if Code.ensure_loaded?(Ecto) do defmodule Phoenix.Sync.Writer do @moduledoc """ Provides [write-path sync](https://electric-sql.com/docs/guides/writes) support for Phoenix- or Plug-based apps. Imagine you're building an application on sync. You've used the [read-path sync utilities](../../../README.md#read-path-sync) to sync data into the front-end. If the client then changes the data locally, these writes can be batched up and sent back to the server. `#{inspect(__MODULE__)}` provides a principled way of ingesting these local writes and applying them to Postgres. In a way that works-with and re-uses your existing authorization logic and your existing `Ecto.Schema`s and `Ecto.Changeset` validation functions. This allows you to build instant, offline-capable applications that work with [local optimistic state](https://electric-sql.com/docs/guides/writes). ### Requirements `#{inspect(__MODULE__)}` requires `Ecto`. It's perfectly possible to use `Phoenix.Sync` without any database connection (syncing via a HTTP) but the write-path requires both a database and `Ecto`. ## Controller example For example, take a project management app that's using [@TanStack/db](https://github.com/TanStack/db) to batch up local optimistic writes and POST them to the `Phoenix.Controller` below: defmodule MutationController do use Phoenix.Controller, formats: [:json] alias #{inspect(__MODULE__)} alias #{inspect(__MODULE__)}.Format def mutate(conn, %{"transaction" => transaction} = _params) do user_id = conn.assigns.user_id {:ok, txid, _changes} = Writer.new() |> Writer.allow( Projects.Project, check: reject_invalid_params/1, load: &Projects.load_for_user(&1, user_id), validate: &Projects.Project.changeset/2 ) |> Writer.allow( Projects.Issue, # Use the sensible defaults: # validate: Projects.Issue.changeset/2 # etc. ) |> Writer.apply( transaction, Repo, format: Format.TanstackDB ) json(conn, %{txid: txid}) end end The controller constructs a `#{inspect(__MODULE__)}` instance and pipes it through a series of `allow/3` calls, registering functions against `Ecto.Schema`s (in this case `Projects.Project` and `Projects.Issue`) to validate and authorize each of these mutation operations before applying them as a single transaction. This controller can become a single, unified entry point for ingesting writes into your application. You can extend the pipeline with `allow/3` calls for every schema that you'd like to be able to ingest changes to. The [`check`, `load`, `validate`, etc. callbacks](#module-callbacks) to the allow function are designed to allow you to re-use your authorization and validation logic from your existing `Plug` middleware and `Ecto.Changeset` functions. > #### Warning {: .warning} > > The mutation operations received from clients MUST be considered as **untrusted**. > > Though the HTTP operation that uploaded them will have been authenticated and > authorized by your existing Plug middleware as usual, the actual content of the > request that is turned into writes against your database needs to be validated > very carefully against the privileges of the current user. > > That's what `#{inspect(__MODULE__)}` is for: specifying which resources can be > updated and registering functions to authorize and validate the mutation payload. ## Usage levels ([high](#module-high-level-usage), [mid](#module-mid-level-usage), [low](#module-low-level-usage-diy)) You don't need to use `#{inspect(__MODULE__)}` to ingest write operations using Phoenix. Phoenix already ships with primitives like `Ecto.Multi` and `c:Ecto.Repo.transaction/2`. However, `#{inspect(__MODULE__)}` provides: - a number of convienience functions that simplify ingesting mutation operations - a high-level pipeline that dries up a lot of common boilerplate and allows you to re-use your existing `Plug` and `Ecto.Changeset` logic ### High-level usage The controller example above uses a higher level pipeline that dries up common boilerplate, whilst still allowing flexibility and extensibility. You create an ingest pipeline by instantiating a `#{inspect(__MODULE__)}` instance and piping into `allow/3` and `apply/4` calls: {:ok, txid, _changes} = #{inspect(__MODULE__)}.new() |> #{inspect(__MODULE__)}.allow(MyApp.Todo) |> #{inspect(__MODULE__)}.allow(MyApp.OtherSchema) |> #{inspect(__MODULE__)}.apply(transaction, Repo, format: MyApp.MutationFormat) Or, instead of `apply/4` you can use seperate calls to `ingest/3` and then `transaction/2`. This allows you to ingest multiple formats, for example: {:ok, txid} = #{inspect(__MODULE__)}.new() |> #{inspect(__MODULE__)}.allow(MyApp.Todo) |> #{inspect(__MODULE__)}.ingest(changes, format: MyApp.MutationFormat) |> #{inspect(__MODULE__)}.ingest(other_changes, parser: &MyApp.MutationFormat.parse_other/1) |> #{inspect(__MODULE__)}.ingest(more_changes, parser: {MyApp.MutationFormat, :parse_more, []}) |> #{inspect(__MODULE__)}.transaction(MyApp.Repo) And at any point you can drop down / eject out to the underlying `Ecto.Multi` using `to_multi/1` or `to_multi/3`: multi = #{inspect(__MODULE__)}.new() |> #{inspect(__MODULE__)}.allow(MyApp.Todo) |> #{inspect(__MODULE__)}.to_multi(changes, format: MyApp.MutationFormat) # ... do anything you like with the multi ... {:ok, changes} = Repo.transaction(multi) {:ok, txid} = #{inspect(__MODULE__)}.txid(changes) ### Mid-level usage The pattern above uses a lower-level `transact/4` function. This abstracts the mechanical details of transaction management whilst still allowing you to handle and apply mutation operations yourself: {:ok, txid} = #{inspect(__MODULE__)}.transact( my_encoded_txn, MyApp.Repo, fn %{operation: :insert, relation: [_, "todos"], change: change} -> MyApp.Repo.insert(...) %{operation: :update, relation: [_, "todos"], data: data, change: change} -> MyApp.Repo.update(Ecto.Changeset.cast(...)) %{operation: :delete, relation: [_, "todos"], data: data} -> # we don't allow deletes... {:error, "invalid delete"} end, format: #{inspect(__MODULE__.Format.TanstackDB)}, timeout: 60_000 ) However, with larger applications, this flexibility can become tiresome as you end up repeating boilerplate and defining your own pipeline to authorize, validate and apply changes with the right error handling and return values. ### Low-level usage (DIY) For the more advanced cases, if you're comfortable parsing, validating and persisting changes yourself then the simplest way to use `#{inspect(__MODULE__)}` is to use `txid!/1` within `c:Ecto.Repo.transaction/2`: {:ok, txid} = MyApp.Repo.transaction(fn -> # ... save your changes to the database ... # Return the transaction id. #{inspect(__MODULE__)}.txid!(MyApp.Repo) end) This returns the database transaction ID that the changes were applied within. This allows you to return it to the client, which can then monitor the read-path sync stream to detect when the transaction syncs through. At which point the client can discard its local optimistic state. A convinient way of doing this is to parse the request data into a list of `#{inspect(__MODULE__)}.Operation`s using a `#{inspect(__MODULE__)}.Format`. You can then apply the changes yourself by matching on the operation data: {:ok, %Transaction{operations: operations}} = #{inspect(__MODULE__)}.parse_transaction( my_encoded_txn, format: #{inspect(__MODULE__.Format.TanstackDB)} ) {:ok, txid} = MyApp.Repo.transaction(fn -> Enum.each(txn.operations, fn %{operation: :insert, relation: [_, "todos"], change: change} -> # insert a Todo %{operation: :update, relation: [_, "todos"], data: data, change: change} -> # update a Todo %{operation: :delete, relation: [_, "todos"], data: data} -> # for example, if you don't want to allow deletes... raise "invalid delete" end) #{inspect(__MODULE__)}.txid!