#log_job_exception gets wrong exception info when job can be retried

[yiichou]

Sidekiq version: 6.4.1
Rails version: 6.0.3.6

If a failed job is able to be retried, Sidekiq will

1. catch the really exception
2. append the exception message to the job and push it into retry queue
3. raise a new exception Sidekiq::JobRetry::Skip

So, when Sidekiq::LogstashJobLogger tries to parse the failed job by #log_job_exception,
the received exception is Sidekiq::JobRetry::Skip.

Finally, the recorded error_message, error and error_backtrace are wrong.

[yiichou]

I have a patch to solve this problem, it tries to get really exception message from the enqueued retry job

def log_job_exception(job, started_at, exc)
  payload = setup_payload(job)

  payload['duration'] = elapsed(started_at)
  payload['message'] += ": fail: #{payload['duration']} sec"
  payload['job_status'] = 'fail'

  if exc.is_a? Sidekiq::JobRetry::Skip
    Sidekiq::RetrySet.new.scan(payload['jid'], 1) do |retried_job|
      payload['error_message'] = retried_job['error_message']
      payload['error_class'] = retried_job['error_class']
    end
  else
    payload['error_message'] = exc.message
    payload['error_class'] = exc.class
    payload['error_backtrace'] = %('#{exc.backtrace.join("\n")}')
  end

  process_payload(payload)
end

[iMacTia]

Hi @yiichou, thanks for raising this.
I think I saw Sidekiq::JobRetry::Skip exceptions before, but they never posed to be a problem, this might be a combination of sidekiq/ruby/rails versions.

Do you think you might be able to come up with a PR that reproduces the issue so we can run it against all the CI matrix versions?