fix(electric): Handle connector errors during startup (#1290)

Previously, we would ignore any non-raising connector startup failures,
failing to even log them.

With this change we make sure all connectors are running and log a fatal
error when that's not the case.

Side note: there's an ongoing work in
#1226 on adding an e2e test
to verify the correct output of such errors.

.changeset/grumpy-goats-work.md

@@ -0,0 +1,5 @@
+---
+"@core/electric": patch
+---
+
+The sync service now exits when a database connector fails to initialise. Previously, some initialisation errors would result in Electric running without a single database connection and not logging any erors about that.

components/electric/lib/electric/application.ex

@@ -53,20 +53,26 @@ defmodule Electric.Application do
 
     opts = [strategy: :one_for_one, name: Electric.Supervisor]
 
-    with {:ok, supervisor} <- Supervisor.start_link(children, opts) do
-      Application.get_env(:electric, Electric.Replication.Connectors, [])
-      |> Enum.each(fn {name, config} ->
-        Connectors.start_connector(
-          PostgresConnector,
-          config
-          |> Keyword.put(:origin, to_string(name))
-          |> Keyword.put(:write_to_pg_mode, write_to_pg_mode)
-        )
-      end)
-
+    with {:ok, supervisor} <- Supervisor.start_link(children, opts),
+         connectors = Application.get_env(:electric, Electric.Replication.Connectors, []),
+         :ok <- start_connectors(connectors, write_to_pg_mode) do
       {:ok, supervisor}
+    else
+      error -> log_supervisor_startup_error(error)
+    end
+  end
+
+  defp start_connectors([], _write_to_pg_mode), do: :ok
+
+  defp start_connectors([{name, config} | connectors], write_to_pg_mode) do
+    connector_config =
+      config
+      |> Keyword.put(:origin, to_string(name))
+      |> Keyword.put(:write_to_pg_mode, write_to_pg_mode)
+
+    with {:ok, _pid} <- Connectors.start_connector(PostgresConnector, connector_config) do
+      start_connectors(connectors, write_to_pg_mode)
     end
-    |> log_supervisor_startup_error()
   end
 
   defp maybe_add_child(children, nil), do: children
@@ -94,15 +100,15 @@ defmodule Electric.Application do
     end
   end
 
-  defp log_supervisor_startup_error({:ok, _sup_pid} = ok), do: ok
-
   defp log_supervisor_startup_error(
          {:error, {:shutdown, {:failed_to_start_child, child_id, reason}}} = error
        ) do
     _ = log_child_error(child_id, reason)
     error
   end
 
+  defp log_supervisor_startup_error(error), do: error
+
   @spec log_child_error({atom, atom}, term) :: no_return
   defp log_child_error(
          {ThousandIsland, :replication_tcp_server_listener},
@@ -126,6 +132,20 @@ defmodule Electric.Application do
     )
   end
 
+  defp log_child_error(
+         Electric.Replication.PostgresConnectorMng,
+         {{:badmatch, {:error, :nxdomain}}, _stracktrace}
+       ) do
+    Electric.Errors.print_fatal_error(
+      :init,
+      "Failed to resolve the database domain name to an IP address.",
+      """
+      Double-check the value of DATABASE_URL and make sure to set
+      DATABASE_USE_IPV6=true if your database is only reachable using IPv6.
+      """
+    )
+  end
+
   defp log_child_error(child_id, reason) do
     Electric.Errors.failed_to_start_child_error(:init, child_id, reason)
     |> Electric.Errors.print_fatal_error()

components/electric/lib/electric/replication/postgres_connector.ex

@@ -18,7 +18,7 @@ defmodule Electric.Replication.PostgresConnector do
     name = name(origin)
     Electric.reg(name)
 
-    children = [%{id: :mng, start: {PostgresConnectorMng, :start_link, [connector_config]}}]
+    children = [{PostgresConnectorMng, connector_config}]
     Supervisor.init(children, strategy: :one_for_all)
   end