MountainWest RubyConf 2012

Summarized using AI

Schemaless SQL - The Best of Both Worlds

Will Leinweber • September 18, 2012 • Earth

In the video titled "Schemaless SQL - The Best of Both Worlds," Will Leinweber discusses the advantages of combining the flexibility of schemaless databases with the structure of relational databases, particularly using Postgres. Leinweber shares his experience transitioning from simple CRUD applications with Active Record to utilizing CouchDB, where he appreciated handling document storage seamlessly. However, upon moving to Heroku and learning Postgres, he became excited about its robust features, which include built-in concurrency control, full-text search, and geolocation functions.

Key Points:

  • Schema with SQL: Leinweber introduces the concept of "schema with SQL," which allows for retaining relational database advantages while integrating document features.
  • Hstore Usage: He introduces Hstore, a key-value store type within Postgres, allowing for more flexible data handling without compromising the benefits of SQL. Key functionalities include:
    • The ability to index Hstore data.
    • Querying capabilities for Hstore attributes, enabling users to extract and manipulate data efficiently.
    • Examples of using Hstore for dynamic queries and updates, providing a robust mechanism to interact with semi-structured data.
  • PLV8 Integration: The talk highlights PLV8, which embeds Google's V8 JavaScript engine into Postgres, allowing developers to write JavaScript directly in database functions. This enhances performance and flexibility, with specific points on:
    • Easier creation of complex functions.
    • Example implementation using a Fibonacci function to demonstrate performance benefits.
  • Community and Future Potential: Leinweber emphasizes the vibrant community around these technologies, encouraging innovation and contributing to the enhancements of Postgres.

Conclusion:

Will concludes by urging developers to consider the advantages of integrating features like Hstore and PLV8 into their database management strategies. He reiterates the importance of balancing well-defined schemas with the flexibility provided by document-like storage methods, indicating that this approach can optimize both development and data handling.

In summary, the video highlights a transformative perspective on handling data in relational databases, advocating for a blended approach that leverages both the strengths of traditional SQL and the agility of schemaless data models.

Schemaless SQL - The Best of Both Worlds
Will Leinweber • September 18, 2012 • Earth

Schemaless database are a joy to use because they make it easy to iterate on your app, especially early on. The relational model isn't always the best fit for evolving and messy data in the real world. And let's be honest—it's painful to persist rich domain models across millions of little tables. However as your data model stabilizes, the lack of well-defined schemas becomes painful. On the other hand, relational databases are proven, robust, and powerful. How are we supposed to pick one or the other? Simple: pick both.

Recent advances in Postgres allow for a new hybrid approach. The hstore datatype brings key/value pairs to Postgres, and PLV8 embeds the super-fast V8 JavaScript engine into Postgres. These in turn make Postgres the best document database in the world. This talk will explore when and how to take advantage of them, and how we use these techniques at Heroku.

Help us caption & translate this video!

