mongo process is managed by Supervisor

def start(_type, _args) do
    # List all child processes to be supervised
    database_config = Application.get_env(:content_proxy, :db_config)

    children = [
      ContentProxyWeb.Endpoint,
      %{
        id: Mongo,
        start: { Mongo, :start_link, [database_config] }
      }
    ]

    {:ok, _} = Logger.add_backend(Sentry.LoggerBackend)

    opts = [strategy: :one_for_one, name: ContentProxy.Supervisor]
    Supervisor.start_link(children, opts)
  end

According to Supervisor doc Supervisor — Elixir v1.20.2, when Supervisor shutdown a child process, he waits until 5000 milliseconds before to kill it. This timeout is only for worker or infinity child spec.

an atom that defines how a child process should be terminated (see the "Shutdown values" section below). This key is optional and defaults to 5000 if the type is :worker or :infinity if the type is :supervisor.

Mongodb defines itself as a worker process (mongodb/lib/mongo.ex at master · elixir-mongo/mongodb · GitHub)

  @spec start_link(Keyword.t()) :: {:ok, pid} | {:error, Mongo.Error.t() | atom}
  def start_link(opts) do
    opts
    |> UrlParser.parse_url()
    |> Topology.start_link()
  end

  def child_spec(opts, child_opts \\ []) do
    Supervisor.Spec.worker(Mongo, [opts], child_opts)
  end

According to MongoDB source code (mongodb/lib/mongo.ex at master · elixir-mongo/mongodb · GitHub), he waits until 5000 for the initial connection, so I can’t say if the problem is when Supervisor tries to terminate mongodb process or when MongoDB process tries to connect the MongoDB host.

This has been something I have been trying to track down for a while now. Unfortunately I have never been able to replicate it. There is an issue open with a fair amount of traffic.

What version of the mongodb package are you currently using? You say that it sometimes happens and sometimes not. Can you say roughly how frequently it happens? Something like 20-30%. Would it be possible for me to gain access to your database to see if I am able to replicate the issue? If I can replicate it, there is a possibility I can track down what is happened and fix it.