xGreat article. Two points that require some elaboration…(in my mind, at least)

1. What was nginx used for? You already had an http load balancer to distribute load between your clusters, so not sure what nginx was needed for.
2. Why not stream the data straight to dataflow for processing? What was the advantage of Pub Sub here (I could see Pub Sub if you did not have a custom node.js backend for ingestion).

Thanks