[Feature Request] Add support for multiple backends to persist trace data

[ayansen]

Currently haystack traces supports only the cassandra backend. Adding a new backend would mean writing a new driver in haystack-traces-commons module and using it in the reader and indexer.
This design has a problem when we create multiple drivers all of the dependencies get added in the main jar and can cause conflict.
This issue is created to discuss if we can come up with a cleaner way to add new backends for traces

[ayansen]

Proposed design for the new architecture is documented here :
https://github.com/ExpediaDotCom/haystack/wiki/Haystack-Traces:-Supporting-Multiple-Backends

[codefromthecrypt]

ps I'm curious how this got to be more difficult :) you know grpc has a really dirty classpath by itself!

in zipkin we have had to do some dodging as well. looking forward to the datastax 4 driver as it no longer conflicts with guava (though it would with netty).

Anyway curious which things bit you and also curious if the cure ends up better than the disease. Appreciate you sharing your experience however this turns out.

[ayansen]

@adriancole yes grpc dependencies did bite us. I had to exclude the guava dependency from the datastax driver https://github.com/ExpediaDotCom/haystack-traces/blob/master/backends/cassandra/pom.xml#L26. Fortunately the grpc version 1.7.1 uses guava version 19.0 which is compatible with the datastax driver. Having said that I'd really like the grpc code to clean up the dependencies.

[codefromthecrypt]

FYI in zipkin we are likely to move to armeria for grpc endpoints