(MyApp.Repo) end, timeout: 60_000) ## Transactions The `txid` in the return value from `apply/4` and `txid/1` / `txid!/1` allows the Postgres transaction ID to be returned to the client in the response data. This allows clients to monitor the read-path sync stream and match on the arrival of the same transaction id. When the client receives this transaction id back through its sync stream, it knows that it can discard the local optimistic state for that transaction. (This is a more robust way of managing optimistic state than just matching on instance IDs, as it allows for local changes to be rebased on concurrent changes to the same date from other users). `#{inspect(__MODULE__)}` uses `Ecto.Multi`'s transaction update mechanism under the hood, which means that either all the operations in a client transaction are accepted or none are. See `to_multi/1` for how you can hook into the `Ecto.Multi` after applying your change data. > #### Compatibility {: .info} > > `#{inspect(__MODULE__)}` can only return transaction ids when connecting to > a Postgres database (a repo with `adapter: Ecto.Adapters.Postgres`). You can > use this module for other databases, but the returned txid will be `nil`. ## Client Libraries `#{inspect(__MODULE__)}` is not coupled to any particular client-side implementation. See Electric's [write pattern guides and example code](https://electric-sql.com/docs/guides/writes) for implementation strategies and examples. Instead, `#{inspect(__MODULE__)}` provides an adapter pattern where you can register a `format` adapter or `parser` function to parse the expected payload format from a client side library into the struct that `#{inspect(__MODULE__)}` expects. The currently supported format adapters are: - [TanStack/db](https://github.com/TanStack/db) "A reactive client store for building super fast apps on sync" Integration: #{inspect(__MODULE__)}.new() |> #{inspect(__MODULE__)}.ingest(mutation_data, format: #{inspect(__MODULE__.Format.TanstackDB)}) |> #{inspect(__MODULE__)}.transaction(Repo) ## Usage Much as every controller action must be authenticated, authorized and validated to prevent users writing invalid data or data that they do not have permission to modify, mutations **MUST** be validated for both correctness (are the given values valid?) and permissions (is the current user allowed to apply the given mutation?). This dual verification -- of data and permissions -- is performed by a pipeline of application-defined callbacks for every model that you allow writes to: - [`check`](#module-check) - a function that performs a "pre-flight" sanity check of the user-provided data in the mutation; this should just validate the data and not usually hit the database; checks are performed on all operations in a transaction before proceeding to the next steps in the pipeline; this allows for fast rejection of invalid data before performing more expensive operations - [`load`](#module-load) - a function that takes the original data and returns the existing model from the database, if it exists, for an update or delete operation - [`validate`](#module-validate) - create and validate an `Ecto.Changeset` from the source data and mutation changes; this is intended to be compatible with using existing schema changeset functions; note that, as per any changeset function, the validate function can perform both authorization and validation - [`pre_apply` and `post_apply`](#module-pre_apply-and-post_apply) - add arbitrary `Ecto.Multi` operations to the transaction based on the current operation See `apply/4` and the [Callbacks](#module-callbacks) for how the transaction is processed internally and how best to use these callback functions to express your app's authorization and validation requirements. Calling `new/0` creates an empty writer configuration with the given mutation parser. But this alone does not permit any mutations. In order to allow writes from clients you must call `allow/3` with a schema module and some callback functions. # create an empty writer configuration writer = #{inspect(__MODULE__)}.new() # allow writes to the `Todos.Todo` table # using `Todos.check_mutation/1` to validate mutation data before # touching the database writer = #{inspect(__MODULE__)}.allow(writer, Todos.Todo, check: &Todos.check_mutation/1) If the table name on the client differs from the Postgres table, then you can add a `table` option that specifies the client table name that this `allow/3` call applies to: # `client_todos` is the name of the `todos` table on the clients writer = #{inspect(__MODULE__)}.allow( writer, Todos.Todo, validate: &Todos.validate_mutation/2, table: "client_todos" ) ## Callbacks ### Check The `check` option should be a 1-arity function whose purpose is to test the mutation data against the authorization rules for the application and model before attempting any database access. If the changes are valid then it should return `:ok` or `{:error, reason}` if they're invalid. If any of the changes fail the auth test, then the entire transaction will be rejected. This is the first line of defence against malicious writes as it provides a quick check of the data from the clients before any reads or writes to the database. Note that the writer pipeline checks all the operations before proceeding to load, validate and apply each operation in turn. def check(%#{inspect(__MODULE__.Operation)}{} = operation) do # :ok or {:error, "..."} end ### Load The `load` callback takes the `data` in `update` or `delete` mutations (i.e.: the original data before changes), and uses it to retrieve the original `Ecto.Struct` model from the database. It can be a 1- or 2-arity function. The 1-arity version receives just the `data` parameters. The 2-arity version receives the `Ecto.Repo` that the transaction is being applied to and the `data` parameters. # 1-arity version def load(%{"column" => "value"} = data) do # Repo.get(...) end # 2-arity version def load(repo, %{"column" => "value"} = data) do # repo.get(...) end If not provided defaults to using `c:Ecto.Repo.get_by/3` using the primary key(s) defined on the model. For `insert` operations this load function is not used. Instead, the original struct is created by calling the `__struct__/0` function on the `Ecto.Schema` module. ### Validate The `validate` callback performs the usual role of a changeset function: to validate the changes against the model's data constraints using the functions in `Ecto.Changeset`. It should return an `Ecto.Changeset` instance (or possibly the original schema struct in the case of `delete`s). If any of the transaction's changeset's are marked as invalid, then the entire transaction is aborted. If not specified, the `validate` function is defaulted to the schema model's standard `changeset/2` function if available. The callback can be either a 2- or 3-arity function. The 2-arity version will receive the `Ecto.Schema` struct returned from the `load` function and the mutation changes. The 