Chris Oliver

Hey Toshiki, this is a great question!

You're right on the tradeoffs here. Here's some thoughts I have on it:

STI is great when you have very similar models that are unlikley to change or won't have many different columns. For example, you wouldn't want to do this if you had two types of User models but each one had 15 different attributes unique to each other. Then you have 30 columns on one table and you'd only ever use half of them at a time. That means you've got a mess of attributes that you don't use and makes for a confusing time later on.

In your case, the models are pretty similar and you don't have that many additional columns so STI is not a bad fit. Yes, you'll have a couple columns that are null, but that's not a big deal.

As for polymorphic associations, they've absolutely got their downsides too. The primary one being additional complexity and no ability to query efficiently. Since every record has the class name stored inside the column's value, you can't add foreign keys which is very unfortunate. It's great for things like comments where you often will be querying for Movies, Actors, or other models and their associated comments and never the reverse of looking at comments and trying to find their "commentable" records. You can index the commentable_type and commentable_id columns together to make finding a Movie and all the associated comments a fast query, but there's no way to do uniqueness or anything like that on a database level.

Another downside with the polymorphic associations are that you now have a whole bunch more tables and you can already see it feels overkill in this situation. As any project grows, more tables means it's harder and harder to wrap your head around how things work as time goes on.

So what I would say is this:

• If you imagine your models aren't going to change hardly at all in the future (it seems like they won't), then STI doesn't seem like too bad of a solution here. You've got very similar models, and there's only 3, and you only add a couple columns unique to each.
• If there's a possiblity you might be adding more columns or several new Invoice Item models...then you might be better off going with polymorphic associations instead so you don't have one bloated table with 60 columns and most of them being null values.

Then the real conclusion here I would probably give you is: start simple and just pick one that seems most intuitive to you right now.

You honestly can't go too wrong with either approach here. The one thing that you're trying to optimize for is the future. You can have a best guess as to what will change in the future, so use the path that seems most in line with that. If something changes and it turns out the solution you chose wasn't ideal, then you can always write migrations to move data into another structure and change your models along with it.

You will always have that option to reorganize your data so even if you pick the wrong solution, you can always fix it and refactor.

Great stuff! I would do the same thing. Since I hadn't paid too close attention that you were using a gem, this makes more sense for that situation. My bad! :)

You'll find out pretty quick, and sometimes you might get exceptions you might want to ignore or whatever like a bot typed in weird crap into the URL that Rails can't parse and it's fine to ignore. There's always those odd things to deal with. I've been really happy with it on those similar apps where I want to know if something goes wrong, but it definitely doesn't justify money or running another service for it.

Let me know how it goes! And now I really want to check out Errbit myself...

I haven't used errbit, but I've deployed exception_notification to like 30 apps and it's always worked fantastic. Errbit looks cool, but that requires you to run a separate app?

The nice part about exception_notification is it's super simple and can post to email or slack or whatever is convenient for you without much setup. The downside is it doesn't collect messages anywhere, so you'd have to track them inside Slack or Email or whatever and that can be a fairly big downside if you have to manage a lot of exceptions.

One of the downsides of doing that is you're faking out objects so not all your code is fully tested there. VCR would only fake out the HTTP response and you could fully test all of your code instead. Sometimes that can save you from some future troubles where a mock object wouldn't.

On the test double side, I've used rspec-mocks before and it worked pretty well. Most of what I was doing I just used the readme for https://github.com/rspec/rspec-mocks#test-doubles

Basically creating the double and then passing that object into the classes I wanted to test. Then you can expect it to have certain methods called and such which is pretty convenient and mock out the return values.

Those are what can get you into trouble if your object's API changes in the app but you forget to update your test doubles to match exactly. That's why I usually recommend not using them as much as you can so you know you're always operating on the same exact objects the app will have.

One thing that comes to mind here is actually the VCR gem. You can use it to record network requests and save them so that your tests can just replay a real request. It sounds like you're describing something that might be a perfect fit for VCR: https://github.com/vcr/vcr

Have you looked into that at all?

I was going to say, I don't think you can do that because ActionCable doesn't go through Rails controllers. You'd want to set self.current_user equal to the logic that finds the user from the session.

What was your solution? I'm sure other people would love to know it as well!

Awesome post, thanks for sharing Michael! :D I need to do an episode on Cloudfront sometime.

Like Andrew said, Searchkick is perfect for this. Each model gets an index inside ElasticSearch and then searchkick lets you query over those by simply doing this: https://github.com/ankane/searchkick#multiple-indices All that should make sense after watching the screencast. 😎

Ransack isn't quite the same in that it doesn't index them in a way that you could search multiple models together. You'd actually have to search each table separately and then combine the results in your view. It's nothing fancy as it just does database queries which makes it really easy to implement without any separate services, but more rudimentary in a sense.

