Extract database connection logic about split

[rentalcustard]

👍

[andrew]

I've started working on extracting the redis calls on this branch: https://github.com/andrew/split/tree/db-adapters

[josedonizetti]

What is missing for this to be finished? I can help on that.

[andrew]

I basically need to get the tests passing on the mongo adapter, refactor it a little and create the ability to switch adapters in the configuration block.

[josedonizetti]

Ok. I'll do that, and send a pull request for you to review!

[andrew]

Awesome, thanks!

The redis adapter may need to be changed as well to make a nicer internal API for others to create adapters.

[josedonizetti]

The mongodb tests are green on my fork, branch db-adapters. But there's still lots of refactoring to do. But first I will impement the support for mysql. I want to run a stress-test between those versions and see how they go, and then improve the interface. Let me know your opnition about that.

[andrew]

Sounds good!

[thomas-atwood]

Given the default use of Postgres on heroku — how about a version that runs with Postgres. Not sure whether it should run in the same Postgres database as the rest of the application, or in a separate one.

[buddhamagnet]

Best approach would be to abstract the Redis interface into an adapter and merge. Then the road is clear to build new adapters. I guess we are quite a way along the road and bearing in mind the amount of work that has been done in master since merging may well be tricky.

[buddhamagnet]

On that note, just had a dig into that branch and it strikes me that "backend" is a better description of the data source than "database" as we could be storing the alternatives etc anywhere. Have just pushed up a branch along those lines.

[buddhamagnet]

Actually, going to trash that branch and look again.

[buddhamagnet]

Down to two failing specs, @andrew you may be in a better position to deal. If this branch is good when the specs pass, we have a bang up to date adapter branch that defaults to redis, which means we can start rolling in the refactoring suggested by code climate hot on its heels.

[buddhamagnet]

Also added an example driver.rb file for new adapters.

[buddhamagnet]

One spec left to pass!

[buddhamagnet]

The other thing to note with this branch is, as we have now abstracted the data source, we can't test using MockRedis as we are now calling adapter methods (ie we call backend.clean - which does not exist for MockRedis). So back to testing against the real backend for a while.

[woodhull]

Why is this a good idea?

Seems totally reasonable to require redis.

[buddhamagnet]

Er...because we are building an adapter layer so the backend may not be redis?

[woodhull]

Sure. I know. It just doesn't seem like the sort of thing that I would allow if I were maintaing the gem.

I tend to think that it would make things fairly complex to maintain for questionable benefit. See discussion of the same issue for sidekiq: https://github.com/mperham/sidekiq/wiki/FAQ

[buddhamagnet]

Take it up with @andrew. Oh and happy new year!

[andrew]

When I first built split we used redis because that fitted the use case at the time, but we're not really using any special parts of redis so there is not really much reason to limit it.

There have been projects where I've added redis to the stack simply to use split, where I'd prefer not to add a second database to manage in production if I can avoid it.

[andrew]

I'm going to close this as we've not made much progress on it in a while.