@callback avg!(Ash.Query.t(), field :: atom(), opts :: Keyword.t()) ::
  term() | no_return()

Get the avg of a given field from the given query, raising any errors

@callback avg(Ash.Query.t(), field :: atom(), opts :: Keyword.t()) ::
  {:ok, term()} | {:error, Ash.Error.t()}

Get the avg of a given field from the given query

@callback bulk_create!(
  [map()],
  resource :: Ash.Resource.t(),
  action :: atom(),
  params :: Keyword.t()
) :: Ash.BulkResult.t() | no_return()

Creates many records, raising on any errors. See bulk_create/2 for more.

• :upsert? (boolean/0) - If a conflict is found based on the primary key, the record is updated in the database (requires upsert support) The default value is false.

• :upsert_identity (atom/0) - The identity to use when detecting conflicts for upsert?, e.g. upsert_identity: :full_name. By default, the primary key is used. Has no effect if upsert?: true is not provided

• :upsert_fields (list of atom/0) - The fields to upsert. If not set, the action's upsert_fields is used. Unlike singular create, bulk_create with upsert? requires that upsert_fields be specified explicitly in one of these two locations.

• :sorted? (boolean/0) - Wether or not to sort results by their input position, in cases where return_records?: true was provided. The default value is false.

• :return_records? (boolean/0) - Wether or not to return all of the records that were inserted. Defaults to false to account for large inserts. The default value is false.

• :return_errors? (boolean/0) - Wether or not to return all of the errors that occur. Defaults to false to account for large inserts. The default value is false.

• :batch_size (pos_integer/0) - The number of records to include in each batch. Defaults to the default_limit or max_page_size of the action, or 100.

• :return_stream? (boolean/0) - If set to true, instead of an Ash.BulkResult, a mixed stream is returned.
  Potential elements:
  {:notification, notification} - if return_notifications? is set to true {:ok, record} - if return_records? is set to true {:error, error} - an error that occurred. May be changeset or an invidual error. The default value is false.

• :stop_on_error? (boolean/0) - If true, the first encountered error will stop the action and be returned. Otherwise, errors will be skipped. The default value is false.

• :notify? (boolean/0) - Wether or not to send notifications out. If this is set to true then the data layer must return the results from each batch. This may be intensive for large bulk actions. The default value is false.

• :transaction - Wether or not to wrap the entire execution in a transaction, each batch, or not at all.
  Keep in mind:
  before_transaction and after_transaction hooks attached to changesets will have to be run inside the transaction if you choose transaction: :all. The default value is :batch.

• :timeout (timeout/0) - A positive integer, or :infinity. If none is provided, the timeout configured on the api is used (which defaults to 30_000).

• :tracer (atom/0) - A tracer that implements the Ash.Tracer behaviour. See that module for more.

• :verbose? (boolean/0) - Log engine operations (very verbose!) The default value is false.

• :authorize? - If an actor option is provided (even if it is nil), authorization happens automatically. If not, this flag can be used to authorize with no user.

• :stacktraces? (boolean/0) - For Ash errors, whether or not each error has a stacktrace. See the error_handling guide for more. The default value is true.

• :tenant (term/0) - A tenant to set on the query or changeset

• :actor (term/0) - If an actor is provided, it will be used in conjunction with the authorizers of a resource to authorize access

• :return_notifications? (boolean/0) - Use this if you're running ash actions in your own transaction and you want notifications to happen still.
  If a transaction is ongoing, and this is false, notifications will be discarded, otherwise the return value is {:ok, result, notifications} (or {:ok, notifications})
  To send notifications later, use Ash.Notifier.notify(notifications). It sends any notifications that can be sent, and returns the rest. The default value is false.

• :notification_metadata (term/0) - Metadata to be merged into the metadata field for all notifications sent from this operation. The default value is %{}.

