Ecto

Shardex.Adapter.Ecto requires :ecto (usually via :ecto_sql).

Dynamic mode: one repo module, many databases

shards: [
  shard_1: [
    primary: {Shardex.Adapter.Ecto, repo: MyApp.Repo, config: [database: "s1"]},
    replica: {Shardex.Adapter.Ecto, repo: MyApp.Repo, config: [database: "s1", hostname: "replica-1"]}
  ]
]

Shardex starts MyApp.Repo once per pool, named MyApp.Shards.shard_1.primary (etc.), with config merged over the repo's application config. run/3 sets the dynamic repo, calls your function with MyApp.Repo, and restores the previous dynamic repo afterwards — also when the function raises:

MyApp.Shards.run(org_id, fn repo -> repo.get(Contact, id) end)

Dynamic repos are per process. Code that spawns processes inside run must set the dynamic repo again there. run_batch(..., max_concurrency: n) already does this for each task.

In a Plug

If you prefer to set the dynamic repo for the whole request:

def call(conn, _opts) do
  case MyApp.Shards.ref(conn.assigns.org_id) do
    {:ok, dynamic_name} ->
      MyApp.Repo.put_dynamic_repo(dynamic_name)
      conn

    {:error, :maintenance} ->
      conn |> send_resp(503, "maintenance") |> halt()

    {:error, _reason} ->
      conn |> send_resp(500, "no shard for this organization") |> halt()
  end
end

The %Shardex.Pool{meta: %{repo: MyApp.Repo, dynamic_name: name}} of a shard (lookup/1 returns the shard; its pools are in shard.roles) carries both values if you need them.

Module mode: one repo module per shard

shards: [
  shard_1: {Shardex.Adapter.Ecto, repo: MyApp.RepoShard1},
  shard_2: {Shardex.Adapter.Ecto, repo: MyApp.RepoShard2}
]

The ref is the module, and run/3 simply calls fun.(MyApp.RepoShard1).

Shardex starts each repo module under its own supervisor. If your application already starts the repo (e.g. it is in your supervision tree), pass start: false. Otherwise Shardex finds the repo's name already registered, treats that process as its pool and routes to it, but does not own it: maintenance(shard, :stop) will not stop it.

Repos started elsewhere

Add start: false to route to a repo you already start in your own tree.

Migrations

Shardex does not run migrations. With dynamic mode, iterate over MyApp.Shards.shards() and run Ecto.Migrator against each pool's dynamic_name, or keep per-shard repo modules for migrations.