3-arity version will receive the `load`ed struct, the changes and the operation. # 2-arity version def changeset(%Todo{} = data, %{} = changes) do data |> Ecto.Changeset.cast(changes, [:title, :completed]) |> Ecto.Changeset.validate_required(changes, [:title]) end # 3-arity version def changeset(%Todo{} = data, %{} = changes, operation) when operation in [:insert, :update, :delete] do # ... end #### Primary keys Whether the params for insert operations contains a value for the new primary key is application specific. It's certainly not required if you have declared your `Ecto.Schema` model with a primary key set to `autogenerate: true`. It's worth noting that if you are accepting primary key values as part of your `insert` changes, then you should use UUID primary keys for your models to prevent conflicts. ### pre_apply and post_apply These callbacks, run before or after the actual `insert`, `update` or `delete` operation allow for the addition of side effects to the transaction. They are passed an empty `Ecto.Multi` struct and which is then [merged](`Ecto.Multi.merge/2`) into the writer's transaction. They also allow for more validation/authorization steps as any operation within the callback that returns an "invalid" operation will abort the entire transaction. def pre_or_post_apply(%Ecto.Multi{} = multi, %Ecto.Changeset{} = change, %#{inspect(__MODULE__)}.Context{} = context) do multi # add some side-effects # |> Ecto.Multi.run(#{inspect(__MODULE__)}.operation_name(context, :image), fn _changes -> # with :ok <- File.write(image.name, image.contents) do # {:ok, nil} # end # end) # # validate the current transaction and abort using an {:error, value} tuple # |> Ecto.Multi.run(#{inspect(__MODULE__)}.operation_name(context, :my_validation), fn _changes -> # {:error, "reject entire transaction"} # end) end Note the use of `operation_name/2` when adding operations. Every name in the final `Ecto.Multi` struct must be unique, `operation_name/2` generates names that are guaranteed to be unique to the current operation and callback. ### Per-operation callbacks If you want to re-use an existing function on a per-operation basis, then in your write configuration you can define both top-level and per operation callbacks: #{inspect(__MODULE__)}.allow( Todos.Todo, load: &Todos.fetch_for_user(&1, user_id), check: &Todos.check_mutation(&1, user_id), validate: &Todos.Todo.changeset/2, update: [ # for inserts and deletes, use &Todos.Todo.changeset/2 but for updates # use this function validate: &Todos.Todo.update_changeset/2, pre_apply: &Todos.pre_apply_update_todo/3 ], insert: [ # optional validate, pre_apply and post_apply # overrides for insert operations ], delete: [ # optional validate, pre_apply and post_apply # overrides for delete operations ], ) ## End-to-end usage The combination of the `check`, `load`, `validate`, `pre_apply` and `post_apply` functions can be composed to provide strong guarantees of validity. The aim is to provide an interface as similar to that used in controller functions as possible. Here we show an example controller module that allows updating of `Todo`s via a standard `HTTP PUT` update handler and also via `HTTP POST`s to the `mutation` handler which applies optimistic writes via this module. We use the `load` function to validate the ownership of the original `Todo` by looking up the data using both the `id` and the `user_id`. This makes it impossible for user `a` to update todos belonging to user `b`. defmodule MyController do use Phoenix.Controller, formats: [:html, :json] alias #{inspect(__MODULE__)} # The standard HTTP PUT update handler def update(conn, %{"todo" => todo_params}) do user_id = conn.assigns.user_id with {:ok, todo} <- fetch_for_user(params, user_id), {:ok, params} <- validate_params(todo, todo_params, user_id), {:ok, updated_todo} <- Todos.update(todo, params) do redirect(conn, to: ~p"/todos/\#{updated_todo.id}") end end # The HTTP POST mutations handler which receives JSON data def mutations(conn, %{"transaction" => transaction} = _params) do user_id = conn.assigns.user_id {:ok, txid, _changes} = Writer.new() |> Writer.allow( Todos.Todo, check: &validate_mutation(&1, user_id), load: &fetch_for_user(&1, user_id), ) |> Writer.apply(transaction, Repo, format: Writer.Format.TanstackDB) json(conn, %{txid: txid}) end # Included here for completeness but in a real app would be a # public function in the Todos context. # Because we're validating the ownership of the Todo here we add an # extra layer of auth checks, preventing one user from modifying # the Todos of another. defp fetch_for_user(%{"id" => id}, user_id) do from(t in Todos.Todo, where: t.id == ^id and t.user_id == ^user_id) |> Repo.one() end defp validate_mutation(%Writer.Operation{} = op, user_id) do with :ok <- validate_params(op.data, user_id) do validate_params(op.changes, user_id) end end defp validate_params(%{"user_id" => user_id}, user_id), do: :ok defp validate_params(%{} = _params, _user_id), do: {:error, "invalid user_id"} end Because `Phoenix.Sync.Write` leverages `Ecto.Multi` to do the work of applying changes and managing errors, you're also free to extend the actions that are performed with every transaction using `pre_apply` and `post_apply` callbacks configured per-table or per-table per-action (insert, update, delete). See `allow/3` for more information on the configuration options for each table. The result of `to_multi/1` or `to_multi/3` is an `Ecto.Multi` instance so you can also just append operations using the normal `Ecto.Multi` functions: {:ok, txid, _changes} = Writer.new() |> Writer.allow(Todo, ...) |> Writer.to_multi(transaction, parser: &my_transaction_parser/1) |> Ecto.Multi.insert(:my_action, %Event{}) |> Writer.transaction(Repo) """ import Kernel, except: [apply: 2, apply: 3] defmodule Context do @moduledoc """ Provides context within callback functions. ## Fields * `index` - the 0-indexed position of the current operation within the transaction * `changes` - the current `Ecto.Multi` changes so far * `operation` - the current `#{inspect(__MODULE__.Operation)}` * `callback` - the name of the current callback, `:pre_apply` or `:post_apply` * `schema` - the `Ecto.Schema` module associated with the current operation """ @derive {Inspect, except: [:writer, :action]} defstruct [:index, :writer, :changes, :operation, :callback, :schema, :action, :pk] end defmodule Error do @moduledoc false defexception [:message, :operation] @impl true def message(e) do "Operation #{inspect(e.operation)} failed: #{e.message}" end end require Record require Logger Record.defrecordp(:opkey, schema: nil, operation: nil, index: 0, pk: nil) alias __MODULE__.Operation alias __MODULE__.Transaction alias __MODULE__.Format @type pre_post_func() :: (Ecto.Multi.t(), Ecto.Changeset.t(), context() -> Ecto.Multi.t()) @operation_options [ validate: [ type: {:fun, 2}, doc: """ A 2-arity function that returns a changeset for the given mutation data. Arguments: - `schema` the original `Ecto.Schema` model returned from the `load` function - `changes` a map of changes from the mutation operation Return value: - an `Ecto.Changeset` """, type_spec: quote(do: (Ecto.Schema.t(), data() -> Ecto.Changeset.t())) ], pre_apply: [ type: {:fun, 3}, doc: """ An optional callback that allows for the pre-pending of operations to the `Ecto.Multi`. Arguments and return value as per the global `pre_apply` callback. """, type_spec: quote(do: pre_post_func()), type_doc: "`t:pre_post_func/0`" ], post_apply: [ type: {:fun, 3}, doc: """ An optional callback that allows for the appending of operations to the `Ecto.Multi`. Arguments and return value as per the global `post_apply` callback. """, type_spec: quote(do: pre_post_func()), type_doc: "`t:pre_post_func/0`" ] ] @operation_options_schema NimbleOptions.new!