Publishing separate or combined SDK + collector layers

[anuraaga]

I want to confirm what we want to publish for layers. Currently, the terraform scripts built a separate layer for collector and SDK. This makes it easy for users to add only the SDK, and use a separate exporter like Zipkin for example without including the collector. But it means that for collector users, we end up using 2 out of the 5 max lambda layers of a function. We could

• Only publish combined layers
• Only publish 2 separate layers
• Publish separate layers for SDK and collector, and also combined ones for each languages

If the layer limit was much higher, I would recommend publishing separate layers to treat as lego blocks for a user to add to their function. But because of the limit, I wonder what folks think of these options. I am starting to lean towards only publishing combined layers at first, and possibly going with both combined and independent layers if there is demand. Thoughts @codeboten @wangzlei?

[anuraaga]

We discussed in the SIG and are going with 2 separate layers for now - this mainly is to make it simpler for downstream customization, which may only need to customize the collector without maintaining several language layers. Keeping things separate in upstream will allow such downstreams to only provide collector layers while pointing users to the SDK layer here.