Scheduled zero arg jobs fail

[nsweeting]

It appears as though scheduled zero-arg jobs end up getting their [] arg list cast into a {} with the LPUSH command.

Verk.Supervisor.start_link

defmodule Job do
  def perform do
    IO.inspect "hello"
  end
end

job = %Verk.Job{class: Job, args: [], queue: "default"}

Currently this results in:

15:02:39.544 [info]  Elixir.Job 13837351251859802746 start
15:02:39.548 [debug] Job failed reason: %ArgumentError{message: "argument error"}
15:02:39.548 [info]  Elixir.Job 13837351251859802746 fail: 5 ms
15:02:39.552 [error] GenServer #PID<0.196.0> terminating
** (stop) :failed
Last message: {:"$gen_cast", {:perform, %Verk.Job{args: %{}, class: "Elixir.Job", enqueued_at: nil, error_backtrace: nil, error_message: nil, failed_at: nil, finished_at: nil, jid: "13837351251859802746", max_retry_count: 25, original_json: "{\"error_message\":null,\"max_retry_count\":25,\"jid\":\"13837351251859802746\",\"class\":\"Elixir.Job\",\"retried_at\":null,\"queue\":\"default\",\"finished_at\":null,\"enqueued_at\":null,\"error_backtrace\":null,\"args\":{},\"failed_at\":null,\"retry_count\":0}", queue: "default", retried_at: nil, retry_count: 0}, #PID<0.197.0>}}
State: nil

And the redis console output:

1518724956.160306 [0 127.0.0.1:54702] "ZADD" "schedule" "1518724956" "{\"retry_count\":0,\"retried_at\":null,\"queue\":\"default\",\"max_retry_count\":25,\"jid\":\"13837351251859802746\",\"finished_at\":null,\"failed_at\":null,\"error_message\":null,\"error_backtrace\":null,\"enqueued_at\":null,\"class\":\"Elixir.Job\",\"args\":[]}"
1518724957.987005 [0 lua] "LPUSH" "queue:default" "{\"error_message\":null,\"max_retry_count\":25,\"jid\":\"13837351251859802746\",\"class\":\"Elixir.Job\",\"retried_at\":null,\"queue\":\"default\",\"finished_at\":null,\"enqueued_at\":null,\"error_backtrace\":null,\"args\":{},\"failed_at\":null,\"retry_count\":0}"

[edgurgel]

Thanks for the bug report! Yeah this looks wrong 🤔

[krasio]

Looks like this comes from the cjson package used in the lua scripts:

127.0.0.1:6379> EVAL 'return cjson.encode(cjson.decode(\'{"args": []}\'))' 0
"{\"args\":{}}"

See also https://github.com/mpx/lua-cjson/issues/11#issuecomment-287617268.

Not sure what can be done. 😞

@nsweeting The only workaround I can think of at the moment is to have at least one argument for such jobs even if it's not used.

defmodule Job do
  def perform(1) do
    IO.inspect "hello"
  end
end

job = %Verk.Job{class: Job, args: [1], queue: "default"}

Not Ideal but better than getting that error.

[edgurgel]

Maybe we could try to "patch" the JSON when decoding? We know that args is an array. I wonder what's the best place to do this...

[krasio]

What about something like this - https://github.com/edgurgel/verk/pull/154?