V7 Beta 15 on_conflict: with Hash does not work on server

[jasonbekolay]

Describe the bug
Version 7 Beta 15 does not work on the server side only when given a Hash for on_conflict:. Server crashes when a conflict is detected.

Expected behavior
Conflict strategy is executed on the server when a conflict is detected.

Current behavior
Job crashes with stacktrace:

2020-05-06T17:25:16.817Z 67576 TID-owvovowmo WARN: NoMethodError: undefined method `to_sym' for {"client"=>"log", "server"=>"reschedule"}:Hash
Did you mean?  to_s
               to_set
2020-05-06T17:25:16.817Z 67576 TID-owvovowmo WARN: /Users/jasonbekolay/.rvm/gems/ruby-2.6.5@yapp/gems/sidekiq-unique-jobs-7.0.0.beta15/lib/sidekiq_unique_jobs/on_conflict.rb:34:in `find_strategy'
/Users/jasonbekolay/.rvm/gems/ruby-2.6.5@yapp/gems/sidekiq-unique-jobs-7.0.0.beta15/lib/sidekiq_unique_jobs/lock/base_lock.rb:148:in `server_strategy'
/Users/jasonbekolay/.rvm/gems/ruby-2.6.5@yapp/gems/sidekiq-unique-jobs-7.0.0.beta15/lib/sidekiq_unique_jobs/lock/while_executing.rb:40:in `block in execute'
/Users/jasonbekolay/.rvm/gems/ruby-2.6.5@yapp/gems/sidekiq-unique-jobs-7.0.0.beta15/lib/sidekiq_unique_jobs/logging.rb:103:in `block in with_logging_context'
/Users/jasonbekolay/.rvm/gems/ruby-2.6.5@yapp/gems/sidekiq-unique-jobs-7.0.0.beta15/lib/sidekiq_unique_jobs/logging.rb:118:in `block in with_configured_loggers_context'
/Users/jasonbekolay/.rvm/gems/ruby-2.6.5@yapp/gems/sidekiq-5.2.8/lib/sidekiq/logging.rb:48:in `with_context'

Worker class

class TestJob
  include Sidekiq::Worker

  sidekiq_options lock: :until_and_while_executing, on_conflict: { client: :log, server: :reschedule }

  def perform
    print "."
    sleep 1
    puts "."
  end
end

Additional context
I'm using Sidekiq 5.2.8. I believe the issue is that job hash is completed stringified on Sidekiq pulls it from Redis, and on_server_conflict is expecting symbols for keys in the hash.

[mhenrixon]

I recommend you upgrade to v7.0.16. It was fixed there

[jasonbekolay]

Thank you for the quick response!

I don't see a release for 7.0.16. I tried against master and I'm still able to reproduce my issue. My reproduction steps are:

Worker class:

class TestJob
  include Sidekiq::Worker

  sidekiq_options lock: :until_and_while_executing, on_conflict: { client: :log, server: :reschedule }

  def perform(i)
    sleep 1
    puts "\n ----- #{i} -----\n"
  end
end

Start up multiple sidekiq servers. I have been running 4.

For my client, I boot up a rails console and run the following:

rand = Random.new
while true do
  i = rand.rand(100)
  puts "Queueing #{i}"
  3.times { Sidekiq::Client.enqueue(TestJob, i) }
  sleep 0.1
  3.times { Sidekiq::Client.enqueue(TestJob, i) }
  while Sidekiq::Queue.new.size > 0 do
    print "."
    sleep 1
  end
  puts "Done #{i}"
end

The client is solid, it works as expected when there's a conflict. If there's a conflict on the server side, it fails instead of rescheduling the job.

I think the issue is that on the server side, the job_hash gets serialized to JSON then deserialized, so the keys for the on_conflict hash are no longer symbols. This is what I'm trying to get it in #500

Thanks again for your help!

[mhenrixon]

Ok, I see you point, the solution is much simpler than I thought. Will try to create a new release today.

[jasonbekolay]

Thanks! I've been running in production with #500 for the past few days and it has been working well. I'll switch to the latest release whenever you're able to create it.

[mhenrixon]

Released as v7.0.0.beta16

[jasonbekolay]

Thank you for the release. Unfortunately the release broke client locking for me, as the on_conflict hash is not deeply stringified on the client side. Solution in #506 is working for me.