(@operation_options) @type data() :: %{binary() => any()} @type mutation() :: %{required(binary()) => any()} @type operation() :: :insert | :update | :delete @operations [:insert, :update, :delete] @type txid() :: Transaction.id() @type context() :: %Context{ index: non_neg_integer(), changes: Ecto.Mult.changes(), schema: Ecto.Schema.t(), operation: :insert | :update | :delete, callback: :load | :validate | :pre_apply | :post_apply } @type operation_opts() :: unquote([NimbleOptions.option_typespec(@operation_options_schema)]) @operation_schema [ type: :keyword_list, keys: @operation_options, doc: NimbleOptions.docs(NimbleOptions.new!(@operation_options)), type_spec: quote(do: operation_opts()), type_doc: "`t:operation_opts/0`" ] @parse_schema_options [ format: [ type: :atom, doc: """ A module implementing the `#{inspect(__MODULE__.Format)}` behaviour. See `#{inspect(__MODULE__.Format)}`. """, type_spec: quote(do: Format.t()), type_doc: "[`Format.t()`](`t:#{inspect(Format)}.t/0`)" ], parser: [ type: {:or, [{:fun, 1}, :mfa]}, doc: """ A function that parses some input data and returns a [`%Transaction{}`](`#{inspect(__MODULE__.Transaction)}`) struct or an error. See `c:#{inspect(__MODULE__.Format)}.parse_transaction/1`. """, type_doc: "`#{inspect(Format)}.parser_fun() | mfa()`", type_spec: quote(do: Format.parser_fun()) ] ] @parse_schema NimbleOptions.new!(@parse_schema_options) @allow_schema NimbleOptions.new!( table: [ type: {:or, [:string, {:list, :string}]}, doc: """ Override the table name of the `Ecto.Schema` struct to allow for mapping between table names on the client and within Postgres. If you pass just a table name, then any schema prefix in the client tables is ignored, so Writer.allow(Todos, table: "todos") will match client operations for `["public", "todos"]` and `["application", "todos"]` etc. If you provide a 2-element list then the mapping will be exact and only client relations matching the full `[schema, table]` pair will match the given schema. Writer.allow(Todos, table: ["public", "todos"]) Will match client operations for `["public", "todos"]` but **not** `["application", "todos"]` etc. Defaults to `Model.__schema__(:source)`, or if the Ecto schema module has specified a `namespace` `[Model.__schema__(:prefix), Model.__schema__(:source)]`. """, type_doc: "`String.t() | [String.t(), ...]`", type_spec: quote(do: String.t() | [String.t(), ...]) ], accept: [ type: {:list, {:in, @operations}}, doc: """ A list of actions to accept. A transaction containing an operation not in the accept list will be rejected. Defaults to accepting all operations, `#{inspect(@operations)}`. """, type_spec: quote(do: [operation(), ...]) ], check: [ type: {:fun, 1}, doc: """ A function that validates every %#{inspect(__MODULE__.Operation)}{} in the transaction for correctness. This is run before any database access is performed and so provides an efficient way to prevent malicious writes without hitting your database. Defaults to a function that allows all operations: `fn _ -> :ok end`. """, type_spec: quote(do: (Operation.t() -> :ok | {:error, term()})) ], before_all: [ type: {:fun, 1}, doc: """ Run only once (per transaction) after the parsing and `check` callback have completed and before `load` and `validate` functions run. Useful for pre-loading data from the database that can be shared across all operation callbacks for all the mutations. Arguments: - `multi` an `Ecto.Multi` struct Return value: - `Ecto.Multi` struct with associated data Defaults to no callback. """, type_spec: quote(do: (Ecto.Multi.t() -> Ecto.Multi.t())) ], load: [ type: {:or, [{:fun, 1}, {:fun, 2}]}, doc: """ A 1- or 2-arity function that accepts either the mutation operation's data or an `Ecto.Repo` instance and the mutation data and returns the original row from the database. Arguments: - `repo` the `Ecto.Repo` instance passed to `apply/4` or `transaction/3` - `data` the original operation data Valid return values are: - `struct()` - an `Ecto.Schema` struct, that must match the module passed to `allow/3` - `{:ok, struct()}` - as above but wrapped in an `:ok` tuple - `nil` - if no row matches the search criteria, or - `{:error, String.t()}` - as `nil` but with a custom error string A return value of `nil` or `{:error, reason}` will abort the transaction. This function is only used for updates or deletes. For inserts, the `__struct__/0` function defined by `Ecto.Schema` is used to create an empty schema struct. ## Examples # load from a known Repo load: fn %{"id" => id} -> MyApp.Repo.get(Todos.Todo, id) # load from the repo passed to `#{__MODULE__}.transaction/2` load: fn repo, %{"id" => id} -> repo.get(Todos.Todo, id) If not provided defaults to `c:Ecto.Repo.get_by/3` using the table's schema module and its primary keys. """, type_spec: quote( do: (Ecto.Repo.t(), data() -> Ecto.Schema.t() | {:ok, Ecto.Schema.t()} | nil | {:error, String.t()}) | (data() -> Ecto.Schema.t() | {:ok, Ecto.Schema.t()} | nil | {:error, String.t()}) ) ], validate: [ type: {:or, [{:fun, 2}, {:fun, 3}]}, doc: """ a 2- or 3-arity function that returns an `Ecto.Changeset` for a given mutation. ### Callback params - `data` an Ecto.Schema struct matching the one used when calling `allow/2` returned from the `load` function. - `changes` a map of changes to apply to the `data`. - `operation` (for 3-arity callbacks only) the operation action, one of `:insert`, `:update` or `:delete` At absolute minimum, this should call `Ecto.Changeset.cast/3` to validate the proposed data: def my_changeset(data, changes, _operation) do Ecto.Changeset.cast(data, changes, @permitted_columns) end Defaults to the given model's `changeset/2` function if defined, raises if no changeset function can be found. """, type_spec: quote( do: (Ecto.Schema.t(), data() -> Ecto.Changeset.t()) | (Ecto.Schema.t(), data(), operation() -> Ecto.Changeset.t()) ) ], pre_apply: [ type: {:fun, 3}, doc: """ an optional callback that allows for the pre-pending of operations to the `Ecto.Multi` representing a mutation transaction. If should be a 3-arity function. ### Arguments - `multi` - an empty `%Ecto.Multi{}` instance that you should apply your actions to - `changeset` - the changeset representing the individual mutation operation - `context` - the current change [context](`#{__MODULE__.Context}`) The result should be the `Ecto.Multi` instance which will be [merged](`Ecto.Multi.merge/2`) with the one representing the mutation operation. Because every action in an `Ecto.Multi` must have a unique key, we advise using the `operation_name/2` function to generate a unique operation name based on the `context`. def pre_apply(multi, changeset, context) do name = #{inspect(__MODULE__)}.operation_name(context, :event_insert) Ecto.Multi.insert(multi, name, %Event{todo_id: id}) end Defaults to no `nil`. """, type_spec: quote(do: pre_post_func()), type_doc: "`t:pre_post_func/0`" ], post_apply: [ type: {:fun, 3}, doc: """ an optional callback function that allows for the appending of operations to the `Ecto.Multi` representing a mutation transaction. See the docs for `:pre_apply` for the function signature and arguments. Defaults to no `nil`. """, type_spec: quote(do: pre_post_func()), type_doc: "`t:pre_post_func/0`" ], insert: Keyword.put(@operation_schema, :doc, """ Callbacks for validating and modifying `insert` operations. Accepts definitions for the `validate`, `pre_apply` and `post_apply` functions for `insert` operations that will override the top-level equivalents. See the documentation for `allow/3`. The only difference with these callback functions is that the `action` parameter is redundant and therefore not passed. Defaults to `[]`, using the top-level functions for all operations. """), update: Keyword.put(@operation_schema, :doc, """ Callbacks for validating and modifying `update` operations. See the documentation for `insert`. """), delete: Keyword.update!