Asynchronous logging doesn't work properly for TCP and UDP

[warnickr]

When synchronous logging is disabled (sync: false), nothing gets logged to TCP and UDP endpoints. I think this is because the socket is never flushed after being written to. If I modify LogStashLogger::Device::Connectable#flush as follows then it works.

def flush(*args)
  if args.empty?
    buffer_flush
  else
    messages, group = *args
    write_batch(messages, group)
  end

  @io.flush
end

[warnickr]

I created a pull request that addresses the issue.

[dwbutler]

There is a detailed explanation of this behavior in the troubleshooting section of the readme here. If you require sockets to be flushed immediately, it's up to you to set the sync: true setting. Most applications should keep the default setting of sync: false and allow Ruby / the OS to decide when to flush its internal buffers.

[warnickr]

That makes sense. In my experience, the unpredictable and often extended delay renders sync:false unusable. Wouldn't it be useful to allow asynchronous logging that batches log writes while still maintaining some level or predictability?

[dwbutler]

sync: false is a good default for applications that have a high log volume, because it improves performance. I can understand why you may want more predictable logging behavior in a low volume application.

There are several possibilities.

• Implement an optional mechanism where the underlying IO device is also regularly flushed.
• Implement an optional Rack middleware which flushes after each request. (Very useful for web applications.)