Skip to content

Latest commit

 

History

History
145 lines (107 loc) · 3.63 KB

README.md

File metadata and controls

145 lines (107 loc) · 3.63 KB

DawdleDB

DawdleDB uses Dawdle and SQS to capture change notifications from PostgreSQL.

Installation

The package can be installed by adding dawdle_db to your list of dependencies in mix.exs:

def deps do
  [
    {:dawdle_db, "~> 0.8.0"}
  ]
end

Documentation can be found at https://hexdocs.pm/dawdle_db.

Usage

DawdleDB can be run in either watcher mode or listener mode. In watcher mode it captures PostgreSQL notifications and encodes them into an Elixir struct which is then posted to SQS. In listener mode, the events are pulled out of SQS and handed off to handlers for processing.

For best results, there should only be one instance of DawdleDB running in watcher mode; though you can have many instances running in listener mode. In normal use, DawdleDB can handle this internally. You can start the watcher from your application's start callback and DawdleDB will use Swarm to ensure that there is only one instance of the watcher running. The function DawdleDB.start_watcher/1 takes the name of your Ecto repo and will use the repo's configuration to connect to the database.

defmodule MyApp do
  use Application

  def start(_type, _args) do
    children = [
      # ...
    ]

    opts = [strategy: :one_for_one, name: MyApp.Supervisor]
    sup = Supervisor.start_link(children, opts)

    DawdleDB.start_watcher(MyApp.Repo)

    sup
  end
end

If you want to run the DawdleDB watcher in a separate application, a model application that can be deployed as-is or customized for a particular application can be found at https://github.com/hippware/dawdle_db_watcher.

Initial setup

DawdleDB relies on database triggers to fire the appropriate notifications. There are helpers defined in DawdleDB.Migration to simplify initial setup and defining the triggers.

For example, if you already have a users table and you wish to receive notifications on insert, update, and delete, you could create a migration like the following:

defmodule MyApp.DawdleDBSetup do
  use Ecto.Migration

  import DawdleDB.Migration

  def up do
    create_watcher_events_table()
    update_notify("users", [:insert, :update, :delete])
  end

  def down do
    remove_notify("users", [:insert, :update, :delete])
    drop_watcher_events_table()
  end
end

Handlers

Once the watcher events table has been created, and triggers have been setup, define a handler based on DawdleDB.Handler.

defmodule MyApp.UserHandler do
  use DawdleDB.Handler, type: MyApp.User

  alias MyApp.User

  def handle_insert(%User{} = new) do
    # Do something when a user is created
    :ok
  end

  def handle_update(%User{} = new, %User{} = old) do
    # Do something when a user is updated
    :ok
  end

  def handle_delete(%User{} = old) do
    # Do something when a user is deleted
    :ok
  end
end

A note about the :map type

In Ecto, maps stored with atom-typed keys will always be returned from queries with string-typed keys. DawdleDB intentionally mimics this behaviour for consistency, meaning that the records contained in events with a :map type field may be different from the originally inserted data (if you used atoms as keys).

In other words, if you insert this:

%MyRecord{
  map: %{a: "b"}
}

The record contained in the insertion event will be

%MyRecord{
  map: %{"a" => "b"}
}

(which is the same as the record you would receive from a standard Ecto query).

It's worth noting that the Ecto documentation recommends using string keys at all times anyway.