@callback bulk_create(
  [map()],
  resource :: Ash.Resource.t(),
  action :: atom(),
  params :: Keyword.t()
) ::
  Ash.BulkResult.t()
  | Enumerable.t(
      {:ok, Ash.Resource.record()}
      | {:error, Ash.Changeset.t() | Ash.Error.t()}
      | {:notification, Ash.Notifier.Notification.t()}
    )

Creates many records.

assumptions

Assumptions

We assume that the input is a list of changesets all for the same action, or a list of input maps for the same action with the :resource and :action option provided to illustrate which action it is for.

performance-feasibility

Performance/Feasibility

The performance of this operation depends on the data layer in question. Data layers like AshPostgres will choose reasonable batch sizes in an attempt to handle large bulk actions, but that does not mean that you can pass a list of 500k inputs and expect things to go off without a hitch (although it might). If you need to do large data processing, you should look into projects like GenStage and Broadway. With that said, if you want to do things like support CSV upload and you place some reasonable limits on the size this is a great tool. You'll need to test it yourself, YMMV.

Passing return_records?: true can significantly increase the time it takes to perform the operation, and can also make the operation completely unreasonable due to the memory requirement. If you want to do very large bulk creates and display all of the results, the suggestion is to annotate them with a "bulk_create_id" in the data layer, and then read the records with that bulk_create_id so that they can be retrieved later if necessary.

changes-validations

Changes/Validations

Changes will be applied in the order they are given on the actions as normal. Any change that exposes the bulk_change or bulk_validate callback will be applied on the entire list.

after-action-hooks

After Action Hooks

The following requirements must be met for after_action hooks to function properly. If they are not met, and an after_action hook being applied to a changeset in a change.

1. return_records? must be set to true.
2. The changeset must be setting the primary key as part of its changes, so that we know which result applies to which changeset.

It is possible to use after_action hooks with bulk_change/3, but you need to return the hooks along with the changesets. This allows for setting up after_action hooks that don't need access to the returned record, or after_action hooks that can operate on the entire list at once. See the documentation for that callback for more on how to do accomplish that.

• :upsert? (boolean/0) - If a conflict is found based on the primary key, the record is updated in the database (requires upsert support) The default value is false.

• :upsert_identity (atom/0) - The identity to use when detecting conflicts for upsert?, e.g. upsert_identity: :full_name. By default, the primary key is used. Has no effect if upsert?: true is not provided

• :upsert_fields (list of atom/0) - The fields to upsert. If not set, the action's upsert_fields is used. Unlike singular create, bulk_create with upsert? requires that upsert_fields be specified explicitly in one of these two locations.

• :sorted? (boolean/0) - Wether or not to sort results by their input position, in cases where return_records?: true was provided. The default value is false.

• :return_records? (boolean/0) - Wether or not to return all of the records that were inserted. Defaults to false to account for large inserts. The default value is false.

• :return_errors? (boolean/0) - Wether or not to return all of the errors that occur. Defaults to false to account for large inserts. The default value is false.

• :batch_size (pos_integer/0) - The number of records to include in each batch. Defaults to the default_limit or max_page_size of the action, or 100.

• :return_stream? (boolean/0) - If set to true, instead of an Ash.BulkResult, a mixed stream is returned.
  Potential elements:
  {:notification, notification} - if return_notifications? is set to true {:ok, record} - if return_records? is set to true {:error, error} - an error that occurred. May be changeset or an invidual error. The default value is false.

• :stop_on_error? (boolean/0) - If true, the first encountered error will stop the action and be returned. Otherwise, errors will be skipped. The default value is false.

• :notify? (boolean/0) - Wether or not to send notifications out. If this is set to true then the data layer must return the results from each batch. This may be intensive for large bulk actions. The default value is false.

• :transaction - Wether or not to wrap the entire execution in a transaction, each batch, or not at all.
  Keep in mind:
  before_transaction and after_transaction hooks attached to changesets will have to be run inside the transaction if you choose transaction: :all. The default value is :batch.

• :timeout (timeout/0) - A positive integer, or :infinity. If none is provided, the timeout configured on the api is used (which defaults to 30_000).

• :tracer (atom/0) - A tracer that implements the Ash.Tracer behaviour. See that module for more.

• :verbose? (boolean/0) - Log engine operations (very verbose!) The default value is false.

• :authorize? - If an actor option is provided (even if it is nil), authorization happens automatically. If not, this flag can be used to authorize with no user.

• :stacktraces? (boolean/0) - For Ash errors, whether or not each error has a stacktrace. See the error_handling guide for more. The default value is true.

