The managed cloud provider is a provider extension to the DockerParallel package. It aims to provide container management tools for the non-kubernetes cloud provider. The cloud provider only needs to define functions to run and stop the worker. The managed cloud provider can automatically scale the number of workers up or down according to the user's request. Below shows the structure of the managed cloud provider
where CloudProvider is the base class defined in DockerParallel package. CloudProvider dose nothing but define a few generics(the virtual class in C++ language) .ManagedCloudProvider inherits from CloudProvider, it takes over the control of the worker container by defining the method for setDockerWorkerNumber and getDockerWorkerNumbers(The red font). The provider who inherits from ManagedCloudProvider does not need to manage the workers itself, it only needs to define the method for the generics exposed by ManagedCloudProvider, namely
The name should be self-explained. Please note that only runDockerWorkerContainers, getDockerWorkerStatus and killDockerWorkerContainers are required for making the provider works. The rest are optional and have the default methods(based on getDockerWorkerStatus).
runDockerWorkerContainers should deploy the worker on the cloud. The generic for runDockerWorkerContainers is
getGeneric("runDockerWorkerContainers")
#> nonstandardGenericFunction for "runDockerWorkerContainers" defined from package "ManagedCloudProvider"
#>
#> function (provider, cluster, container, hardware, workerNumber,
#> verbose)
#> {
#> standardGeneric("runDockerWorkerContainers")
#> }
#> <bytecode: 0x00000000160c5560>
#> <environment: 0x00000000160d6b18>
#> Methods may be defined for arguments: provider
#> Use showMethods(runDockerWorkerContainers) for currently available ones.
where provider is the cloud provider object and cluster is the DockerCluster object, container specifies which container should be used for the worker, hardware indicates the hardware requirement for each worker, workerNumber is the number of the workers that should be deployed. verbose is the verbose level. The return value of the function should be a character vector of length workerNumber with each element corresponding to a handle of a worker. The handle should be used to manage the lifecycle of the workers. It is possible to return the same handle for multiple workers if they share the same lifecycle(e.g. multiple workers share the same container).
getDockerWorkerStatus is for checking the status of the workers, a worker can have three possible status, namely initializing, running and stopped. The generic for getDockerWorkerStatus is
getGeneric("getDockerWorkerStatus")
#> nonstandardGenericFunction for "getDockerWorkerStatus" defined from package "ManagedCloudProvider"
#>
#> function (provider, cluster, workerHandles, verbose)
#> {
#> standardGeneric("getDockerWorkerStatus")
#> }
#> <bytecode: 0x00000000166c8940>
#> <environment: 0x00000000166d0ed8>
#> Methods may be defined for arguments: provider
#> Use showMethods(getDockerWorkerStatus) for currently available ones.
where workerHandles is a vector of handles returned by runDockerWorkerContainers, the elements of workerHandles are unique. The return value should be a character vector indicating the status of each worker.
The generics IsDockerWorkerInitializing, IsDockerWorkerRunning and IsDockerWorkerStopped check whether the worker is in a specific status. Once you have getDockerWorkerStatus defined these methods are automatically available. Therefore, you do not have to define them unless you have any faster implementation.
killDockerWorkerContainers will kill the workers given the worker handles. If a handle is shared by multiple workers, all workers with the same handle should be killed. The generic is
getGeneric("killDockerWorkerContainers")
#> nonstandardGenericFunction for "killDockerWorkerContainers" defined from package "ManagedCloudProvider"
#>
#> function (provider, cluster, workerHandles, verbose)
#> {
#> standardGeneric("killDockerWorkerContainers")
#> }
#> <bytecode: 0x00000000161c14b0>
#> <environment: 0x00000000161e69d8>
#> Methods may be defined for arguments: provider
#> Use showMethods(killDockerWorkerContainers) for currently available ones.
The return value is a logical vector indicating whether the killing process is successful. Please note that the method should accept the invalid handle. That is, if the worker has already been killed or the handle does not exist, the method should not raise any error and the killing action should be denoted as successful.