This is a great question Karim. :D Here's some thoughts from when I was consulting:

• The most important thing for me was to just get work at first. Any work at all would lead me to more people and build up my portfolio. Over time I'd "trade up" and get better and better jobs ignoring a lot of ones I probably would have picked up before that were bigger time sinks than I wanted.
• I actually just emailed a random guy on a job board and got my first consulting project that way. The majority of my work afterwards was actually from spending a lot of time in the startup community in town. All kinds of people want to start a business and need a developer to help them build it. There is always a lot of work available if you're around those people.
• I'd charge pretty much any rate you feel comfortable with. The ideal is to know you can deliver on their project with a fixed cost, but that's rarely possible to do well. Too much changes in most projects so to stay safe you can always charge hourly. To make things a bit easier on yourself, consider charging per day or per week and they have to book either a full day or full week and that way you're not losing out. You'll do well to find out the client's budget and then work with that (see what you think you can deliver for that, it might not be the full project or whatever if it's too small but you can help them start). Raise your rate by $5/hr each new client or something as you get more experienced.
• Lots of different projects was great for me to keep improving skill wise. And then for each new client I'd try different things. Like I went from hourly to daily to weekly and then tried some fixed fee projects so each new client I'd tweak one piece of the wway I worked to see how much it improved my life and if I liked it or not.
• I didn't really do any marketing because the best clients I got were actually from referrals so I focused on that. Rarely did I get good work that would show up randomly so I didn't put any time into marketing which was nice for me. A lot of people will do marketing for this, but you'll need to do something totally unique to stand out from all the other developers who claim they can "take your ideas and make them real" or any of the thousands of variations every consultant's website has. 😛

Happy to answer some more questions on this and I would love to hear about other people's experiences. I spent like 7 years consulting and made a really good living doing it.

Exactly, you can just replace that with however you find the current user normally. 👍

That turned out nice and simple! :) I hadn't actually customized it to be username/pw but that looks pretty much exactly like what I imagined you'd have to do. Nice and simple, thanks for sharing your solution. 👍

Hey Naveen,

In this case, it looks like you created a select without a matching form tag. The CSRF token actually gets generated by the form_tag and form_for helpers so if you just wrap your select with form_tag, you'll be fine. You can then move your URL, method, and remote options to the form_tag and have that all organized nicely and that should fix your issue.

The problem you saw was that, while you can try to put the remote: true options on a select tag directly, it isn't going to generate the CSRF token to send along with.

This is exactly what the counter_cache option does inside ActiveRecord, it's just hidden away. You get the performance benefit from querying, with the slight tradeoff of inserting/destroying records taking 1ms or so more. That's totally fine because most applications are much more read heavy than write heavy.

As for the conditional counter caching, you're right, nothing in Rails to do that. You could probably turn that into a concern that could be reusable that accepts a proc you call to replace the `if user` portion so it could be reused and customizable in any fashion. Might even make for a nice little gem!

Another tweak you could do with that is actually do the COALESCE like you mentioned, but like Rails does, ignore the count and actually just add 1 or subtract 1 from the current value. This way you don't have to also query for the total count which would speed up your implementation slightly. It does assume the value is accurate, which is totally fine since like we saw in the episode that you can just set that when you add the new column. This way you're making a micro update to just increment or decrement the counter column rather than doing that plus selecting the count. Adding or subtracting 1 each time should only ever take a millisecond or less while the count might take several.

This was a great question!

Haha! It just might be. :) Someone was also asking about counter caches in Slack so it seems like perfect timing!

Just coffeescript because they removed curly braces. They use tabbing to denote which block it is a part of instead (like Python).

Hmm, not sure. It looks correct assuming each of those lines is tabbed over properly

$(document).on "turbolinks:load", ->
  $("#new_message").on "keypress", (e) ->
    console.log e.keyCode

You can add your own ID or class to any element you have in your form and change that query to match. I just use that because it's auto-generated and easy to use, so you can use the autogenerated one for your form, or specify your own and that should do the trick. 👍

Sounds like a good pull request for someone to submit to the repo if they have time! *hint hint*

Masud, I think you'll have to benchmark your app and monitor memory usage and things to find out what it's actually using and if it's going to surpass your limits. There's no real way to tell ahead of time, but you can also try deploying the change and making it easy to roll back in case you do surpass your limits.

If it does look like it's going to be way too much, you can always try using http://anycable.io/ which can handle 20,000 idle ActionCable connections within 1GB of memory.