FriendFeed Implements Schema-less Storage Atop MySQL
Bret Taylor of FriendFeed describes a "schemaless solution" to the problem of "storing data with flexible schemas and building new indexes on the fly," for a rapid growth website. The problem itself comes from the need to constantly add new features and update the underlying database structure, and the millions of existing records stored in that database, to support both the old and new features. FriendFeed's solution was to build "a schemaless solution on top of MySQL instead of moving to a different technology base. Bret describes the basic problem:
In particular, making schema changes or adding indexes to a database with more than 10 - 20 million rows completely locks the database for hours at a time. Removing old indexes takes just as much time, and not removing them hurts performance because the database will continue to read and write to those unused blocks on every INSERT, pushing important blocks out of memory. There are complex operational procedures you can do to circumvent these problems (like setting up the new index on a slave, and then swapping the slave and the master), but those procedures are so error prone and heavyweight, they implicitly discouraged our adding features that would require schema/index changes. Since our databases are all heavily sharded, the relational features of MySQL like JOIN have never been useful to us, so we decided to look outside of the realm of RDBMS.
After reviewing several possible solutions, they decided to use a custom "schema-less" solution on top of MySQL instead of opting for a completely different persistence solution.
Their solution was to separate the primary data and indices to that data. "Our datastore stores schema-less bags of properties ... We index data in these entities by storing indexes in separate MySQL tables." This resulted in a trade-off between volume and efficiency.
As a result, we end up having more tables than we had before, but adding and removing indexes is easy. We have heavily optimized the process that populates new indexes (which we call "The Cleaner") so that it fills new indexes rapidly without disrupting the site.
Separating the data and indexes raised issues of consistency and atomicity. Instead of a strict transaction protocol, they made the database tables canonical and used the indexes only for references and reapplied filters to the actual database reads. They enhanced the automated process that keeps tables updated, called 'the Cleaner' to continuously update and fix indexes with updated entities having priority. Inconsistencies, although possible, were cleaned up in less than two seconds on average.
Using Average Page Latency as a metric, Bret reported the following trends.
- Overall -- a significant decrease in the face of a rising trend.
- Last 24 hours -- stable even during peak hours.
- Prior week -- "dramatic" drop.
Bret's post generated a lot of feedback. One theme, "modern RDBMS do not share the limitations of MySQL when it comes to schema evolution." which ignores the cost issue behind the original choice. Other readers responded with alternate solutions.
It is interesting that no one pointed out the similarity between FriendFeed's solution and the "ancient" technology of ISAM - Indexed Sequential Access Method. ISAM used the same basic architecture - separating data and indices with automatic updates of the indices when data was changed.
it begs the question