ConnectionManagement middleware drains connection pool when Exceptions are raised

[orslumen]

As ActiveRecord::ConnectionAdapters::ConnectionManagement middleware does not rescue from Exception (but only from StandardError), the Connection Pool quickly runs out of connections when multiple erroneous Requests come in right after each other.

class TestController < ActionController::Base
  def eat_connection
    # do something with the DB
    raise Exception('lost a connection')
  end

  def release_all_connections
    head :ok
  end
end

Call /test/eat_connection as many times as you have connections in your pool and the application becomes unresponsive, until the connections timeout and are returned to the pool.

Call /test/eat_connection a few times, followed by /test/release_all_connections and the entire pool becomes available again. This is probably what happens with most applications, but it is risky imo that someone who knows how to cause an Exception could potentially bring down the entire application.

Was it a conscious choice to write rescue instead of rescue Exception?

[senny]

The relevant commit is 3b2a032 . It could be an oversight. Let's wait for @tenderlove to elaborate.

[thedarkone]

Yep, plain rescue not catching all exceptions has bitten me multiple times :( (wish Matz wouldn't have made it this way). We should check the rest of the Rails code base for any "naked" rescue clauses.

[tenderlove]

Yep, this is a bug. We need to rescue from Exception rather than a naked rescue.