http://amara.org/v/FGg7/

MountainWest RubyConf 2012

00:00:14.330 All right, so I'm going to talk about an idea I call "schema with SQL". This allows you to have all the advantages of a relational database while also incorporating some nice document features.
00:00:21.180 I'm Will Leinweber, and I'm probably best known for my amazing gem bundle, which installs Bundler when you install it. It's quite popular. I work at Heroku on the Postgres team, where we run numerous Postgres databases. I really like Postgres, and I want to share some of the cool things you can do with it.
00:00:27.270 To provide some context, I started out doing plain Active Record and building simple CRUD apps. That worked well for a while until I found CouchDB, which I fell in love with. It has great features like a nice HTTP RESTful interface and MapReduce capabilities, but what I loved most was the ability to store documents as regular objects in Ruby.
00:00:53.010 Using tables and maintaining a third-normalized form was often painful. My web applications didn't always deal with relational data; making many tables in pursuit of normalization just felt wrong to me at times. The last project I worked on involved a tool for artists to version their songs and present those versions to their fans. Each song had a title and pointers back to the artist, along with other metadata. Versions of the songs included details like changes between versions and the date created, along with a list of tracks.
00:01:35.440 While trying to model this in SQL, I faced a problem with the file part—justifying the presence of a files table didn’t seem logical since a file only made sense in context. Although I could have worked around this in SQL, my logic wasn't up to the task, so I used CouchDB, and it worked great.
00:01:58.569 Then I came to Heroku and began learning Postgres, which is a fantastic database that features multi-version concurrency control and phenomenal full-text search capabilities. The geolocation features and the latest version from Postgres includes K-nearest neighbors, allowing you to find the closest items on an index without sifting through many records.
00:02:06.110 Postgres also has built-in listen/notify features, and its various data types are incredibly robust. I could deliver an entire talk on the amazing features within Postgres. However, the aspect I missed greatly was transaction handling—being able to roll back changes. Now, whenever I face a daunting migration, I spin up a fork of my database and execute a migration to see if it works. If not, I can roll it back and keep trying until I get it right, which saves me substantial time.
00:02:44.350 However, I still missed the document model. I should say that I used to miss documents, but I don't anymore—and I'll explain why.
00:02:51.010 I found two features, Hstore and PLV8. One can be used today, while the other is still somewhat cutting-edge and not quite ready for production use. I’m excited about both. Let’s start with Hstore. Hstore provides a key-value column type within Postgres, which means you can have your standard table with one column designated as an Hstore document.
00:03:11.350 One fantastic aspect of Hstore is that you can create an index on it. How many of you in your applications have a generic serialized table where you treat a serialized column as an opaque object without being able to query it? Well, now you can actually query it!
00:03:40.970 Let me show you how Hstore works. Here's an example query. It selects a string and casts it into an Hstore type. For example, if 'X' is 'A' and 'Y' is 'B', you can use the arrow operator to extract values from the Hstore. For instance, if you're asking for the 'A' key, it returns 'X'. Also, you can concatenate two Hstore objects, but be cautious because if keys overlap, the existing keys will be overwritten.
00:04:06.670 You can also subtract keys from an Hstore document; this works as long as the keys match specifically. Using this knowledge can lead to significant flexibility in how you handle your data.
00:04:42.950 There are many more operators available; for more details, the Postgres documentation is quite helpful. Another exciting feature is that you can utilize Hstore in a WHERE clause to select products based on certain attributes or conditions. For example, you can select all products that are red based on an Hstore attribute.
00:05:14.160 You can also perform updates. For instance, you could change all color attributes of products to blue and add new attributes if they do not exist. You can even join other tables based on Hstore attributes, so you have all the powerful features of SQL available while accessing serialized columns.
00:05:42.490 The standout feature is the ability to create an index! With a functional index, you can access specific keys quickly, which significantly enhances performance when querying. Hstore indices can greatly optimize how you interact with your semi-structured data.
00:06:07.290 Using a Postgres 9.1 database, all you need to do to integrate Hstore is create the extension, and you're set. Active Record now includes built-in support for Hstore, so you can utilize this feature seamlessly in your applications.
00:06:44.890 One of my colleagues demonstrated this capability through a live demo. You can check out that Hstore demo to see how it behaves in practice.
00:07:14.830 Next, I’d like to discuss PLV8, which embeds the V8 JavaScript engine into Postgres. For those unfamiliar, V8 is Google's JavaScript engine, which is also what Node.js uses. The benefits of PLV8 are substantial for those of us in the Ruby community, as it allows for the writing of trusted JavaScript code within Postgres.
00:08:09.970 Installing PLV8 is straightforward. You can build it from source, and once installed, it enables the use of JavaScript inside your database functions. This is particularly exciting for creating functions that handle complex operations or data transformations.
00:08:56.379 To demonstrate, I want to present a simple Fibonacci function implemented in PLV8. The approach of using JavaScript to perform calculations can greatly improve execution speed compared to using traditional procedural SQL. The ability to leverage JavaScript alongside SQL presents a powerful combination for developers.
00:09:28.550 While PLV8 might be experimental, it opens up opportunities for optimizing Postgres functionality and enhancing application performance. The community is vibrant and maintaining these enhancements contributors can positively impact how we use databases.
00:09:47.799 As we continue, I'd like to showcase how JavaScript can be specified in SQL. Here’s an example where you can create a JavaScript function that interacts with JSON data stored in your database. Storing data in JSON format can be beneficial for flexible data structures.
00:10:37.799 In conclusion, utilizing features like Hstore and PLV8 can revolutionize how we handle data in relational databases, allowing us to benefit from great flexibility and powerful performance. The importance of combining these document-like features with well-defined schemas in Postgres can't be overstated.
00:11:23.650 I encourage everyone here to explore these options and think critically about where your data exists and how best to utilize it. There are ample opportunities in considering the locality of your data and optimizing your data layer accordingly.
00:11:46.470 Thank you for your time, and I'm happy to answer any questions!
Explore all talks recorded at MountainWest RubyConf 2012
+11