Rails feature that you've never heard about: schema cache

I’ve been doing Rails development already for more than five years and only now I learned about the schema cache, although this feature is mostly relevant for apps under the massive scale.

What is it for?

On boot, the Rails process makes SHOW FULL FIELDS query to get information about the database structure and columns types. This is required in order to know that the created_at column is DATETIME and we have to cast all values as a date time objects.

Now imagine that you have a hundred of Unicorn processes and you restart them in your production cluster. Every process will query MySQL with the same query: SHOW FULL FIELDS. Keep in mind that this query is quite expensive because it’s not optimized as may be SELECT with an index.

A hundred application servers all making an expensive query at the same time may also kill your database! To avoid it, Rails provides with the schema cache feature. The idea is:

  1. Serialize data about the schema (tables, columns, and types) into a file
  2. Distribute that file over all application servers
  3. Load the data from file to avoid hitting the database

Here is the chart with a number of queries not using index: you can totally see the moment when we enabled the schema cache!

Implementation

In Rails <= 5.0, schema cache is serialized and persisted in Marshal. In Rails 5.1, I changed schema cache to use YAML to preserve compatibility of serialized cache between different Rails version.

Schema cache is enabled by default for all Rails apps, but it won’t be used unless you have prepared the dump file on your app server with rake db:schema:cache:dump.

Anyway, the SchemaCache class is only 100 LOC and I suggest that you check it out.

Why you may need it?

Using schema cache may not be worth it with only a couple of application servers, but you may start using it as you scale your app and the number of Unicorns grows.

Further reading

I have a plan to write more about Rails features that are not well known. Please let me know if you are interested in this topic.

Comments

comments powered by Disqus