• :tenant (term/0) - A tenant to set on the query or changeset

• :actor (term/0) - If an actor is provided, it will be used in conjunction with the authorizers of a resource to authorize access

• :return_notifications? (boolean/0) - Use this if you're running ash actions in your own transaction and you want notifications to happen still.
  If a transaction is ongoing, and this is false, notifications will be discarded, otherwise the return value is {:ok, result, notifications} (or {:ok, notifications})
  To send notifications later, use Ash.Notifier.notify(notifications). It sends any notifications that can be sent, and returns the rest. The default value is false.

• :notification_metadata (term/0) - Metadata to be merged into the metadata field for all notifications sent from this operation. The default value is %{}.

@callback can(
  action_or_query_or_changeset ::
    Ash.Query.t()
    | Ash.Changeset.t()
    | {Ash.Resource.t(), atom() | Ash.Resource.Actions.action()},
  actor :: term(),
  opts :: Keyword.t()
) :: {:ok, boolean() | :maybe} | {:error, term()}

Returns wether or not the user can perform the action, or :maybe, returning any errors.

In cases with "runtime" checks (checks after the action), we may not be able to determine an answer, and so the value :maybe will be returned from can/2. The can? function assumes that :maybe means false. Keep in mind, this is just for doing things like "can they do this" in a UI, so assuming :maybe is false is fine. The actual action invocation will be properly checked regardless. If you have runtime checks, you may need to use can instead of can?, or configure what :maybe means.

options

Options

• maybe_is - What to treat :maybe results as, defaults to true if using can?, or :maybe if using can.
• run_queries? - In order to determine authorization status for changesets that use filter checks, we may need to run queries (almost always only one query). Set this to false to disable (returning :maybe instead). The default value is true.
• data - A record or list of records. For authorizing reads with filter checks, this can be provided and a filter check will only be true if all records match the filter. This is detected by running a query.

@callback can?(
  query_or_changeset_or_action ::
    Ash.Query.t()
    | Ash.Changeset.t()
    | {Ash.Resource.t(), atom() | Ash.Resource.Actions.action()},
  actor :: term(),
  opts :: Keyword.t()
) :: boolean() | no_return()

Returns wether or not the user can perform the action, or raises on errors.

See can/3 for more info.

@callback create!(Ash.Changeset.t(), params :: Keyword.t()) ::
  Ash.Resource.record()
  | {Ash.Resource.record(), [Ash.Notifier.Notification.t()]}
  | no_return()

Create a record. See create/2 for more information.

@callback create(Ash.Changeset.t(), params :: Keyword.t()) ::
  {:ok, Ash.Resource.record()}
  | {:ok, Ash.Resource.record(), [Ash.Notifier.Notification.t()]}
  | {:error, term()}

Create a record.

• :upsert? (boolean/0) - If a conflict is found based on the primary key, the record is updated in the database (requires upsert support) The default value is false.

• :upsert_identity (atom/0) - The identity to use when detecting conflicts for upsert?, e.g. upsert_identity: :full_name. By default, the primary key is used. Has no effect if upsert?: true is not provided

• :upsert_fields (list of atom/0) - The fields to upsert. If not set, the action's upsert_fields is used, and if that is not set, then any fields not being set to defaults are written.

• :timeout (timeout/0) - A positive integer, or :infinity. If none is provided, the timeout configured on the api is used (which defaults to 30_000).

• :tracer (atom/0) - A tracer that implements the Ash.Tracer behaviour. See that module for more.

• :verbose? (boolean/0) - Log engine operations (very verbose!) The default value is false.

• :action (term/0) - The action to use, either an Action struct or the name of the action

• :authorize? - If an actor option is provided (even if it is nil), authorization happens automatically. If not, this flag can be used to authorize with no user.

• :stacktraces? (boolean/0) - For Ash errors, whether or not each error has a stacktrace. See the error_handling guide for more. The default value is true.

