Architecture of Nublado controller¶
The Nublado controller is a FastAPI application with three basic functions:
Prepull images to every Kubernetes node.
Manage user lab pods and their supporting environment.
Manage user file server pods and their supporting environment.
Internally, the code structure uses the handler-services-storage code structure documented in SQR-072. To try to keep the size of the major services, such as the user lab manager, manageable, there are several layers of services that coordinate.
This diagram attempts to provide a guide to the overall code structure. The Kubernetes storage layer is further subdivided into per-resource storage layers, which are not shown in this graph to try to keep the graph more concise.
flowchart LR
subgraph services
prepuller(Prepuller)
image(ImageService)
lab(LabManager)
fileserver(FileserverManager)
image --> docker-source(DockerImageSource)
image --> gar-source(GARImageSource)
prepuller --> prepull-builder(PrepullerBuilder)
prepuller --> image
lab --> image
lab --> lab-builder(LabBuilder)
fileserver --> fileserver-builder(FileserverBuilder)
lab-builder --> volume-builder(VolumeBuilder)
fileserver-builder --> volume-builder
end
subgraph storage
docker(DockerStorageClient)
gar(GARStorageClient)
metadata(MetadataStorage)
fileserver-storage(FilserverStorage)
fileserver-storage --> kubernetes(Kubernetes)
lab-storage(LabStorage)
lab-storage --> kubernetes
node-storage(NodeStorage)
end
docker-source --> docker
gar-source --> gar
fileserver --> fileserver-storage
lab --> metadata
lab --> lab-storage
prepuller --> metadata
prepuller --> kubernetes
image --> node-storage
main(app) --> handlers
handlers --> image
handlers --> prepuller
handlers --> lab
handlers --> fileserver
background(BackgroundTaskManager)
background --> image
background --> prepuller
background --> lab
background --> fileserver
Internal structure¶