Neo4j Lessons Learned Followup
It’s been just over two months since my last post, where I wrote about building and self-hosting a site on a budget with Rails and Neo4j. Since then, I haven’t made many changes to the site itself — a couple styling tweaks, some new admin controls — but I’m happy to report that it’s been rock solid, extremely fast, and yielded unanimously positive feedback. Go team! There are two updates to that article that I want to address.
First, while researching the optimization of Cypher queries, I came across a mention of a simple way to improve performance when using the embedded database. I checked against the code in neo4j-core and found that it wasn’t following this best practice, so I made the change. It’s been committed but since there hasn’t been (and won’t be) a new version of the gem pushed on the v2.x branch, you should update your Gemfile to reflect it.
gem ‘neo4j-core’, git: ‘[email protected]:andreasronge/neo4j-core.git’, ref: ‘99644256a6’
I do so much caching in Rails can’t claim I saw a huge change on my server, but my tests in the console confirmed that it worked. It changes the way the query engine works and lets the database cache queries correctly. It certainly can’t help.
Second, I neglected to mention a really important lesson from my experience:
Never trust relationship order to be consistent and predictable.
Phillymetal.com uses a node for shows, additional nodes for bands, with a relationship (“playing_shows” or something) between them. The ORDER of the bands is very important. When I tested, I found that it was consistently returning the bands in the order that they were specified after I imported the data.
Unfortunately, a few days after the site launched, I discovered that on my production server, new show submissions would be in the wrong order immediately after the form was submitted, then they’d flip to the right order after the show was modified in any way. It was odd. After a little while later, I found the problem, but it doesn’t matter because I was wrong to assume it would just be in the right order in the first place.
If you want a series of relationships to appear in the right order, put a property on the relationship that you can use to sort. In my case, I actually had been saving the index from the submitted form just in case, I just wasn’t actually using it to sort, so my fix was easy. Don’t make the same mistake.
If you’re in NYC and want to talk about any of this stuff (or at least listen to me talk about it), I’ll be presenting at the NYC Neo4j Meetup on July 21, 2014. Info here. It’s just going to be a discussion and Q&A. I expect we’ll cover a lot of the same stuff as these blog posts but considering how smart the crowd at the last Meetup was, I can only imagine that I’ll learn a bit from it, too.
That’s it for now. Lots more to write about later, hope someone finds this stuff useful!