event}"
:ok = Extreme.PersistentSubscription.ack(subscription, event, correlation_id)
end
You must track the subscription
PID returned from the Extreme.connect_to_persistent_subscription/5
function as part of the process state when using a GenServer
subscriber.
def handle_info({:on_event, event, correlation_id}, %{subscription: subscription} = state) do
Logger.debug "New event added to stream 'people': #{inspect event}"
:ok = Extreme.PersistentSubscription.ack(subscription, event, correlation_id)
{:noreply, state}
end
Events can also be not acknowledged. They can be not acknowledged with a nack_action of :Park, :Retry, :Skip, or :Stop.
def handle_info({:on_event, event, correlation_id}, %{subscription: subscription} = state) do
Logger.debug "New event added to stream 'people': #{inspect event}"
if needs_to_retry do
:ok = Extreme.PersistentSubscription.nack(subscription, event, correlation_id, :Retry)
else
:ok = Extreme.PersistentSubscription.ack(subscription, event, correlation_id)
end
{:noreply, state}
end
- Fork it
- Create your feature branch (
git checkout -b my-new-feature
) - Commit your changes (
git commit -am 'add some feature'
) - Push to the branch (
git push origin my-new-feature
) - Create new Pull Request