Add concurrency extension for ActiveJob

[bensheldon]

CreatesGoodJob::ActiveJobExtensions::Concurrency module that can be included into ActiveJob classes.

When enqueuing jobs and concurrency is reached/exceeded, jobs are not enqueued and the MyJob.perform_later returns false using native callback behavior of throw :abort

When performing jobs and concurrency is reached/exceeded, jobs are retried using an incremental backoff, using native ActiveJob retry_on ... attempts: Float::INFINITY, wait: :exponentially_longer functionality.

Closes #206.

[reczy]

Does this create a race condition that needs to be accounted for? Say the enqueue_concurrency limit is 1, and the actual enqueue concurrency goes from 1 to 0 between the above two lines of code.

[bensheldon]

I think the race condition is addresses because this is wrapped with a pg_advisory_lock on the key, which should function as a database-wide mutex.

But It's also gonna introduce latency because the PG connection will wait until it can acquire an advisory lock before running the code. If done on the webserver when under heavy load and there's a database statement timeout... that could be bad. I hope that jobs that are likely to need enqueue limits will be triggered from a backend process (rake task, upcoming cron feature, etc).

[reczy]

@bensheldon thank you for this!