Interesting compromise from the github issue: "In other to reduce latency we decided to compress and flush on each data frame. The downside is that it degrades

Interesting compromise from the github issue: "In other to reduce latency we decided to compress and flush on each data frame. The downside is that it degrades the compression ratio. The second option is to let gzip to compress until the buffer gets full. I think we could make it configurable."