Creates batch adapter

[tpitale]

• Size-bounded queue using ruby stdlib SizedQueue for thread-safe queueing

TODO:

• test what net/http, faraday adapters do when given an array (must match batch body for GA)
• discuss clear behavior
• usage documentation
• update our usage of net/http to support array or string

Resolves #57

[tpitale]

@wjordan: Here's a quick spike. One issue (seemingly present in all the queueing/buffer code I looked at) is: how/when to trigger the "final" flushing of the buffer.

I've added a method called clear on the adapter, I could delegate to this from the tracker, but the application using this would have to somehow know when to call it. One example, a rails app, is probably not aware when the server crashes/quits. Events in this case would remain un-sent to GA. Seems less than ideal.

[wjordan]

@tpitale Awesome progress! This looks feasible. Flushing a partially-filled buffer can be handled a few different ways:

1. Async flush operation the application can manually invoke to flush a partially-filled batch, e.g. at the end of a per-request ensure block (ruby-stud :force option)
2. Configurable interval timer that automatically flushes a partially-filled buffer after a maximum delay
3. Blocking flush operation the application can manually invoke to definitively flush the remaining buffered items e.g. on teardown, ensure block, etc. (ruby-stud :final option)

the last one is already covered by the existing clear method. For 1-2, we would need clear to support a partial flush without killing flushing_thread. This could be added to this PR by pushing a Null Object to the queue (to wake up the waiting flushing_thread), maybe something like:

FLUSH_OBJECT = Object.new.freeze
def clear(final=false)
  @queue << FLUSH_OBJECT
  if final
    @flushing_thread.kill
    @flushing_thread.join
    flush(true)
  end
end

def flush(final=false)
  param_array = []
  while param_array.size < @size && (object = @queue.deq(final)) != FLUSH_OBJECT
    param_array << object
  end
  rescue ThreadError
  ensure
    @adapter.post(param_array)
  end
end

[tpitale]

Okay, so you're thinking the use case would be to flush it at the end of a request, so that N events would get queued up and make a minimal number of calls to Google.

I think I can make some tweaks to make that work.

[sebbean]

@tpitale i commend how long you've been working on this thing! it has really help us out in our usecase (analytics for an iframe embed interactive widget, frontend analytics is being blocked)

[tpitale]

@wjordan @sebbean Are either of you using this? It has been sitting for a long time, and I'm not sure if anyone really needs it anymore …?

[tpitale]

@bots-business I could change this PR so that no flushing_thread is created if the flush_timeout is explicitly set to nil. This way the tracker and adapter are still thread-safe, but no additional thread is created to perform auto-flushing of the queue.

[tpitale]

Closing this as old and not sure it's necessary yet in V4.