ActionCable can deplete AR's connection pool

[AndreasWurm]

Normally ActionCable::ServerWorker::ActiveRecordConnectionManagement takes care to ensure connections are checked back into the pool. This is not ensured when using ActiveRecord from within a ActionCable::Channel::Streams stream_form/stream_for callback.

Observed this behaviour in rails 5.0.0.beta2.

Example:

class MyChannel < ApplicationCable::Channel
  def subscribed
    queue = "my_channels_queue"
    stream_from queue, -> (message) do
      # open a db connection, this will not be returned to the pool
      ActiveRecord::Base.connection
      transmit ActiveSupport::JSON.decode(message), via: queue
    end
  end
end

[maclover7]

If a user is manually calling ActiveRecord::Base.connection, shouldn't they have to return it to Active Record when they are done using the connection? Not sure it is Action Cable's responsibility to cleanup here...

[AndreasWurm]

well its just an example.. if I do MyModel.find(1) or any AR call instead the same thing happens, and my expectation was it just works and its taken care of.

The Readme states

You have access to your full domain model written with Active Record or your ORM of choice.

and has examples using active record without managing connections manually, and it works everywere else since ActionCable::ServerWorker::ActiveRecordConnectionManagement takes care of things, just not in the provided example. Personally I consider this a bug since I dont think anyone would expect that for this method its their responsibility to manage the pool, when everywhere else its taken care of.

[zhu1230]

+1

[benoitongit]

Same issue, I get the following error after using ActiveRecord from an ApplicationCable::Channel class:
"ActiveRecord::ConnectionTimeoutError - could not obtain a connection from the pool within 5.000."
Tested with last commit: 22c318d
Also good to point out that this issue does not happen when testing with this version: 49f6ce6

[dhh]

@jeremy Same issue as we were seeing, it appears.

[doits]

After updating to 5.0.0.beta3, I get a lot of ActiveRecord::ConnectionTimeoutError when running a large test suite of my rails app (including a lot of feature specs). With 5.0.0.beta2 it does not happen. Only mysterious thing: I don't use ActionCable at all. So either I hit a different bug or it is the same but the source doesn't lie in ActionCable but goes deeper.

[dhh]

@jeremy ^

[chashmeetsingh]

I tried many configurations for the gem file bouncing between beta2 and 3 and found that the issue is caused by the active job gem beta3 , beta2 works just fine
Hope this helps!
https://github.com/chashmeetsingh/WebsocketSample