Notifiers allow you to tap into create, update and destroy actions on a resource. Notifiers are called after the current transaction is committed, which solves a lot of problems that can happen from performing a certain kind of side effect in your action code.
A common example of one such issue is using Phoenix PubSub to notify another part of your app (often a LiveView or phoenix channel) of a change. If you send a message to another process while your transaction is still open, and that process tries to look up a record you just created, it won't find it yet, because your transaction is still open!
Notifiers are a solution for a certain kind of side effect, what we call "at most once" effects. An example is sending an event to an analytics system, or our pusbub example above. It is "okay" if the event is fired and some error in that process prevents it from being sent.
In these cases you are looking for something other than a notifier. For example, you may want to look into integrating https://hexdocs.pm/oban into your application, allowing you to commit a "job" in the same transaction as your changes, to be processed later.
Alternatively, you could look into using Reactor
, which is designed for writing "sagas" and has first-class support
for Ash via the AshReactor
extension.
If the notifier is also an extension, include it in the notifiers
key:
defmodule MyResource do
use Ash.Resource,
notifiers: [ExampleNotifier]
end
Configuring a notifier for a specific action or actions can be a great way to avoid complexity in the implementation of a notifier. It allows you to avoid doing things like pattern matching on the action, and treat it more like a change module, that does its work whenever it is called.
create :create do
notifiers [ExampleNotifier]
end
When your notifier is not an extension, and you want it to run on all actions, include it this way to avoid unnecessary compile time dependencies:
defmodule MyResource do
use Ash.Resource,
simple_notifiers: [ExampleNotifier]
end
Ash comes with a builtin pub_sub notifier: Ash.Notifier.PubSub
. See the module documentation for more.
A notifier is a simple extension that must implement a single callback notify/1
. Notifiers do not have to implement an Ash DSL extension, but they may in order to configure how that notifier should behave. See Ash.Notifier.Notification
for the currently available fields on a notification.
For more information on creating a DSL extension to configure your notifier, see the docs for Spark.Dsl.Extension
.
Notifiers should not do intensive synchronous work. If any heavy work needs to be done, they should delegate to something else to handle the notification, like sending it to a GenServer or GenStage.
defmodule ExampleNotifier do
use Ash.Notifier
def notify(%Ash.Notifier.Notification{resource: resource, action: %{type: :create}, actor: actor}) do
if actor do
Logger.info("#{actor.id} created a #{resource}")
else
Logger.info("A non-logged in user created a #{resource}")
end
end
end
Domain calls involving resources who's datalayer supports transactions (like Postgres), notifications are saved up and sent after the transaction is closed. For example, the domain call below ultimately results in many many database calls.
Post
|> Ash.Changeset.for_update(:update, %{})
|> Ash.Changeset.manage_relationship(:related_posts, [1, 2, 3], type: :append)
|> Ash.Changeset.manage_relationship(:related_posts, [4, 5], type: :remove)
|> Ash.Changeset.manage_relationship(:comments, [10], type: :append)
|> Ash.update!()
Ash.Changeset.manage_relationship
doesn't leverage bulk operations yet, so it performs the following operations:
- a read of the currently related posts
- a read of the currently related comments
- a creation of a post_link to relate to 1
- a creation of a post_link to relate to 2
- a creation of a post_link to relate to 3
- a destruction of the post_link related to 4
- a destruction of the post_link related to 5
- an update to comment 10, to set its
post_id
to this post
If all three of these resources have notifiers configured, we need to send a notification for each operation (notifications are not sent for reads). For data consistency reasons, if a data layer supports transactions, all writes are done in a transaction. However, if you try to read the record from the database that you have just received a notification about before the transaction has been closed, in a different process, the information will be wrong. For this reason, Ash accumulates notifications until they can be sent.
If you need to perform multiple operations against your resources in your own transaction, you will have to handle that case yourself. To support this, Ash.create/2
, Ash.update/2
and Ash.destroy/2
support a return_notifications?: true
option. This causes the domain call to return {:ok, result, notifications}
in the successful case. Here is an example of how you might use it.
result =
Ash.DataLayer.transaction(resource, fn ->
{:ok, something, notifications1} = create_something()
{:ok, result, notifications2} = create_another_thing(something)
{:ok, notifications3} = destroy_something(something)
{result, Enum.concat([notifications1, notifications2, notifications3])}
end)
case result do
{:ok, value, notifications} ->
Ash.Notifier.notify(notifications)
value
{:error, error} ->
handle_error(error)
end