[Proposal] Schema cache dump #5162

Merged
merged 5 commits into from Mar 8, 2012

4 participants

@kennyj

In my experience, if we had many models (ex. one hundred), Rails boot was slowly.
According to production log, it seems that AR's schema data loading is slowly especially.

Thus I've implemented schema cache dumping. Please review it.
I guess this implementation has many fixing point ;)

Usage:

$ edit config/environments/production.rb
config.use_schema_cache_dump = true
$ RAILS_ENV=production bundle rake db:schema:cache:dump
=> generate db/schema_cache.dump
$ RAILS_ENV=production rails s
@tenderlove
Ruby on Rails member

I like this idea, but can we change a few things?

First, can we just implement marshal_dump and marshal_load on the SchemaCache object? Second, I'm not sure that loading every model is the best idea for the schema cache. What about asking for all the tables and populating the cache that way? For example:

schema_cache.connection.tables.each do |table|
  schema_cache.populate(table)
end

Maybe not a populate method, but something. I don't really like the idea of requiring every model in order to get the schema cache.

I have another idea that is related to this: can we enable schema caching by default? We can use the migration version to determine if the cache should be expired. Maybe add a version method to the schema cache object.

Anyway, I really like this feature.

@kennyj

Thank you for comment ! I'll improve the implement :)

@kennyj

Hi @tenderlove

Done!
Please review new some commits.

@tenderlove tenderlove merged commit 447ecb0 into rails:master Mar 8, 2012
@josevalim

This configuration should not be here. It is specific to Active Record and therefore should be defined in Active Record railtie.

Certainly, I agree with you.. Do you mean kennyj@82bd05a ?

@dhh
Ruby on Rails member

Can you provide some benchmarks for this optimization? How much does it actually speed things up?

@kennyj

I'll provide it, but I've many works during this week. Please, just wait a moment a few days.

@kennyj

Sorry for keeping you waiting for this reply.

I tested about this performance.
But this result was not expected one.

・building environment steps
https://gist.github.com/3730757
・test result
https://gist.github.com/3730759

In my experience on Oracle, the queries to data dictionary were very slow when having many data.
Thus, by similar approache, we solved that problem.

I'll try to research a little more.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment