New user usability: do not return scope from block / raise on empty block

[wjessop]

Issue 1, queries sent to the replica have missing conditions

When I use .on_slave in my local dev environment (I have two databases set up locally, one replicating from the other) I don't get the results I expect. For instance this query:

=> UserActivity.where("id > 2001").on_slave.count
20012

runs this query on the replica:

2020-06-24 21:25:36.059 BST [86763] LOG:  execute <unnamed>: SELECT COUNT(*) FROM "user_activities"

You can see that it's missing the where clause. If I re-run the query without the .on_slave call:

=> UserActivity.where("id > 2001").count
18025

The call is sent to the primary as expected, but this time with the clause intact:

2020-06-24 21:25:16.601 BST [87228] LOG:  execute <unnamed>: SELECT COUNT(*) FROM "user_activities" WHERE (id > 2001)

If I use the block syntax then the query is sent to the replica and maintains the condition as expected and returns the correct result:

=> ActiveRecord::Base.on_slave { UserActivity.where("id > 2001").count }
18025

2020-06-24 21:34:46.614 BST [86763] LOG:  execute <unnamed>: SELECT COUNT(*) FROM "user_activities" WHERE (id > 2001)

Issue 2, queries sent to the wrong server

If I make a query like this (note on_slave now appears before the .where):

UserActivity.on_slave.where("id > 2001").count
18025

The result is correct, but the query gets sent to the primary and not the replica:

2020-06-24 21:45:45.708 BST [88178] LOG:  execute <unnamed>: SELECT COUNT(*) FROM "user_activities" WHERE (id > 2001) 

Local setup

This is the database.yml I have in place:

development: &default
  adapter: postgresql
  database: connector_development
  encoding: utf8
  min_messages: warning
  pool: <%= Integer(ENV["DB_POOL"] || 5) %>
  reaping_frequency: <%= Integer(ENV["DB_REAPING_FREQUENCY"] || 10) %>
  timeout: 5000
  variables:
    statement_timeout: 5000 # ms
  connect_timeout: 1
  checkout_timeout: 1
  prepared_statements: false
  slave:
    host: 127.0.0.1
    port: 5433

We are using Ruby 2.6.5 and Rails 5.0 and active_record_shards 3.17.0.

[wjessop]

OK, more weirdness. This Ruby:

ActiveRecord::Base.on_slave { UserActivity.where("id > 2003") }

runs on the primary and returns the correct data, whereas this ruby:

=> ActiveRecord::Base.on_slave { UserActivity.where("id > 2003").count }
18023

runs on the replica, though in this instance does return the correct data.

[grosser]

UserActivity.where("id > 2003") returns a scope ... so the query has not yet been executed
it is executed when your irb session tries to print it

-> it would be good to blow up when a scope is returned since that is most likely not what the user intended

UserActivity.where("id > 2001").on_slave.count does not work because .on_slave without a block returns User
-> we should blow up when no block was given

[wjessop]

Thanks for the reply @grosser!

I don't really understand why using a scope would prevent .on_slave from working, does that mean you can't use any conditions at all except on the primary?

I've gone back to the docs but I can't see how to get arbitrary queries to run on the replica. my expectation is that I can just switch queries to the replica using .on_slave, but that doesn't seem to be the case, so what am I missing?

[grosser]

The query needs to execute within the block, just like you showed

UserActivity.on_slave { UserActivity.where("id > 2001").count }

but if the block returns a proxy, it is not yet executed

UserActivity.on_slave { UserActivity.where("id > 2001") }.count

[wjessop]

OK, thanks!

[grosser]

let's leave this open since these are pretty ugly gotchas for new users :)