The goal of this document is to provide a quick-start guide on the design of lmctfy in order to make contributing to the codebase easier.
The repository has a slew of directories and a handful of files at the top level. Most of the top level files are documentation, licensing, and the build/test rules. The other directories are described below.
The core of lmctfy can be found in /lmctfy and the public API can be found in /include.
/lmctfy: The implementation of the public API and the core abstractions (described below)./lmctfy/cli: The implementation of the lmctfy command line tool./lmctfy/controllers: Controllers for all resources lmctfy supports./lmctfy/resources: Resource handlers for all resources lmctfy supports.
This comprises the majority of the top-level directories. These are base and generic C++ libraries and utilities.
/base: Base libraries./file: File-handling utilities./gmock: The gMock and gTest frameworks./strings: String-handling utilities./thread: Wrapper for thread operations./util: Generic utilities.
The lmctfy code abstracts system interactions with the system API found in /system_api. This abstraction lets us mock out the system for unit and integration testing.
At the core of lmctfy is the public C++ library API (the CLI is a thin wrapper around this). The public API has the ContainerApi which functions as a factory of Container objects (both of these are in lmctfy.h). These make heavy use of different specifications found in lmctfy.proto. All of these specifications are split into resources that lmctfy supports. The aim is for these to not change significantly over time and to be high level enough to adapt to underlying kernel changes.
The C++ library is implemented by some logic wrapping a set of handlers. The most prevalent of these are the resource handlers which each implements the support and policies of a single resource (their interface is in resource_handler.h). One other top-level handler is the tasks handler which manages tracking of tasks (interface at tasks_handler.h).
Each resource handler defines its own specification and implements the isolation that specification mentions. It does this through a series of controllers which interact with the kernel. The ResouceHandler itself decides how it will use the controllers to implement the desired isolation (i.e.: the CpuResourceHandler creates a /batch top-level cpu cgroup (through the CpuController) to isolate all batch workload).
The controllers are what actually interacts with the kernel to implement the desired isolation. They provide a slightly higher API to the resource handlers in order to encapsulate changes in the kernel. The granularity of a controller is a cgroup or a namespace (e.g.: CpuController for the cpu hierarchy and MemoryController for the memory hierarchy).
lmctfy requires a canonical tasks hierarchy it uses to keep track of groups of tasks. The TasksHandler contains this logic. The current implementation is the CgroupTasksHandler which uses the freezer cgroup hierarchy to track tasks.
Feel free to e-mail the project's mailing list with any questions: [email protected].