I would like to point out that the idea here is actually to not continuously allocate new []byte instances multiple times per each push.

You can see my original comment here

As well as somewhat relevant one about usage of sync.Pool here.

Depending on how well this works we might want to reintroduce sync.Pool. Or even just recompressing/rencoding the body on each time we need to send it.

tl;dr: the point is not just to use the snappy-framing support, but to use it to optimize how we send the packet.