Closed stuarteberg closed 2 years ago
With the new neuronjson data type, DVID becomes the central store and clio-store will proxy to DVID. Since DVID can run on larger centralized servers, we can also have in-memory annotations for "head" versions with slower disk access for non-head versions.
As we've discussed elsewhere, it would be nice if we could consolidate our "proofreading" body annotations with the "biology" body annotations, but simply storing all of it in clio-store.
In theory, we could simply require all clients to learn how to talk to clio-store directly (including NeuTu and various batch scripts). But it might be more convenient if DVID could act as a proxy to clio-store. To our clients, the details about how to fetch the right version of the body annotations (for a given UUID) would be hidden. They'd just provide a UUID, a segmentation instance name, and a body ID. They'd get a JSON document in response.