( @operation_schema, :keys, &Keyword.put(&1, :validate, type: {:or, [{:fun, 1}, {:fun, 2}]}) ) |> Keyword.put(:doc, """ Callbacks for validating and modifying `delete` operations. See the documentation for `insert`. """) ) defstruct ingest: [], mappings: %{} @type allow_opts() :: [unquote(NimbleOptions.option_typespec(@allow_schema))] @type parse_opts() :: [unquote(NimbleOptions.option_typespec(@parse_schema))] @type schema_config() :: %{required(atom()) => term()} @type ingest_change() :: {Format.t(), Format.parser_fun(), Format.transaction_data()} @type repo_transaction_opts() :: keyword() @type transact_opts() :: [parse_opts() | repo_transaction_opts()] @type t() :: %__MODULE__{ ingest: [ingest_change()], mappings: %{(binary() | [binary(), ...]) => schema_config()} } @doc """ Create a new empty writer. Empty writers will reject writes to any tables. You should configure writes to the permitted tables by calling `allow/3`. """ @spec new() :: t() def new do %__MODULE__{} end @doc """ Allow writes to the given `Ecto.Schema`. Only tables specified in calls to `allow/3` will be accepted by later calls to `transaction/3`. Any changes to tables not explicitly defined by `allow/3` calls will be rejected and cause the entire transaction to be rejected. ## Examples # allow writes to the Todo table using # `MyApp.Todos.Todo.check_mutation/1` to validate operations #{inspect(__MODULE__)}.new() |> #{inspect(__MODULE__)}.allow( MyApp.Todos.Todo, check: &MyApp.Todos.check_mutation/1 ) # A more complex configuration adding an `post_apply` callback to inserts # and using a custom query to load the original database value. #{inspect(__MODULE__)}.new() |> #{inspect(__MODULE__)}.allow( MyApp.Todos..Todo, load: &MyApp.Todos.get_for_mutation/1, check: &MyApp.Todos.check_mutation/1, insert: [ post_apply: &MyApp.Todos.post_apply_insert_mutation/3 ] ) ## Supported options #{NimbleOptions.docs(@allow_schema)} """ @spec allow(t(), module(), allow_opts()) :: t() def allow(writer, schema, opts \\ []) def allow(%__MODULE__{} = writer, schema, opts) when is_atom(schema) do {schema, table, pks} = validate_schema!(schema) config = NimbleOptions.validate!(opts, @allow_schema) key = config[:table] || table load_fun = load_fun(schema, pks, opts) check_fun = check_fun(opts) accept = Keyword.get(config, :accept, @operations) |> MapSet.new() table_config = %{ schema: schema, table: table, pks: pks, accept: accept, check: check_fun } table_config = Enum.reduce(@operations, table_config, fn action, table_config -> Map.put( table_config, action, action_config(schema, config, action, load: load_fun, table: key, pks: pks) ) end) Map.update!(writer, :mappings, &Map.put(&1, key, table_config)) end defp validate_schema!(module) do if !Code.ensure_loaded?(module), do: raise(ArgumentError, message: "Unknown module #{module}") if !(function_exported?(module, :__changeset__, 0) && function_exported?(module, :__schema__, 1)), do: raise(ArgumentError, message: "Not an Ecto.Schema module #{module}") table = if prefix = module.__schema__(:prefix), do: [prefix, module.__schema__(:source)], else: module.__schema__(:source) {module, table, module.__schema__(:primary_key)} end defp action_config(schema, config, action, extra) do validate_fun = get_in(config, [action, :validate]) || config[:validate] || default_changeset!(schema) || raise(ArgumentError, message: "No validate/3 or validate/2 defined for #{action}s") # nil-hooks are just ignored pre_apply_fun = get_in(config, [action, :pre_apply]) || config[:pre_apply] post_apply_fun = get_in(config, [action, :post_apply]) || config[:post_apply] before_all_fun = config[:before_all] Map.merge( Map.new(extra), %{ schema: schema, before_all: before_all_fun, validate: validate_fun, pre_apply: pre_apply_fun, post_apply: post_apply_fun } ) end defp load_fun(schema, pks, config) do load_fun = case config[:load] do nil -> fn repo, change -> key = Map.new(pks, fn col -> {col, Map.fetch!(change, to_string(col))} end) repo.get_by(schema, key) end fun when is_function(fun, 1) -> fn _repo, change -> fun.(change) end fun when is_function(fun, 2) -> fun _invalid -> raise(ArgumentError, message: "`load` should be a 1- or 2-arity function") end fn _repo, :insert, _change, _ctx -> schema.__struct__() repo, _action, change, ctx -> # look for the data in the changes so far. uses the multi changes from # append_changeset_result/3 pk = Map.new(pks, fn col -> {to_string(col), change |> Map.fetch!(to_string(col)) |> to_string()} end) Enum.filter(ctx.changes, fn {opkey(schema: ^schema, pk: ^pk), _value} = change -> change _ -> nil end) |> Enum.max_by(fn {opkey(index: index), _} -> index end, &>=/2, fn -> nil end) |> case do nil -> # not in changeset, so call load load_fun.(repo, change) {opkey(operation: :delete), _} -> # last op on this key was a delete nil {_opkey, value} -> {:ok, value} end end end defp check_fun(opts) do case opts[:check] do nil -> fn _ -> :ok end fun1 when is_function(fun1, 1) -> fun1 fun when is_function(fun) -> info = Function.info(fun) raise ArgumentError, message: "Invalid check function. Expected a 1-arity function but got arity #{info[:arity]}" invalid -> raise ArgumentError, message: "Invalid check function. Expected a 1-arity function but got #{inspect(invalid)}" end end defp default_changeset!(schema) do cond do function_exported?(schema, :changeset, 3) -> &schema.changeset/3 function_exported?