Race condition for creating ConnectionPool?

[mlangenberg]

https://github.com/arthurnn/apn_sender/blob/master/lib/apn/connection.rb#L5

As you probably know, Sidekiq runs in a multithreaded environment. I believe calling APN.notify_sync from multiple threads will still cause a race condition for the conditional assignment and creation of the client.

    def connection_pool
      @pool ||= ConnectionPool.new(size: (pool_size || 1), timeout: (pool_timeout || 5)) do
        APN::Client.new(host: host,
                        port: port,
                        certificate: certificate,
                        password: password)
      end
    end

    def with_connection(&block)
      connection_pool.with(&block)
    end

APN::Client itself is already lazily opening a socket connect, so in my opinion it does not hurt to initialize the ConnectionPool in an earlier stage.

Another interesting idea is to see if the APN::Client could be a Celluloid::Actor of its own.

[thbar]

Any update on this? I'm discovering apn_sender and wonder if I can use it safely with Sidekiq (which it seems to support).

[arthurnn]

This is a race-condition that theoretically could happen, but I was not able to simulate yet... we use in production with Sidekiq without any problems so far.

[thbar]

Thanks for the reply!