Skip to content

Latest commit

 

History

History
111 lines (89 loc) · 3.57 KB

design.md

File metadata and controls

111 lines (89 loc) · 3.57 KB
Raw

Containerfs Design

Architecture

See the later sections for more details of each Containerfs component. image

Volume

Volume is a Containerfs instance. One Containerfs cluster can host millions of volumes. A volume has one matadata table and unlimited number of block groups

Inode

Inode is a data structure recording the file or directory attributes, indexed with a unique 64bit integer.

Matadata Table

A metadata table is a sorted key-value map from "parent inodeID + name" to inode attributes. One volume is associated with a metadata table. Metadata tables are replicated via Raft, and not sharded by design.

Block Group

Fix-sized replicated storage unit of file extents

MetaNode

Hosts one or multiple metadata tables

DataNode

Hosts one or multiple block groups

Volume Manager

VolMgr holds all the cluster-level metadata, like the volume space quota, nodes status.

Client

FUSE
Linux kernel

Communication

Core Functions

Volume create

open I/O flow

write I/O flow

read I/O flow

Struct

volume namespace

 type KvStateMachine struct {
 	id      uint64
 	applied uint64
 	raft    *raft.RaftServer
 	DentryLocker sync.RWMutex
 	dentryData   map[string][]byte
 	inodeLocker  sync.RWMutex
 	inodeData    map[string][]byte
 	BlockGroupLocker sync.RWMutex
 	blockGroupData   map[string][]byte
 	chunkID uint64
 	inodeID uint64
 }

dentry

 type Dirent struct {
 	InodeType bool   `protobuf:"varint,1,opt,name=InodeType" json:"InodeType,omitempty"`
 	Inode     uint64 `protobuf:"varint,2,opt,name=Inode" json:"Inode,omitempty"`
 }

inode

 type InodeInfo struct {
 	ModifiTime int64        `protobuf:"varint,1,opt,name=ModifiTime" json:"ModifiTime,omitempty"`
 	AccessTime int64        `protobuf:"varint,2,opt,name=AccessTime" json:"AccessTime,omitempty"`
 	Link       uint32       `protobuf:"varint,3,opt,name=Link" json:"Link,omitempty"`
 	FileSize   int64        `protobuf:"varint,4,opt,name=FileSize" json:"FileSize,omitempty"`
 	Chunks     []*ChunkInfo `protobuf:"bytes,5,rep,name=Chunks" json:"Chunks,omitempty"`
 }

block group

 type BlockGroup struct {
 	BlockGroupID uint32       `protobuf:"varint,1,opt,name=BlockGroupID" json:"BlockGroupID,omitempty"`
 	FreeSize     int64        `protobuf:"varint,2,opt,name=FreeSize" json:"FreeSize,omitempty"`
 	Status       int32        `protobuf:"varint,3,opt,name=Status" json:"Status,omitempty"`
 	BlockInfos   []*BlockInfo `protobuf:"bytes,4,rep,name=BlockInfos" json:"BlockInfos,omitempty"`
 }
 
 type BlockInfo struct {
 	BlockID      uint32 `protobuf:"varint,1,opt,name=BlockID" json:"BlockID,omitempty"`
 	DataNodeIP   int32  `protobuf:"varint,2,opt,name=DataNodeIP" json:"DataNodeIP,omitempty"`
 	DataNodePort int32  `protobuf:"varint,3,opt,name=DataNodePort" json:"DataNodePort,omitempty"`
 	Status       int32  `protobuf:"varint,4,opt,name=Status" json:"Status,omitempty"`
 }

chunk

 type ChunkInfo struct {
 	ChunkID      uint64  `protobuf:"varint,1,opt,name=ChunkID" json:"ChunkID,omitempty"`
 	ChunkSize    int32   `protobuf:"varint,2,opt,name=ChunkSize" json:"ChunkSize,omitempty"`
 	BlockGroupID uint32  `protobuf:"varint,3,opt,name=BlockGroupID" json:"BlockGroupID,omitempty"`
 	Status       []int32 `protobuf:"varint,4,rep,packed,name=Status" json:"Status,omitempty"`
 }