Christopher Smith

Fountain codes require more buffering of data on both the sender and receiver side, which means increased latency. There is certainly a case for it, but it seems in conflict...

Google does do 204's, both with DoubleClick and just on their search home page (when I go to the Google home page it actually calls out to https://www.google.com/gen_204 ;-), as...

GA is a different story, because they've got backwards compatibility needs from hell (all those "utm" parameters date back from when it was Urchin!). They still support no javascript style...

Makes sense to me. I was going to implement it as another handler that was available and let the path from the client side dictate the response.

I find it simpler just to have a Kafka Connector that swallows from the divolte producer's topic and reroutes the event as needed (generally publishing to another Kafka topic). Sure...

Yeah, I grok the use case for having both. I hadn't thought about using the HDFS client API with the s3 URL. I may try that out.

As requested, a pull against master.

Take a look at #111, I think it avoids losing the micro-batch (though perhaps there is a case where the entire batch needs to be discarded). Sure it is _possible_...

Actually, I'm struggling to imagine what state is in the _thread_ that will get cleaned up by discarding it and drawing blank.

What's needed here is a way to represent the device mapping for ephemeral storage. Ec2 supports it.