Event store implemented in Elixir. Uses PostgreSQL as the underlying storage engine.
Requires Elixir v1.5 and PostgreSQL v9.5 or newer.
EventStore supports running on a cluster of nodes.
MIT License
This README and the following guides follow the
master
branch which may not be the currently published version. Read docs for the latest published version of EventStore on Hex.
- Getting started
- Using the EventStore
- Running on a cluster
- Event serialization
- Upgrading an EventStore
- Used in production?
- Backup and administration
- Benchmarking performance
- Contributing
- Need help?
Define an event store module:
defmodule MyApp.EventStore do
use EventStore, otp_app: :my_app
end
Append events to a stream:
alias MyApp.EventStore
defmodule ExampleEvent do
defstruct [:key]
end
stream_uuid = UUID.uuid4()
expected_version = 0
events = [
%EventStore.EventData{
event_type: "Elixir.ExampleEvent",
data: %ExampleEvent{key: "value"},
metadata: %{user: "[email protected]"}
}
]
:ok = EventStore.append_to_stream(stream_uuid, expected_version, events)
Read all events from a single stream, starting at the stream's first event:
alias MyApp.EventStore
{:ok, events} = EventStore.read_stream_forward(stream_uuid)
More: Using the EventStore
Subscribe to events appended to all streams:
alias MyApp.EventStore
{:ok, subscription} = EventStore.subscribe_to_all_streams("example_subscription", self())
# Wait for the subscription confirmation
receive do
{:subscribed, ^subscription} ->
IO.puts("Successfully subscribed to all streams")
end
receive_loop = fn loop ->
# Receive a batch of events appended to the event store
receive do
{:events, events} ->
IO.puts("Received events: #{inspect events}")
# Acknowledge successful receipt of events
EventStore.ack(subscription, events)
end
loop.(loop)
end
# Infinite receive loop
receive_loop.(receive_loop)
In production use you would use a GenServer
subscriber process and the handle_info/2
callback to receive events.
More: Subscribe to streams
Yes, this event store is being used in production.
PostgreSQL is used for the underlying storage. Providing guarantees to store data securely. It is ACID-compliant and transactional. PostgreSQL has a proven architecture. A strong reputation for reliability, data integrity, and correctness.
You can use any standard PostgreSQL tool to manage the event store data:
Run the benchmark suite using mix with the bench
environment, as configured in config/bench.exs
. Logging is disabled for benchmarking.
MIX_ENV=bench mix do es.reset, app.start, bench
Example output:
## AppendEventsBench
benchmark name iterations average time
append events, single writer 100 20288.68 µs/op
append events, 10 concurrent writers 10 127416.90 µs/op
append events, 20 concurrent writers 5 376836.60 µs/op
append events, 50 concurrent writers 2 582350.50 µs/op
## ReadEventsBench
benchmark name iterations average time
read events, single reader 500 3674.93 µs/op
read events, 10 concurrent readers 50 44653.98 µs/op
read events, 20 concurrent readers 20 73927.55 µs/op
read events, 50 concurrent readers 10 188244.80 µs/op
## SubscribeToStreamBench
benchmark name iterations average time
subscribe to stream, 1 subscription 100 27687.97 µs/op
subscribe to stream, 10 subscriptions 50 56047.72 µs/op
subscribe to stream, 20 subscriptions 10 194164.40 µs/op
subscribe to stream, 50 subscriptions 5 320435.40 µs/op
After running two benchmarks you can compare the runs:
MIX_ENV=bench mix bench.cmp -d percent
You can also produce an HTML page containing a graph comparing benchmark runs:
MIX_ENV=bench mix bench.graph
Taking the above example output, the append events benchmark is for writing 100 events in a single batch. That's what the µs/op average time is measuring. For a single writer it takes on average 0.02s per 100 events appended (4,929 events/sec) and for 50 concurrent writers it's 50 x 100 events in 0.58s (8,586 events/sec).
For reading events it takes a single reader 3.67ms to read 100 events (27,211 events/sec) and for 50 concurrent readers it takes 0.19s (26,561 events/sec).
The purpose of the benchmark suite is to measure the performance impact of proposed changes, as opposed to looking at the raw numbers. The above figures are taken when run against a local PostgreSQL database. You can run the benchmarks against your own hardware to get indicative performance figures for the Event Store.
The benchmark suite is configured to use Erlang's external term format serialization. Using another serialization format, such as JSON, will likely have a negative impact on performance.
Pull requests to contribute new or improved features, and extend documentation are most welcome.
Please follow the existing coding conventions, or refer to the Elixir style guide.
You should include unit tests to cover any changes.
EventStore exists thanks to the following people who have contributed.
- Andrey Akulov
- Ben Smith
- Bruce Williams
- Chris Brodt
- Christian Green
- Craig Savolainen
- David Soff
- Dominik Guzei
- Douglas Vought
- Eamon Taaffe
- Floris Huetink
- Jan Vereecken
- Kaz Walker
- Olafur Arason
- Ole Michaelis
- Paul Iannazzo
- Raphaël Lustin
- Samuel Roze
- Simon Harris
- Stuart Corbishley
- Victor Oliveira Nascimento
- Yamil Díaz Aguirre
Please open an issue if you encounter a problem, or need assistance.
For commercial support, and consultancy, please contact Ben Smith.