SQLite Busy Exceptions not being treated as cache misses

[Th3-M4jor]

We're experimenting with using SolidCache backed by SQLite in production and found that SolidCache::Store::Failsafe does not include SQLite::Busy exceptions as a possible transient error that would be treated is if it were a cache miss.

Since this is something that could still theoretically occur with sharding and a reasonable busy timeout on SQLite configured, should this be added to the list of possible transient ActiveRecord errors?

[djmb]

Oh interesting. I would expect the SQlite::BusyExceptions to be wrapped up as ActiveRecord::StatementTimeout, which we do rescue - see here.

Sounds like this is not working though - do you know why?

[Th3-M4jor]

I'm actually seeing it be wrapped as a ActiveRecord::StatementInvalid when it happens.

I added the below code to a Rails initializer and it seems to solve the issue for us:

module PatchSolidCacheFailSafe
  private

  def failsafe(method, returning: nil)
    super
  rescue ActiveRecord::StatementInvalid => e
    raise e unless e.cause.is_a?(SQLite3::BusyException)

    Rails.logger.warn("Caught SQLite3::BusyException in SolidCache::Store::FailSafe. Treating as cache miss.")

    returning
  end
end

SolidCache::Store::Failsafe.prepend(PatchSolidCacheFailSafe)

Note: This is with Rails 7.2.2.1

[Th3-M4jor]

Dug a little further and it seems that it's because 7-2-stable does not wrap SQLite3::BusyException as an ActiveRecord::StatementTimeout so the catch all in the abstract adapter will convert it to a StatementInvalid

[Th3-M4jor]

We finished upgrading to Rails 8.0.1 and we were able to remove the patch without issue.

However I'm of the opinion that this should still be fixed if Solid Cache wants to continue to support earlier versions of Rails with SQLite.

[djmb]

This is an issue that would need to be fixed in Rails, and since Rails 7.2 is no longer supported for bug fixes, that's unlikely to happen.