Add delayed enqueue for workers that support it

[gaffneyc]

A couple forks have added delay support. I don't fully understand the reasoning or what is causing the need for it. My best assumption is that the worker system is pulling the index job off the queue before the record can be saved to the database.

[vinniefranco]

Hi @gaffneyc

The reason I added it was to deal with a race condition I was having between my services. I would have a 99.5% miss with a index job on it's first try, since the job would try to run before the record actually existed. Since indexing was not something that needed to be async, I added the functionality to have it as a scheduled job with a configurable delay.

[gaffneyc]

@vinniefranco that's interesting, I'm curious what could be causing the index job to be enqueued before the record exists in the database. Sunspot should be calling index during after_commit and we do a check to make sure the record has been persisted.

Was this with ActiveRecord and a single master database? Was the record being saved through an association?

I think adding a delay is the quickest and simplest approach but I would like to understand that root cause if possible.

[gaffneyc]

Actually... it looks like the index is being done in the after_save callback and not after_commit.

[vinniefranco]

Yeah, I figured as much since it had a id passed and yes, you are correct the record was being saved through an association. Single master

[xinuc]

Yup. Pretty much what I've experienced.

The worker would run the index before the actual record saved. Usually, I just need 1 sec delay to get around this problem.

[quake]

I'm using delay_for to resolve race condition issue: sunspot is using after_save callback, index job will start before db commit sometimes.

[coorasse]

Same problem here with Sidekiq. The job starts always before the object is saved. I can't understand if there is a reason not to put the callback in after_commit instead of after_save callback

[britg]

I'll throw my hat in - race conditions started happening after a added additional 'after_save' callbacks to my model. Near 100% lookup miss on first try.

[ghempton]

Hopefully if sunspot/sunspot#534 gets merged this problem will go away.