gourao
commented
8 years ago To be blunt, what the project is trying to accomplish is establish a storage protocol between the scheduling software and the underlying infrastructure. We want storage to be allocated in a container granular manner in a multi host environment.
To that end, there are two entities involved: 1) The actual container runtime software (Docker, Rkt) etc. 2) A scheduler (k8s, mesos) etc.
And we want a spec that allows for any of the above combinations to work with a backend that can basically respect the protocols of the spec.
To give you an example, if I want to deploy postgress in a container in a multi node environment, and my choices are mesos/k8s/swarm for scheduling, Docker, rkt etc for the container format, it is unclear how to allocate infrastructure resources in a generic way.
Hopefully this project can solve that.
I understand that y'all are trying to get something working but it is very unclear what this project is trying to accomplish from reading the spec. I see the swagger API but it is very unclear what the semantics are.
Perhaps a few stories on how the spec might work in practice could be illustrative. For example if I call this method https://github.com/libopenstorage/specs/blob/master/api/openstorage.yaml#L97 what changes on the host? What directory will be mounted? Does that directory have to exist before hand? etc.