• :tenant (term/0) - A tenant to set on the query or changeset

• :actor (term/0) - If an actor is provided, it will be used in conjunction with the authorizers of a resource to authorize access

• :after_action (term/0) - A hook to be run just before the action returns, but before fields are selected (still inside the same transaction, if your data layer supports transactions). This is mostly important if you want to load calculations after the action, which depend on having fields selected, but you want to authorize with the minimal set of fields that are actually being selected. Runs only if the action is successful, and is passed the changeset and result of the action. Should return {:ok, result} or {:error, error}.
  For example, if you had a full_name calculation, but were only selecting, first_name and full_name, you might do something like this:

    MyApp.User
    |> Ash.Changeset.for_create(:create, %{first_name: "first_name", last_name: "last_name"}
    |> Ash.Changeset.select(:first_name))
    |> Api.create(after_action: fn _changeset, user -> Api.load(user, :full_name) end)

  If you tried to load that :full_name calculation after receiving the data, the last_name would not be selected and as such would not be usable in the calculation, regardless of whether or not the calculation includes that field in its select list.

• :return_notifications? (boolean/0) - Use this if you're running ash actions in your own transaction and you want notifications to happen still.
  If a transaction is ongoing, and this is false, notifications will be discarded, otherwise the return value is {:ok, result, notifications} (or {:ok, notifications})
  To send notifications later, use Ash.Notifier.notify(notifications). It sends any notifications that can be sent, and returns the rest. The default value is false.

• :notification_metadata (term/0) - Metadata to be merged into the metadata field for all notifications sent from this operation. The default value is %{}.

@callback destroy!(Ash.Changeset.t() | Ash.Resource.record(), params :: Keyword.t()) ::
  :ok
  | Ash.Resource.record()
  | [Ash.Notifier.Notification.t()]
  | {Ash.Resource.record(), [Ash.Notifier.Notification.t()]}
  | no_return()

Destroy a record. See destroy/2 for more information.

@callback destroy(Ash.Changeset.t() | Ash.Resource.record(), params :: Keyword.t()) ::
  :ok
  | {:ok, Ash.Resource.record()}
  | {:ok, [Ash.Notifier.Notification.t()]}
  | {:ok, Ash.Resource.record(), [Ash.Notifier.Notification.t()]}
  | {:error, term()}

Destroy a record.

• :return_destroyed? (boolean/0) - If true, the destroyed record is included in the return result, e.g {:ok, destroyed} or {:ok, destroyed, notifications} The default value is false.

• :timeout (timeout/0) - A positive integer, or :infinity. If none is provided, the timeout configured on the api is used (which defaults to 30_000).

• :tracer (atom/0) - A tracer that implements the Ash.Tracer behaviour. See that module for more.

• :verbose? (boolean/0) - Log engine operations (very verbose!) The default value is false.

• :action (term/0) - The action to use, either an Action struct or the name of the action

• :authorize? - If an actor option is provided (even if it is nil), authorization happens automatically. If not, this flag can be used to authorize with no user.

• :stacktraces? (boolean/0) - For Ash errors, whether or not each error has a stacktrace. See the error_handling guide for more. The default value is true.

• :tenant (term/0) - A tenant to set on the query or changeset

• :actor (term/0) - If an actor is provided, it will be used in conjunction with the authorizers of a resource to authorize access

• :return_notifications? (boolean/0) - Use this if you're running ash actions in your own transaction and you want notifications to happen still.
  If a transaction is ongoing, and this is false, notifications will be discarded, otherwise the return value is {:ok, result, notifications} (or {:ok, notifications})
  To send notifications later, use Ash.Notifier.notify(notifications). It sends any notifications that can be sent, and returns the rest. The default value is false.

• :notification_metadata (term/0) - Metadata to be merged into the metadata field for all notifications sent from this operation. The default value is %{}.

@callback first!(Ash.Query.t(), field :: atom(), opts :: Keyword.t()) ::
  term() | no_return()

Get the first of a given field from the given query, raising any errors

@callback first(Ash.Query.t(), field :: atom(), opts :: Keyword.t()) ::
  {:ok, term()} | {:error, Ash.Error.t()}

Get the first of a given field from the given query

@callback get!(
  resource :: Ash.Resource.t(),
  id_or_filter :: term(),
  params :: Keyword.t()
) :: Ash.Resource.record() | no_return()

Get a record by a primary key. See get/3 for more.

