Glance is designed to be as adaptable as possible for various back-end storage and registry database solutions. There is a main Glance API server (the glance-api program) that serves as the communications hub between various client programs, the registry of image metadata, and the storage systems that actually contain the virtual machine image data.
From a birdseye perspective, one can visualize the Glance architectural model like so:
digraph birdseye { node [fontsize=10 fontname="Monospace"] a [label="Client A"] b [label="Client B"] c [label="Client C"] d [label="Glance API Server"] e [label="Registry Server"] f [label="Store Adapter"] g [label="S3 Store"] h [label="Swift Store"] i [label="Filesystem Store"] j [label="HTTP Store"] a -> d [dir=both] b -> d [dir=both] c -> d [dir=both] d -> e [dir=both] d -> f [dir=both] f -> g [dir=both] f -> h [dir=both] f -> i [dir=both] f -> j [dir=both] }
A registry server is a service that publishes image metadata for internal consumption by the Glance API server. The Glance registry server uses a SQL database for its metdata storage.
A store is a Python class that inherits from glance.store.Backend and conforms to that class’ API for reading, writing, and deleting virtual machine image data.
Glance currently ships with stores for S3, Swift, RBD, a simple filesystem store, and a read-only HTTP(S) store.
Implementors are encouraged to create stores for other backends, including other distributed storage systems like Sheepdog.