> By using the includes method we've been able to knock our queries down from six to two: The first to find the events, and the second to find the categories for those events.
Am I the only one that thinks rails/activerecord is crazy here for (often) preferring two trips to the db over a join on indexed primary keys?
Are there really (sane) use-cases where that is better?
You mean
SELECT u.* , c.* FROM users u INNER JOIN comments c
ON u.id = c.user_id WHERE u.id = <...>
If the average user record includes a non-trivial amount of data and there are many comments, you're repeating a lot of information in the query results.
disclaimer: I'm currently doing some work for pganalyze.com (the blog host)
I'm not sure about "non-trivial amount of data" - especially if we're talking about a single row (single user) here? I suppose there might be a few blolbs of large json documents?
And I'm not sure I understand "repeating information in query results" - surely we're talking about receiving a query, transforming it and responding with the data?
In a rails+graphql ideal world, that'd maybe mean Parsing the json query, generate some simple ruby (active record) code, fire it off to the db (let ar /arel do its job; generate sql) - serialize to json and write the response?
If there truly is non-trivial data in fields not requested, it might be worth it to naemrrow the columns selected. But I doubt it, in the general case.
For a query like
SELECT u.* , c.* FROM users u INNER JOIN comments c
ON u.id = c.user_id WHERE u.id = <...>
u.id, u.name, c.id, c.text
1 Bob 1 you are right
1 Bob 2 you are wrong
1 Bob 3 you are right again
I didn't consider the kind of expansion you're alluding to - on the other hand i can't ever remember seeing an indication that the amount of data returned is an issue with app performance.
But I've certainly seen the number of round-trips result in real issues. And complicated joins, for that matter.
But you're right, there are always trade-offs, and there are some situations where the current AR approach is not ideal, and a single join would be more efficient. However, I suspect this is the way AR does it because it's right enough often enough that it works well as the default approach.
And in this example, the data is definitely compressible, but the Postgres wire protocol does not implement compression (it leaves that to tls right now, though CRIME makes that a problem). Adding native compression to the protocol has been on the TODO list for a while: https://wiki.postgresql.org/wiki/Todo#Wire_Protocol_Changes_...
Modify the original sample to load two relations: fetch all events with their first 5 categories and all their users. If you use two joins then you get a combinatorial explosion because you are now getting all permutations of the (event, category, user) tuple.
(event1, category1, user1)
(event1, category1, user2)
(event1, category2, user1)
(event1, category2, user2)For instance, if you have a simple active record type:
class User has_many :comments; has_many :likes; end
User.includes(:comments, :likes).find(user_id)
Generates 3 queries, but importantly the number of records returned is appropriate.
User.joins(:comments, :likes).where(users: { id: user_id })
Generates only a single query, but the number of records returned from this join is the product of comments*likes for this particular user.
I find the includes stuff easier to reason about as the amount of associations you need loaded in memory becomes more complex.
Not "user has orders", and also "user has favorite products".
(That is, it's hard/impossible via the active record DSL to make the correct nested join?)
By the way (you may know this) when staying within active record convention, you should never need to single out the ".id" and ".foreign_model_id" fields:
User.joins(:comments, :likes).where(user: some_user)
(where some_user is an instance of User).
I'm still not quite comfortable enough with AR, but I think, leaning on this: https://www.learneroo.com/modules/137/nodes/768
I think what you want in your second case is simply?:
User.joins(:comments, :likes).where(user: some_user).uniq
(I'd have to create some models and watch what ".to_sql" creates to be certain, though)
You encounter this problem whether you use an ORM or not.
Also from my experience the performance problems are completely divorced from ORM/not ORM. Usually the reason why you drop the ORM is that you want to take advantage of features that are not modeled by the ORM. Most query optimizations simply involve trying to do everything in as big of a batch as possible. Whether you do batching with or without the ORM isn't really relevant. The reason why people don't batch their queries when they use ORMs is because it's so trivial to not do so but there is no reason why we can't build better ORMs that make batched queries as easy to use as non batched queries which would get us both performance and convenience at the same time.
And if it is not acceptable, you can still optimize the living excrements out of your GraphQL resolvers.
Also, the burden of doing more than one round trip usually moves to the client. So there's always some trade-off to be made.
Having said that, I don’t mind the second request model myself. It’s actually pretty clean and when you take the dB roundtrips out of it, the dB will generally do the same work anyway.