(schema, :changeset, 2) -> &schema.changeset/2 true -> nil end end @doc """ Add the given changes to the operations that will be applied within a `transaction/3`. Examples: {:ok, txid} = #{inspect(__MODULE__)}.new() |> #{inspect(__MODULE__)}.allow(MyApp.Todo) |> #{inspect(__MODULE__)}.ingest(changes, format: MyApp.MutationFormat) |> #{inspect(__MODULE__)}.ingest(other_changes, parser: &MyApp.MutationFormat.parse_other/1) |> #{inspect(__MODULE__)}.ingest(more_changes, parser: {MyApp.MutationFormat, :parse_more, []}) |> #{inspect(__MODULE__)}.transaction(MyApp.Repo) Supported options: #{NimbleOptions.docs(@parse_schema)} """ @spec ingest(t(), Format.transaction_data(), parse_opts()) :: t() def ingest(writer, changes, opts) do case validate_ingest_opts(opts) do {:ok, format, parser_fun} -> %{writer | ingest: [{format, parser_fun, changes} | writer.ingest]} {:error, message} -> raise Error, message: message end end @doc """ Ingest and write changes to the given repo in a single call. #{inspect(__MODULE__)}.new() |> #{inspect(__MODULE__)}.apply(changes, Repo, parser: &MyFormat.parse/1) is equivalent to: #{inspect(__MODULE__)}.new() |> #{inspect(__MODULE__)}.ingest(changes, parser: &MyFormat.parse/1) |> #{inspect(__MODULE__)}.transaction(Repo) """ @spec apply(t(), Format.transaction_data(), Ecto.Repo.t(), transact_opts()) :: {:ok, txid(), Ecto.Multi.changes()} | Ecto.Multi.failure() def apply(%__MODULE__{} = writer, changes, repo, opts) when is_atom(repo) and is_list(changes) do {writer_opts, txn_opts} = split_writer_txn_opts(opts) writer |> ingest(changes, writer_opts) |> transaction(repo, txn_opts) end @writer_option_keys Keyword.keys(@parse_schema_options) defp split_writer_txn_opts(opts) do Keyword.split(opts, @writer_option_keys) end @doc """ Given a writer configuration created using `allow/3` translate the list of mutations into an `Ecto.Multi` operation. Example: %Ecto.Multi{} = multi = #{inspect(__MODULE__)}.new() |> #{inspect(__MODULE__)}.allow(MyApp.Todos.Todo, check: &my_check_function/1) |> #{inspect(__MODULE__)}.allow(MyApp.Options.Option, check: &my_check_function/1) |> #{inspect(__MODULE__)}.ingest(changes, format: #{inspect(__MODULE__.Format.TanstackDB)}) |> #{inspect(__MODULE__)}.to_multi() If you want to add extra operations to the mutation transaction, beyond those applied by any `pre_apply` or `post_apply` callbacks in your mutation config then use the functions in `Ecto.Multi` to do those as normal. Use `transaction/3` to apply the changes to the database and return the transaction id. `to_multi/1` builds an `Ecto.Multi` struct containing the operations required to write the mutation operations to the database. The order of operation is: ### 1. Parse The transaction data is parsed, using either the `format` or the `parser` function supplied in `ingest/3`. ### 2. Check The user input data in each operation in the transaction is tested for validity via the `check` function. At this point no database operations have taken place. Errors at the parse or `check` stage result in an early exit. The purpose of the `check` callback is sanity check the incoming mutation data against basic sanitization rules, much as you would do with `Plug` middleware and controller params pattern matching. Now that we have a list of validated mutation operations, the next step is: ### 3. Before-all Perform any actions defined in the `before_all` callback. This only happens once per transaction, the first time the model owning the callback is included in the operation list. The following actions happen once per operation in the transaction: ### 4. Load The `load` function is called to retrieve the source row from the database (for `update` and `delete` operations), or the schema's `__struct__/0` function is called to instantiate an empty struct (`insert`). ### 5. Validate The `validate` function is called with the result of the `load` function and the operation's changes. ### 6. Pre-apply The `pre_apply` callback is called with a `multi` instance, the result of the `validate` function and the current `Context`. The result is [merged](`Ecto.Multi.merge/2`) into the transaction's ongoing `Ecto.Multi`. ### 7. Apply The actual operation is applied to the database using one of `Ecto.Multi.insert/4`, `Ecto.Multi.update/4` or `Ecto.Multi.delete/4`, and ### 8. Post-apply Finally the `post_apply` callback is called. Any error in any of these stages will abort the entire transaction and leave your database untouched. """ @spec to_multi(t()) :: Ecto.Multi.t() def to_multi(%__MODULE__{} = writer) do # Want to return a multi here but have that multi fail without contacting # the db if any of the check calls fail. # # Looking at `Ecto.Multi.__apply__/4` it first checks for invalid # operations before doing anything. So i can just add an error to a blank # multi and return that and the transaction step will fail before touching # the repo. writer.ingest |> Enum.reverse() |> Enum.reduce_while(start_multi(), fn {_format, parser_fun, changes}, multi -> with {:ok, %Transaction{} = txn} <- parse_check(writer, parser_fun, changes) do {:cont, Enum.reduce(txn.operations, multi, &append_multi(&2, &1, writer))} else {step, {:error, error}} -> {:halt, Ecto.Multi.error(Ecto.Multi.new(), step, error)} end end) end @doc """ Ingest changes and map them into an `Ecto.Multi` instance ready to apply using `#{inspect(__MODULE__)}.transaction/3` or `c:Ecto.Repo.transaction/2`. This is a wrapper around `ingest/3` and `to_multi/1`. Example: %Ecto.Multi{} = multi = #{inspect(__MODULE__)}.new() |> #{inspect(__MODULE__)}.allow(MyApp.Todos.Todo, check: &my_check_function/1) |> #{inspect(__MODULE__)}.allow(MyApp.Options.Option, check: &my_check_function/1) |> #{inspect(__MODULE__)}.to_multi(changes, format: #{inspect(__MODULE__.Format.TanstackDB)}) """ @spec to_multi(t(), Format.transaction_data(), parse_opts()) :: Ecto.Multi.t() def to_multi(%__MODULE__{} = writer, changes, opts) do writer |> ingest(changes, opts) |> to_multi() end defp parse_check(writer, parser_fun, changes) do with {:parse, {:ok, %Transaction{} = txn}} <- {:parse, parser_fun.(changes)}, {:check, :ok} <- {:check, check_transaction(writer, txn)} do {:ok, txn} end end @doc """ Use the parser configured in the given [`Writer`](`#{inspect(__MODULE__)}`) instance to decode the given transaction data. This can be used to handle mutation operations explicitly: {:ok, txn} = #{inspect(__MODULE__)}.parse_transaction(my_json_tx_data, format: #{inspect(__MODULE__.Format.TanstackDB)}) {:ok, txid} = Repo.transaction(fn -> Enum.each(txn.operations, fn operation -> # do something wih the given operation # raise if something is wrong... end) # return the transaction id #{inspect(__MODULE__)}.txid!(Repo) end) """ @spec parse_transaction(Format.transaction_data(), parse_opts()) :: {:ok, Transaction.t()} | {:error, term()} def parse_transaction(changes, opts) do with {:ok, opts} <- NimbleOptions.validate(opts, @parse_schema), {:ok, _format, parser_fun} <- validate_ingest_opts(opts) do parser_fun.(changes) end end defp validate_ingest_opts(opts) do with {:ok, config} <- NimbleOptions.validate(opts, @parse_schema), format = Keyword.get(config, :format), parser = Keyword.get(config, :parser), {:ok, parser_func} <- parser_fun(format, parser) do {:ok, format, parser_func} end end defp parser_fun(format, parser) do case format_parser(format) do parser when is_function(parser, 1) -> {:ok, parser} nil -> case parser do nil -> {:error, "no valid format or parser"} parser when is_function(parser, 1) -> {:ok, parser} {m, f, a} when is_atom(m) and is_atom(f) and is_list(a) -> {:ok, fn changes -> Kernel.apply(m, f, [changes | a]) end} end {:error, _} = error -> error end end defp format_parser(nil), do: nil defp format_parser(format) when is_atom(format) do if Code.ensure_loaded?(format) && function_exported?(format, :parse_transaction, 1) do Function.capture(format, :parse_transaction, 1) else {:error, "#{inspect(format)} does not implement the #{inspect(__MODULE__.Format)} behaviour"} end end defp check_transaction(%__MODULE__{} = writer, %Transaction{} = txn) do Enum.reduce_while(txn.operations, :ok, fn operation, :ok -> case mutation_actions(operation, writer) do {:ok, actions} -> %{check: check, accept: accept} = actions if MapSet.member?(accept, operation.operation) do case check.