@callback get(
  resource :: Ash.Resource.t(),
  id_or_filter :: term(),
  params :: Keyword.t()
) :: {:ok, Ash.Resource.record()} | {:ok, nil} | {:error, term()}

Get a record by a primary key.

For a resource with a composite primary key, pass a keyword list, e.g MyApi.get(MyResource, first_key: 1, second_key: 2)

• :error? (boolean/0) - Whether or not an error should be returned or raised when the record is not found. If set to false, nil will be returned. The default value is true.

• :load (term/0) - Fields or relationships to load in the query. See Ash.Query.load/2

• :lock (term/0) - A lock statement to add onto the query

• :context (term/0) - Context to be set on the query being run

• :reselect_all? (boolean/0) - Wether or not to reselect all attributes depended on by loads. By default, we only reselect fields that weren't already selected. The default value is false.

• :timeout (timeout/0) - A positive integer, or :infinity. If none is provided, the timeout configured on the api is used (which defaults to 30_000).

• :tracer (atom/0) - A tracer that implements the Ash.Tracer behaviour. See that module for more.

• :verbose? (boolean/0) - Log engine operations (very verbose!) The default value is false.

• :action (term/0) - The action to use, either an Action struct or the name of the action

• :authorize? - If an actor option is provided (even if it is nil), authorization happens automatically. If not, this flag can be used to authorize with no user.

• :stacktraces? (boolean/0) - For Ash errors, whether or not each error has a stacktrace. See the error_handling guide for more. The default value is true.

• :tenant (term/0) - A tenant to set on the query or changeset

• :actor (term/0) - If an actor is provided, it will be used in conjunction with the authorizers of a resource to authorize access

@callback list!(Ash.Query.t(), field :: atom(), opts :: Keyword.t()) ::
  [term()] | no_return()

Get the list of a given field from the given query, raising any errors

@callback list(Ash.Query.t(), field :: atom(), opts :: Keyword.t()) ::
  {:ok, [term()]} | {:error, Ash.Error.t()}

Get list of a given field from the given query

@callback load!(
  record_or_records :: Ash.Resource.record() | [Ash.Resource.record()],
  query :: load_statement(),
  opts :: Keyword.t()
) :: Ash.Resource.record() | [Ash.Resource.record()] | no_return()

Load fields or relationships on already fetched records. See load/3 for more information.

@callback load(
  record_or_records :: Ash.Resource.record() | [Ash.Resource.record()],
  query :: load_statement(),
  opts :: Keyword.t()
) :: {:ok, Ash.Resource.record() | [Ash.Resource.record()]} | {:error, term()}

Load fields or relationships on already fetched records.

Accepts a list of non-loaded fields and loads them on the provided records or a query, in which case the loaded fields of the query are used. Relationship loads can be nested, for example: MyApi.load(record, [posts: [:comments]]).

• :lazy? (boolean/0) - If set to true, values will only be loaded if the related value isn't currently loaded. The default value is false.

• :reselect_all? (boolean/0) - Wether or not to reselect all attributes depended on by loads. By default, we only reselect fields that weren't already selected. The default value is false.

• :timeout (timeout/0) - A positive integer, or :infinity. If none is provided, the timeout configured on the api is used (which defaults to 30_000).

• :tracer (atom/0) - A tracer that implements the Ash.Tracer behaviour. See that module for more.

• :verbose? (boolean/0) - Log engine operations (very verbose!) The default value is false.

• :action (term/0) - The action to use, either an Action struct or the name of the action

• :authorize? - If an actor option is provided (even if it is nil), authorization happens automatically. If not, this flag can be used to authorize with no user.

• :stacktraces? (boolean/0) - For Ash errors, whether or not each error has a stacktrace. See the error_handling guide for more. The default value is true.

