github.com/lalkh/containerd@v1.4.3/design/architecture.md (about) 1 # Architecture 2 3 To ensure clean separation of concerns, we have organized the units of 4 containerd's behavior into _components_. _Components_ are roughly organized 5 into _subsystems_. _Components_ that bridge subsystems may be referred to as 6 _modules_. _Modules_ typically provide cross-cutting functionality, such as 7 persistent storage or event distribution. Understanding these _components_ and 8 their relationships is key to modifying and extending the system. 9 10 This document will cover very high-level interaction. For details on each 11 module, please see the relevant design document. 12 13 The main goal of this architecture is to coordinate the creation and execution 14 of _bundles_. _Bundles_ contain configuration, metadata and root filesystem 15 data and are consumed by the _runtime_. A _bundle_ is the on-disk 16 representation of a runtime container. _Bundles_ are mutable and can be passed 17 to other systems for modification or packed up and distributed. In practice, it 18 is simply a directory on the filesystem. 19 20 ![Architecture](architecture.png) 21 22 Note that while these architectural ideas are important to understand the 23 system, code layout may not reflect the exact architecture. These ideas should 24 be used as a guide for placing functionality and behavior and understanding the 25 thought behind the design. 26 27 ## Subsystems 28 29 External users interact with services, made available via a GRPC API. 30 31 - __*Bundle*__: The bundle service allows the user to extract and pack bundles 32 from disk images. 33 - __*Runtime*__: The runtime service supports the execution of _bundles_, 34 including the creation of runtime containers. 35 36 Typically, each subsystem will have one or more related _controller_ components 37 that implement the behavior of the _subsystem_. The behavior of the _subsystem_ 38 may be exported for access via corresponding _services_. 39 40 ## Modules 41 42 In addition to the subsystems, we have several components that may cross 43 subsystem boundaries, referenced to as components. We have the following 44 components: 45 46 - __*Executor*__: The executor implements the actual container runtime. 47 - __*Supervisor*__: The supervisor monitors and reports container state. 48 - __*Metadata*__: Stores metadata in a graph database. Use to store any 49 persistent references to images and bundles. Data entered into the 50 database will have schemas coordinated between components to provide access 51 to arbitrary data. Other functionality includes hooks for garbage collection 52 of on-disk resources. 53 - __*Content*__: Provides access to content addressable storage. All immutable 54 content will be stored here, keyed by content hash. 55 - __*Snapshot*__: Manages filesystem snapshots for container images. This is 56 analogous to the graphdriver in Docker today. Layers are unpacked into 57 snapshots. 58 - __*Events*__: Supports the collection and consumption of events for providing 59 consistent, event driven behavior and auditing. Events may be replayed to 60 various _modules_ 61 - __*Metrics*__: Each components will export several metrics, accessible via 62 the metrics API. (We may want to promote this to a subsystem. 63 64 ## Client-side components 65 66 Some components are implemented on the client side for flexibility: 67 68 - __*Distribution*__: Functions for pulling and pushing images 69 70 ## Data Flow 71 72 As discussed above, the concept of a _bundle_ is central to containerd. Below 73 is a diagram illustrating the data flow for bundle creation. 74 75 ![data-flow](data-flow.png) 76 77 Let's take pulling an image as a demonstrated example: 78 79 1. Instruct the Distribution layer to pull a particular image. The distribution 80 layer places the image content into the _content store_. The image name and 81 root manifest pointers are registered with the metadata store. 82 2. Once the image is pulled, the user can instruct the bundle controller to 83 unpack the image into a bundle. Consuming from the content store, _layers_ 84 from the image are unpacked into the _snapshot_ component. 85 3. When the snapshot for the rootfs of a container is ready, the _bundle 86 controller_ can use the image manifest and config to prepare the execution 87 configuration. Part of this is entering mounts into the execution config 88 from the _snapshot_ module. 89 4. The prepared bundle is then passed off to the _runtime_ subsystem for 90 execution. It reads the bundle configuration to create a running container. 91