dut3062796s / containerfs

The container-native distributed file system

ContainerFS

Overview

ContainerFS is a container-native distributed filesystem as a unified platform for unstructured data storage.

• unlimited small files and large files.

• append-only or random writes

• scale-out metadata management

• strong consistency

• multi-tenancy: millions of filesystem volumes

• POSIX-compatible

Containerfs has been built and deployed in production since 2013.

Concepts

volume

inode

directory

blob

extent

Architecture

CFS consists of several components:

• the cluster master. single raft replication, managing volumes, metanodes, datanodes, meta-partitions and data-partitions

• metanode. multi-raft replication, a meta partition (inode range) per replication state machine

• datanode. de-clustering of data partitions, two storage engines - Blob Store (BS) and Extent Store (ES), optimized for small and large files respectively.

• client interfaces: FUSE, Java SDK, Go SDK, Linux kernel

replication

master: single-raft

metadata partition: multi-raft

data partition: chained append-only replication for append operations; multi-raft for extent updates

Usage

• mounted by containers for decoupling storage from compute

integrated with Kubernetes to run various workloads from application microservices to complex databases e.g., HBase, MyRocks, ElasticSearch, and ChuBao DB

• Image Store

based on the namespace-less key-file interface, nginx integration

• Object Store

integration with minio

Deployment

rack

set