fix: change CdcRls.WorkerSupervisor to be temporary instead of transient (#1587)

RealtimeChannel already reacts to WorkerSupervisor going down to
reconnect

Having it transient can bring down other unrelated WorkerSupervisor in
some cases

Author: edgurgel

lib/extensions/postgres_cdc_rls/cdc_rls.ex

@@ -99,7 +99,7 @@ defmodule Extensions.PostgresCdcRls do
       %{
         id: tenant,
         start: {Rls.WorkerSupervisor, :start_link, [args]},
-        restart: :transient
+        restart: :temporary
       }
     )
   end

mix.exs

@@ -4,7 +4,7 @@ defmodule Realtime.MixProject do
   def project do
     [
       app: :realtime,
-      version: "2.56.1",
+      version: "2.56.2",
       elixir: "~> 1.18",
       elixirc_paths: elixirc_paths(Mix.env()),
       start_permanent: Mix.env() == :prod,

priv/repo/dev_seeds.exs

@@ -41,30 +41,29 @@ default_db_host = "127.0.0.1"
       })
       |> Repo.insert!()
 
-    publication = "supabase_realtime"
-
-    [
-      "drop publication if exists #{publication}",
-      "drop table if exists public.test_tenant;",
-      "create table public.test_tenant ( id SERIAL PRIMARY KEY, details text );",
-      "grant all on table public.test_tenant to anon;",
-      "grant all on table public.test_tenant to postgres;",
-      "grant all on table public.test_tenant to authenticated;",
-      "create publication #{publication} for table public.test_tenant"
-    ]
-    |> Enum.each(&query!(Repo, &1, []))
-
     tenant
   end)
 
 # Reset Tenant DB
 settings = Database.from_tenant(tenant, "realtime_migrations", :stop)
 settings = %{settings | max_restarts: 0, ssl: false}
 {:ok, tenant_conn} = Database.connect_db(settings)
+publication = "supabase_realtime"
 
 Postgrex.transaction(tenant_conn, fn db_conn ->
   Postgrex.query!(db_conn, "DROP SCHEMA IF EXISTS realtime CASCADE", [])
   Postgrex.query!(db_conn, "CREATE SCHEMA IF NOT EXISTS realtime", [])
+
+  [
+    "drop publication if exists #{publication}",
+    "drop table if exists public.test_tenant;",
+    "create table public.test_tenant ( id SERIAL PRIMARY KEY, details text );",
+    "grant all on table public.test_tenant to anon;",
+    "grant all on table public.test_tenant to postgres;",
+    "grant all on table public.test_tenant to authenticated;",
+    "create publication #{publication} for table public.test_tenant"
+  ]
+  |> Enum.each(&Postgrex.query!(db_conn, &1))
 end)
 
 case Tenants.Migrations.run_migrations(tenant) do

test/realtime/extensions/cdc_rls/cdc_rls_test.exs

@@ -76,21 +76,23 @@ defmodule Realtime.Extensions.CdcRlsTest do
       metadata = [metadata: subscription_metadata]
       :ok = PostgresCdc.subscribe(PostgresCdcRls, pg_change_params, external_id, metadata)
 
+      RealtimeWeb.Endpoint.subscribe(PostgresCdcRls.syn_topic(tenant.external_id))
       # First time it will return nil
       PostgresCdcRls.handle_connect(args)
       # Wait for it to start
-      Process.sleep(3000)
+      assert_receive %{event: "ready"}, 1000
+
+      on_exit(fn -> PostgresCdcRls.handle_stop(external_id, 10_000) end)
       {:ok, response} = PostgresCdcRls.handle_connect(args)
 
       # Now subscribe to the Postgres Changes
       {:ok, _} = PostgresCdcRls.handle_after_connect(response, postgres_extension, pg_change_params)
 
-      on_exit(fn -> PostgresCdcRls.handle_stop(external_id, 10_000) end)
+      RealtimeWeb.Endpoint.unsubscribe(PostgresCdcRls.syn_topic(tenant.external_id))
       %{tenant: tenant}
     end
 
-    @tag skip: "Flaky test. When logger handle_sasl_reports is enabled this test doesn't break"
-    test "Check supervisor crash and respawn", %{tenant: tenant} do
+    test "supervisor crash must not respawn", %{tenant: tenant} do
       sup =
         Enum.reduce_while(1..30, nil, fn _, acc ->
           :syn.lookup(Extensions.PostgresCdcRls, tenant.external_id)
@@ -112,12 +114,11 @@ defmodule Realtime.Extensions.CdcRlsTest do
       Process.exit(sup, :kill)
       assert_receive {:DOWN, _, :process, ^sup, _reason}, 5000
 
-      assert_receive %{event: "ready"}, 5000
+      assert_receive %{event: "postgres_cdc_rls_down"}
 
-      {sup2, _} = :syn.lookup(Extensions.PostgresCdcRls, tenant.external_id)
+      refute_receive %{event: "ready"}, 1000
 
-      assert(sup != sup2)
-      assert Process.alive?(sup2)
+      :undefined = :syn.lookup(Extensions.PostgresCdcRls, tenant.external_id)
     end
 
     test "Subscription manager updates oids", %{tenant: tenant} do