• :tenant (term/0) - A tenant to set on the query or changeset

• :actor (term/0) - If an actor is provided, it will be used in conjunction with the authorizers of a resource to authorize access

@callback max!(Ash.Query.t(), field :: atom(), opts :: Keyword.t()) ::
  term() | no_return()

Get the max of a given field from the given query, raising any errors

@callback max(Ash.Query.t(), field :: atom(), opts :: Keyword.t()) ::
  {:ok, term()} | {:error, Ash.Error.t()}

Get the max of a given field from the given query

@callback min!(Ash.Query.t(), field :: atom(), opts :: Keyword.t()) ::
  term() | no_return()

Get the min of a given field from the given query, raising any errors

@callback min(Ash.Query.t(), field :: atom(), opts :: Keyword.t()) ::
  {:ok, term()} | {:error, Ash.Error.t()}

Get the min of a given field from the given query

@callback page!(Ash.Page.page(), page_request()) :: Ash.Page.page() | no_return()

Fetch a page relative to the provided page.

@callback page(Ash.Page.page(), page_request()) ::
  {:ok, Ash.Page.page()} | {:error, term()}

Fetch a page relative to the provided page.

A page is the return value of a paginated action called via read/2.

@callback read!(Ash.Query.t() | Ash.Resource.t(), params :: Keyword.t()) ::
  [Ash.Resource.record()]
  | {[Ash.Resource.record()], Ash.Query.t()}
  | no_return()

Run an ash query. See read/2 for more.

@callback read(Ash.Query.t(), params :: Keyword.t()) ::
  {:ok, [Ash.Resource.record()]}
  | {:ok, [Ash.Resource.record()], Ash.Query.t()}
  | {:error, term()}

Run a query on a resource.

For more information on building a query, see Ash.Query.

• :page - Pagination options, see the pagination docs for more

• :load (term/0) - A load statement to add onto the query

• :lock (term/0) - A lock statement to add onto the query

• :return_query? (boolean/0) - If true, the query that was ultimately used is returned as a third tuple element.
  The query goes through many potential changes during a request, potentially adding authorization filters, or replacing relationships for other data layers with their corresponding ids. This option can be used to get the true query that was sent to the data layer. The default value is false.

• :reselect_all? (boolean/0) - Wether or not to reselect all attributes depended on by loads. By default, we only reselect fields that weren't already selected. The default value is false.

• :timeout (timeout/0) - A positive integer, or :infinity. If none is provided, the timeout configured on the api is used (which defaults to 30_000).

• :tracer (atom/0) - A tracer that implements the Ash.Tracer behaviour. See that module for more.

• :verbose? (boolean/0) - Log engine operations (very verbose!) The default value is false.

• :action (term/0) - The action to use, either an Action struct or the name of the action

• :authorize? - If an actor option is provided (even if it is nil), authorization happens automatically. If not, this flag can be used to authorize with no user.

• :stacktraces? (boolean/0) - For Ash errors, whether or not each error has a stacktrace. See the error_handling guide for more. The default value is true.

• :tenant (term/0) - A tenant to set on the query or changeset

• :actor (term/0) - If an actor is provided, it will be used in conjunction with the authorizers of a resource to authorize access

pagination

Pagination

Limit/offset pagination

• :offset (non_neg_integer/0) - The number of records to skip from the beginning of the query

• :limit (pos_integer/0) - The number of records to include in the page

• :filter (term/0) - A filter to apply for pagination purposes, that should not be considered in the full count.
  This is used by the liveview paginator to only fetch the records that were already on the page when refreshing data, to avoid pages jittering.

• :count (boolean/0) - Whether or not to return the page with a full count of all records

Keyset pagination

• :before (String.t/0) - Get records that appear before the provided keyset (mutually exclusive with after)

• :after (String.t/0) - Get records that appear after the provided keyset (mutually exclusive with before)

• :limit (pos_integer/0) - How many records to include in the page

• :filter (term/0) - See the filter option for offset pagination, this behaves the same.

• :count (boolean/0) - Whether or not to return the page with a full count of all records

@callback read_one!(Ash.Query.t() | Ash.Resource.t(), params :: Keyword.t()) ::
  Ash.Resource.record()
  | {Ash.Resource.record(), Ash.Query.t()}
  | nil
  | no_return()

Run an ash query, raising on more than one result. See read_one/2 for more.

@callback read_one(Ash.Query.t() | Ash.Resource.t(), params :: Keyword.t()) ::
  {:ok, Ash.Resource.record()}
  | {:ok, Ash.Resource.record(), Ash.Query.t()}
  | {:ok, nil}
  | {:error, term()}

Run a query on a resource, but fail on more than one result.

