If you have N+1 query it means you’re executing one SQL query per element so if you have 50 emails in an inbox, that’d be 50 SQL calls, right? That sounds like a bug. Well in a Russian doll caching setup, it’s not a bug, it’s a feature. The beauty of those individual calls are that they’re individually cached, on their own timeline, and that they’re super simple.
Now, I have been involved with ORMs and databases for over a decade now, and I have spent large parts of my career working to resolve performance problems in database driven applications.
In a word, the idea that having a larger amount of simpler queries is better is nonsense. In particular, it completely ignores the cost of going to the database. Sure, a more complex query may require the database to do additional work, and if you are using caching, then you’ll not have the data in the cache in neat “cache entry per row”. But in practice, this leads to applications doing hundreds of queries per page view, absolute reliance on the cache and tremendous cost at startup.
In RavenDB, most queries are so fast that when we measured, the network component was the largest cost we had to face.
Let us do the numbers, shall we? Let us assume that we are talking about the best case, we have the database machine and the web machine in the same datacenter. Cost of roundtrip in the same datacenter is 0.5 ms. Now, let's go back to the 50 emails example above, shall we? We need to send 50 queries to the database. We’ll assume that we have a perfect database that takes no time at all to answer. That is still 25 ms wasted just on roundtrip times.
And the problem is that this is usually a lot more than 50 queries per page when you adopt this kind of silliness. You typically see hundreds or thousands of them, and the database isn’t really able to answer you in no time, so expect to see much bigger delays in practice.
But wait, I hear you say, this is all about caching, you are ignoring that part. Well, no, I’m not. If you are using a distributed cache, the costs over the network are exactly the same. So this is only relevant if you are using a cache on the same server, but then you run into issues when you have different caches on different machines hold different information. Not to mention that you are now in the wonderful world of having to worry about cache invalidation strategies and aligning them with the business requirements.
And for fun, what happens one of your cache nodes goes down? You got it, you just created a DoS attack on your own database.
On the other hand, you can actually create proper queries, get the data in as few roundtrips as possible and trust the database engine to do its job.