(operation) do :ok -> {:cont, :ok} {:error, reason} -> {:halt, {:error, %Error{message: reason, operation: operation}}} end else {:halt, {:error, %Error{ message: "Action #{inspect(operation.operation)} not in :accept list: #{MapSet.to_list(accept) |> inspect()}", operation: operation }}} end {:error, reason} -> {:halt, {:error, reason}} end end) end @txid_name {:__phoenix_sync__, :txid} @txid_query "SELECT pg_current_xact_id()::xid as txid" defp start_multi do Ecto.Multi.new() |> txid_step() end defp txid_step(multi \\ Ecto.Multi.new()) do Ecto.Multi.run(multi, @txid_name, fn repo, _ -> case repo.__adapter__() do Ecto.Adapters.Postgres -> txid(repo) adapter -> Logger.warning("Unsupported adapter #{adapter}. txid will be nil") {:ok, nil} end end) end defp has_txid_step?(multi) do multi |> Ecto.Multi.to_list() |> Enum.any?(fn {name, _} -> name == @txid_name end) end defp append_multi(multi, %Operation{} = op, %__MODULE__{} = writer) do with {:ok, actions} <- mutation_actions(op, writer), {:ok, action} <- Map.fetch(actions, op.operation) do ctx = %Context{ index: op.index, writer: writer, operation: op, schema: action.schema, action: action } multi |> apply_before_all(action) |> mutation_changeset(op, ctx, action) |> validate_pks(op, ctx, action) |> apply_before(op, ctx, action) |> apply_changeset(op, ctx, action) |> apply_after(op, ctx, action) else {:error, reason} -> Ecto.Multi.error(multi, {:error, op.index}, %Error{message: reason, operation: op}) end end defp apply_before_all(multi, %{before_all: nil} = _action) do multi end defp apply_before_all(multi, %{before_all: before_all_fun} = action) when is_function(before_all_fun, 1) do key = {action.schema, :before_all} Ecto.Multi.merge(multi, fn %{^key => true} = _changes -> Ecto.Multi.new() _changes -> before_all_fun.(Ecto.Multi.new()) |> Ecto.Multi.put({action.schema, :before_all}, true) end) end defp mutation_changeset(multi, %Operation{} = op, %Context{} = ctx, action) do %{schema: schema, validate: changeset_fun} = action %{index: idx, operation: operation, data: lookup_data, changes: change_data} = op Ecto.Multi.run(multi, {:__phoenix_sync__, :changeset, idx}, fn repo, changes -> ctx = %{ctx | changes: changes} case action.load.(repo, operation, lookup_data, ctx) do struct when is_struct(struct, schema) -> apply_changeset_fun(changeset_fun, struct, op, change_data, action) struct when is_struct(struct) -> {:error, %Error{ message: "load function returned an inconsistent value. Expected %#{schema}{}, got %#{struct.__struct__}{}", operation: op }} {:ok, struct} when is_struct(struct, schema) -> apply_changeset_fun(changeset_fun, struct, op, change_data, action) {:ok, struct} when is_struct(struct) -> {:error, %Error{ message: "load function returned an inconsistent value. Expected %#{schema}{}, got %#{struct.__struct__}{}", operation: op }} {:error, reason} -> {:error, %Error{message: reason, operation: op}} nil -> pks = Map.new(action.pks, fn col -> {col, Map.fetch!(lookup_data, to_string(col))} end) {:error, %Error{message: "No original record found for row #{inspect(pks)}", operation: op}} invalid -> {:error, %Error{ message: "Invalid return value from load(), got: #{inspect(invalid)}", operation: op }} end end) end defp apply_changeset_fun(changeset_fun, data, op, change_data, action) do case changeset_fun do fun3 when is_function(fun3, 3) -> {:ok, fun3.(data, change_data, op.operation)} fun2 when is_function(fun2, 2) -> {:ok, fun2.(data, change_data)} # delete changeset/validation functions can just look at the original fun1 when is_function(fun1, 1) -> {:ok, fun1.(data)} _ -> {:error, "Invalid changeset_fun for #{inspect(action.table)} #{inspect(op)}"} end end # inserts don't need the pk fields defp validate_pks(multi, %Operation{operation: :insert}, _ctx, _action) do multi end defp validate_pks(multi, %Operation{index: idx, data: lookup}, _ctx, action) do do_validate_pks(multi, action.pks, lookup, idx) end defp do_validate_pks(multi, pks, data, n) do case Enum.reject(pks, &Map.has_key?(data, to_string(&1))) do [] -> multi keys -> Ecto.Multi.error( multi, {:error, n}, {"Operation data is missing required primary keys: #{inspect(keys)}", data} ) end end defp apply_before(multi, operation, ctx, %{pre_apply: pre_apply_fun} = action) do apply_hook(multi, operation, ctx, {:pre_apply, pre_apply_fun}, action) end defp apply_after(multi, operation, ctx, %{post_apply: post_apply_fun} = action) do apply_hook(multi, operation, ctx, {:post_apply, post_apply_fun}, action) end defp apply_hook(multi, _operation, _ctx, {_, nil}, _action) do multi end defp apply_hook(multi, operation, ctx, {hook_name, hook_fun}, action) do Ecto.Multi.merge(multi, fn changes -> changeset = changeset!(changes, operation) ctx = %{ctx | changes: changes, callback: hook_name} case hook_fun do fun3 when is_function(fun3, 3) -> fun3.(Ecto.Multi.new(), changeset, ctx) _ -> raise "Invalid #{hook_name} for #{inspect(action.table)} #{inspect(operation.operation)}" end |> validate_callback!(operation.operation, action) end) end defp validate_callback!(%Ecto.Multi{} = multi, _op, _action), do: multi defp validate_callback!(value, op, action), do: raise(ArgumentError, message: "Invalid return type #{inspect(value)} for #{op} into #{inspect(action.table)}. Expected %Ecto.Multi{}" ) defp changeset!(changes, %Operation{index: n}) do changeset!(changes, n) end defp changeset!(changes, n) when is_integer(n) do Map.fetch!(changes, {:__phoenix_sync__, :changeset, n}) end defp apply_changeset(multi, %Operation{operation: :insert} = op, ctx, action) do Ecto.Multi.merge(multi, fn changes -> Ecto.Multi.insert(Ecto.Multi.new(), operation_name(ctx), changeset!(changes, op)) end) |> append_changeset_result(ctx, action) end defp apply_changeset(multi, %Operation{operation: :update} = op, ctx, action) do Ecto.Multi.merge(multi, fn changes -> Ecto.Multi.update(Ecto.Multi.new(), operation_name(ctx), changeset!(changes, op)) end) |> append_changeset_result(ctx, action) end defp apply_changeset(multi, %Operation{operation: :delete} = op, ctx, action) do Ecto.Multi.merge(multi, fn changes -> Ecto.Multi.delete(Ecto.Multi.new(), operation_name(ctx), changeset!(changes, op)) end) |> append_changeset_result(ctx, action) end # Put the result of each op into the multi under a deterministic key based on # the primary key. # This allows for quick lookup of data that we've already accessed defp append_changeset_result(multi, ctx, action) do name = operation_name(ctx) Ecto.Multi.merge(multi, fn %{^name => result} -> pk = Map.new(action.pks, &{to_string(&1), Map.fetch!(result, &1) |> to_string()}) Ecto.Multi.put(Ecto.Multi.new(), load_key(ctx, pk), result) end) end defp mutation_actions(%Operation{relation: [prefix, name] = relation} = operation, write) when is_binary(name) and is_binary(prefix) do case write.mappings do %{^relation => actions} -> {:ok, actions} %{^name => actions} -> {:ok, actions} _ -> {:error, %Error{ message: "No configuration for writes to table #{inspect(name)}", operation: operation }} end end defp mutation_actions(%Operation{relation: name} = operation, write) when is_binary(name) do case write.mappings do %{^name => actions} -> {:ok, actions} mappings -> case Enum.filter(Map.keys(mappings), &match?([_, ^name], &1)) do [] -> {:error, %Error{ message: "No configuration for writes to table #{inspect(name)}", operation: operation }} [key] -> {:ok, Map.fetch!