The job gets JID and goes to dead right away

[migalenkom]

Describe the bug
We have one instance with sidekiq and after some time it starts to work incorrectly. We go to console and call HeartbeatWorker.perform_async(false, nil) , it shows JID and goes to dead right away.

HeartbeatWorker.perform_async(false, nil)
=> "72d9a59868d59c268c063780"

In logs, I can see

2020-06-25 08:26:29.174996 W [19701:processor] Sidekiq -- the unique_key: uniquejobs:17efb573488d8206c5190fbaaa8931f8 is not locked, allowing job to silently complete

Expected behavior
The worker processed after it get JID and it goes to dead only when we have a job in progress with the same args.

Current behavior
The worker is not processed it goes to dead

Worker class

class UniqueWorker
  include Sidekiq::Worker
  sidekiq_options lock: :until_executed, on_conflict: :reject, unique_args: -> (*args) { args }
end

class HeartbeatWorker < UniqueWorker
  sidekiq_options queue: :default, retry: false, backtrace: true
  
  def perform(a, b)
    MyClass.do_some(a, b)
  end
end

Additional context
Currently, gem version is 6.0.22, also we tried to 7.0.0.beta22 and received the same issue.

[0xCCD]

Same problem for me !

[dkirilenko]

I have the same issue! Any ideas?

[mhenrixon]

Seems like a problem with generating the lock.

In v7, you can configure lock_info which for instance would tell you which arguments are configured for uniqueness. Maybe that will help with debugging the issue?

I don't see anything apparently wrong but I suspect you have a problem with the unique arguments and what actually ends up with the worker somehow.

[adrianotadao]

I'm having the same problem. Any solution?

[mhenrixon]

@adrianotadao @0xCCD @migalenkom do any of you create one background job from another?

In the original issue description it looks like that is what is going on. To be completely honest, I haven't done a whole lot of experimenting with this. I just assumed it will work as normal.

The issue description doesn't make clear if this is the case; hence the question.

Ah, now I see that we are doing worker inheritance.

I'll experiment with this. The question remains, do you use either inheritance or scheduling one job from another?

[mhenrixon]

It would be brilliant if someone that is having this problem could follow the instructions in #571 and create a failing example for me using v7.0.1. I'll close this for now but I'm here to help (preferably with v7).