This is useful if you have a query that doesn't include a primary key but you know that it will only ever return a single result.

@callback reload!(record :: Ash.Resource.record(), params :: Keyword.t()) ::
  Ash.Resource.record() | no_return()

Refetches a record by primary key. See reload/1 for more.

@callback reload(record :: Ash.Resource.record()) ::
  {:ok, Ash.Resource.record()} | {:error, term()}

Refetches a record by primary key.

@callback run_action!(input :: Ash.ActionInput.t(), opts :: Keyword.t()) ::
  term() | no_return()

Runs a generic action, raising on errors

@callback run_action(input :: Ash.ActionInput.t(), opts :: Keyword.t()) ::
  {:ok, term()} | {:error, term()}

Runs a generic action

@callback stream!(Ash.Query.t(), opts :: Keyword.t()) ::
  Enumerable.t(Ash.Resource.record())

Streams the results of a query.

This utilizes keyset pagination to accomplish this stream, and for that reason, the action for the query must support keyset pagination.

@callback sum!(Ash.Query.t(), field :: atom(), opts :: Keyword.t()) ::
  term() | no_return()

Get the sum of a given field from the given query, raising any errors

@callback sum(Ash.Query.t(), field :: atom(), opts :: Keyword.t()) ::
  {:ok, term()} | {:error, Ash.Error.t()}

Get the sum of a given field from the given query

@callback update!(Ash.Changeset.t(), params :: Keyword.t()) ::
  Ash.Resource.record()
  | {Ash.Resource.record(), [Ash.Notifier.Notification.t()]}
  | no_return()

Update a record. See update/2 for more information.

@callback update(Ash.Changeset.t(), params :: Keyword.t()) ::
  {:ok, Ash.Resource.record()}
  | {:ok, Ash.Resource.record(), [Ash.Notifier.Notification.t()]}
  | {:error, term()}

Update a record.

• :timeout (timeout/0) - A positive integer, or :infinity. If none is provided, the timeout configured on the api is used (which defaults to 30_000).

• :tracer (atom/0) - A tracer that implements the Ash.Tracer behaviour. See that module for more.

• :verbose? (boolean/0) - Log engine operations (very verbose!) The default value is false.

• :action (term/0) - The action to use, either an Action struct or the name of the action

• :authorize? - If an actor option is provided (even if it is nil), authorization happens automatically. If not, this flag can be used to authorize with no user.

• :stacktraces? (boolean/0) - For Ash errors, whether or not each error has a stacktrace. See the error_handling guide for more. The default value is true.

• :tenant (term/0) - A tenant to set on the query or changeset

• :actor (term/0) - If an actor is provided, it will be used in conjunction with the authorizers of a resource to authorize access

• :after_action (term/0) - A hook to be run just before the action returns, but before fields are selected (still inside the same transaction, if your data layer supports transactions). This is mostly important if you want to load calculations after the action, which depend on having fields selected, but you want to authorize with the minimal set of fields that are actually being selected. Runs only if the action is successful, and is passed the changeset and result of the action. Should return {:ok, result} or {:error, error}.
  For example, if you had a full_name calculation, but were only selecting, first_name and full_name, you might do something like this:

    MyApp.User
    |> Ash.Changeset.for_create(:create, %{first_name: "first_name", last_name: "last_name"}
    |> Ash.Changeset.select(:first_name))
    |> Api.create(after_action: fn _changeset, user -> Api.load(user, :full_name) end)

  If you tried to load that :full_name calculation after receiving the data, the last_name would not be selected and as such would not be usable in the calculation, regardless of whether or not the calculation includes that field in its select list.

• :return_notifications? (boolean/0) - Use this if you're running ash actions in your own transaction and you want notifications to happen still.
  If a transaction is ongoing, and this is false, notifications will be discarded, otherwise the return value is {:ok, result, notifications} (or {:ok, notifications})
  To send notifications later, use Ash.Notifier.notify(notifications). It sends any notifications that can be sent, and returns the rest. The default value is false.

• :notification_metadata (term/0) - Metadata to be merged into the metadata field for all notifications sent from this operation. The default value is %{}.