Shards are being restarted while calling `Phoenix.Tracker.graceful_permdown/1`

[tonnenpinguin]

Hey all!

I noticed that when calling Phoenix.Tracker.graceful_permdown/1 shards are being restarted on the Tracker that we're currently in the process of shutting down.
Looking through the source code I think the problem ist that when we are passing the shards to the Supervisor, the child_spec field :restart defaults to :permanent which according to the Supervisor docs
means

• :permanent - the child process is always restarted.

However, I think we actually want this:

• :transient - the child process is restarted only if it terminates abnormally, i.e., with an exit reason other than :normal, :shutdown, or {:shutdown, term}.

as in the Shard graceful_permdown callback we are trying to gracefully stop the Shard GenServer

  def handle_call(:graceful_permdown, _from, state) do
    broadcast_from(state, self(), {:pub, :graceful_permdown, Replica.ref(state.replica)})
    {:stop, :normal, :ok, state}
  end

Returning {:stop, reason, reply, new_state} stops the loop
https://hexdocs.pm/elixir/1.18.2/GenServer.html#c:handle_call/3

Explicitly specifying restart: :transient in https://github.com/phoenixframework/phoenix_pubsub/blob/main/lib/phoenix/tracker.ex#L328 seems to fix the issue.

Does that make sense? If yes, take a look at #195

[josevalim]

Thank you for the issue and the PR. I believe the issue is that we are trying to shutdown the shards individually while we already shutdown the supervisor. So theoretically, a possible fix is to not shutdown the shards at all, and let the supervisor do it.

HOWEVER, doing so leaves the possibility for a permdown shard to consume other events, so I guess we want to keep the early shutdown anyway. In other words, we should merge your PR. Thank you!