(write.mappings, key)} [_ | _] = keys -> {:error, %Error{ message: "Multiple matches for relation #{inspect(name)}: #{inspect(keys)}. Please pass full `[\"schema\", \"name\"]` relation in mutation data", operation: operation }} end end end @doc """ Runs the mutation inside a transaction. Since the mutation operation is expressed as an `Ecto.Multi` operation, see the [`Ecto.Repo` docs](https://hexdocs.pm/ecto/Ecto.Repo.html#c:transaction/2-use-with-ecto-multi) for the result if any of your mutations returns an error. #{inspect(__MODULE__)}.new() |> #{inspect(__MODULE__)}.allow(MyApp.Todos.Todo) |> #{inspect(__MODULE__)}.allow(MyApp.Options.Option) |> #{inspect(__MODULE__)}.ingest( changes, format: #{inspect(__MODULE__)}.Format.TanstackDB ) |> #{inspect(__MODULE__)}.transaction(MyApp.Repo) |> case do {:ok, txid, _changes} -> # return the txid to the client Plug.Conn.send_resp(conn, 200, Jason.encode!(%{txid: txid})) {:error, _failed_operation, failed_value, _changes_so_far} -> # extract the error message from the changeset returned as `failed_value` error = Ecto.Changeset.traverse_errors(failed_value, fn {msg, opts} -> Regex.replace(~r"%{(\w+)}", msg, fn _, key -> opts |> Keyword.get(String.to_existing_atom(key), key) |> to_string() end) end) Plug.Conn.send_resp(conn, 400, Jason.encode!(error)) end Also supports normal fun/0 or fun/1 style transactions much like `c:Ecto.Repo.transaction/2`, returning the txid of the operation: {:ok, txid, todo} = #{inspect(__MODULE__)}.transaction(fn -> Repo.insert!(changeset) end, Repo) """ @spec transaction(t() | Ecto.Multi.t(), Ecto.Repo.t(), keyword()) :: {:ok, txid(), Ecto.Multi.changes()} | Ecto.Multi.failure() def transaction(writer_or_multi, repo, opts \\ []) def transaction(%__MODULE__{ingest: []}, _repo, _opts) do {:error, "no changes ingested"} end def transaction(%__MODULE__{} = writer, repo, opts) do writer |> to_multi() |> transaction(repo, opts) end def transaction(%Ecto.Multi{} = multi, repo, opts) when is_atom(repo) do wrapped_multi = if has_txid_step?(multi) do multi else Ecto.Multi.prepend(multi, txid_step()) end with {:ok, changes} <- repo.transaction(wrapped_multi, opts) do {txid, changes} = Map.pop!(changes, @txid_name) {:ok, txid, changes} end end @doc """ Apply operations from a mutation directly via a transaction. `operation_fun` is a 1-arity function that receives each of the `%#{inspect(__MODULE__.Operation)}{}` structs within the mutation data and should apply them appropriately. It should return `:ok` or `{:ok, result}` if successful or `{:error, reason}` if the operation is invalid or failed to apply. If any operation returns `{:error, _}` or raises then the entire transaction is aborted. This function will return `{:error, reason}` if the transaction data fails to parse. {:ok, txid} = #{inspect(__MODULE__)}.transact( my_encoded_txn, MyApp.Repo, fn %{operation: :insert, relation: [_, "todos"], change: change} -> MyApp.Repo.insert(...) %{operation: :update, relation: [_, "todos"], data: data, change: change} -> MyApp.Repo.update(Ecto.Changeset.cast(...)) %{operation: :delete, relation: [_, "todos"], data: data} -> # we don't allow deletes... {:error, "invalid delete"} end, format: #{inspect(__MODULE__.Format.TanstackDB)}, timeout: 60_000 ) Any of the `opts` not used by this module are passed onto the `c:Ecto.Repo.transaction/2` call. This is equivalent to the below: {:ok, txn} = #{inspect(__MODULE__)}.parse_transaction( my_encoded_txn, format: #{inspect(__MODULE__.Format.TanstackDB)} ) {:ok, txid} = MyApp.Repo.transaction(fn -> Enum.each(txn.operations, fn %{operation: :insert, relation: [_, "todos"], change: change} -> # insert a Todo %{operation: :update, relation: [_, "todos"], data: data, change: change} -> # update a Todo %{operation: :delete, relation: [_, "todos"], data: data} -> # we don't allow deletes... raise "invalid delete" end) #{inspect(__MODULE__)}.txid!(MyApp.Repo) end, timeout: 60_000) """ @spec transact( Format.transaction_data(), Ecto.Repo.t(), operation_fun :: (Operation.t() -> :ok | {:ok, any()} | {:error, any()}), transact_opts() ) :: {:ok, txid()} | {:error, any()} def transact(changes, repo, operation_fun, opts) when is_function(operation_fun, 1) and is_atom(repo) do {parse_opts, txn_opts} = split_writer_txn_opts(opts) with {:ok, %Transaction{} = txn} <- parse_transaction(changes, parse_opts) do repo.transaction( fn -> Enum.reduce_while(txn.operations, :ok, fn op, :ok -> case operation_fun.(op) do {:ok, _result} -> {:cont, :ok} :ok -> {:cont, :ok} {:error, _reason} = error -> {:halt, error} other -> raise ArgumentError, "expected to return :ok, {:ok, _} or {:error, _}, got: #{inspect(other)}" end end) |> case do {:error, reason} -> repo.rollback(reason) :ok -> txid!(repo) end end, txn_opts ) end end @doc """ Extract the transaction id from changes or from a `Ecto.Repo` within a transaction. This allows you to use a standard `c:Ecto.Repo.transaction/2` call to apply mutations defined using `apply/2` and extract the transaction id afterwards. Example {:ok, changes} = #{inspect(__MODULE__)}.new() |> #{inspect(__MODULE__)}.allow(MyApp.Todos.Todo) |> #{inspect(__MODULE__)}.allow(MyApp.Options.Option) |> #{inspect(__MODULE__)}.to_multi(changes, format: #{inspect(__MODULE__)}.Format.TanstackDB) |> MyApp.Repo.transaction() {:ok, txid} = #{inspect(__MODULE__)}.txid(changes) It also allows you to get a transaction id from any active transaction: MyApp.Repo.transaction(fn -> {:ok, txid} = #{inspect(__MODULE__)}.txid(MyApp.Repo) end) Attempting to run `txid/1` on a repo outside a transaction will return an error. """ @spec txid(Ecto.Multi.changes()) :: {:ok, txid()} | :error def txid(%{@txid_name => txid} = _changes), do: {:ok, txid} def txid(changes) when is_map(changes), do: :error def txid(repo) when is_atom(repo) do if repo.in_transaction?() do with {:ok, %{rows: [[txid]]}} = repo.query(@txid_query) do {:ok, txid} end else {:error, %Error{message: "not in a transaction"}} end end @doc """ Returns the a transaction id or raises on an error. See `txid/1`. """ @spec txid!(Ecto.Multi.changes()) :: txid() def txid!(%{@txid_name => txid} = _changes), do: txid def txid!(%{}), do: raise(ArgumentError, message: "No txid in change data") def txid!(repo) when is_atom(repo) do case txid(repo) do {:ok, txid} -> txid {:error, reason} -> raise reason end end @doc """ Return a unique operation name for use in `pre_apply` or `post_apply` callbacks. `Ecto.Multi` requires that all operation names be unique within a transaction. This function gives you a simple way to generate a name for your own operations that is guarateed not to conflict with any other. Example: #{inspect(__MODULE__)}.new() |> #{inspect(__MODULE__)}.allow( MyModel, pre_apply: fn multi, changeset, context -> name = #{inspect(__MODULE__)}.operation_name(context) Ecto.Multi.insert(multi, name, AuditEvent.for_changeset(changeset)) end ) """ def operation_name(%Context{} = ctx) do {ctx.schema, ctx.operation.operation, ctx.index} end @doc """ Like `operation_name/1` but allows for a custom label. """ @spec operation_name(context(), term()) :: term() def operation_name(%Context{} = ctx, label) do {operation_name(ctx), label} end defp load_key(ctx, pk) do opkey(schema: ctx.schema, operation: ctx.operation.operation, index: ctx.index, pk: pk) end end end