Yeah, it seems like there is first bug Thanks

The point is that :warn used by Elixir Logger is now :warning form the Erlang, but there should be translation from old format to the new format. This was done, but in the meantime we have changed code a little to provide given message, and the translation perished. Will provide fix in a moment.

Thanks for these contributions! I installed elixir master using asdf:

asdf install elixir ref:67dec4683de49e6e8153f9c49d0a0a179c550f9f

and gave it a try in a phoenix app where I have:

config :logger, level: :warn

And I get:

Erlang/OTP 22 [erts-10.4.3] [source] [64-bit] [smp:8:8] [ds:8:8:10] [async-threads:1] [hipe]

warning: the level for Erlang's logger was set to :debug, but Elixir's logger was set to :warn. Elixir's logger value will take higher precedence
  (logger 1.10.0-dev) lib/logger/app.ex:96: Logger.App.add_elixir_handler/2
  (logger 1.10.0-dev) lib/logger/app.ex:24: Logger.App.start/2
  (kernel 6.4.1) application_master.erl:277: :application_master.start_it_old/4

** (MatchError) no match of right hand side value: {:error, {:bad_return, {{Logger.App, :start, [:normal, []]}, {:EXIT, {{:badmatch, {:error, {:invalid_level, :warn}}}, [{Logger.App, :add_elixir_handler, 2, [file: 'lib/logger/app.ex', line: 103]}, {Logger.App, :start, 2, [file: 'lib/logger/app.ex', line: 24]}, {:application_master, :start_it_old, 4, [file: 'application_master.erl', line: 277]}]}}}}}
    (mix 1.10.0-dev) lib/mix/cli.ex:106: Mix.CLI.restart_logger/0
    (mix 1.10.0-dev) lib/mix/cli.ex:82: Mix.CLI.run_task/2
    (elixir 1.10.0-dev) lib/code.ex:912: Code.require_file/2

So the first thing to note is that the log level :warn is deprecated, but is there also a new way to configure the log level?

Trying level: :info or level: :warning produce a similar warning, but no match error, and logging seems to work as usual.

Very excited about this. Thanks for all your work!

I’m curious what approach you have in mind for structured logging. So far I’ve been doing it this way:

• At the call site, pass structured data via metadata:

    Logger.log(level, "Processing request",
      method: conn.method,
      path: conn.request_path,
      ...
    )
  

• Define a custom formatter which takes all metadata:

  config :logger, :file_log,
    path: "app.log",
    level: :info,
    format: {MyApp.Logger, :format},
    metadata: :all
  

• In the formatter, generate JSON, roughly like:

  def format(level, message, timestamp, metadata) do
   %{
      "timestamp" => format_timestamp(timestamp),
      "level" => level,
      "message" => IO.iodata_to_binary(message)
    }
    |> add_metadata(metadata)
    |> Jason.encode!()
  end
  

  where add_metadata reads each key/value pair from metadata, converts the value to a printable representation, and adds it to the event map, similar to Honeybadger.JSON.

It works, but it’s not ideal - we had to reimplement functionality from the default formatter, for instance formatting time or converting PIDs to strings. OTP reports also can’t be structured, because they’re a string blob by the time they reach the formatter (not a huge deal though).

Could we achieve JSON logging more easily with the new backend, or with future development?