...
- More efficient reading and writing of persistent data when using the standalone version for it don't need network communication.
- Reduce the impact of its dependency component changes when components are changed. For example, when minio crashes, the system is unavailable.
- Dependency injection can be done with local storage, removing the dependency of minio in unittest on minio.
At the same time, in order to preserve the selectivity of users, minio can also be used as a storage engine for the standalone version. We are going to make the storage engine configurable.
...
For solving Problem 3, We redefine the interface implemented by minio. All file storage needs to implement this interface.
Code Block title ChunkManager interface linenumbers true type ChunkManager interface { GetPath(filePath string) (string, error) GetSize(filePath string) (int64, error) Write(filePath string, content []byte) error MultiWrite(contents map[string][]byte) error Exist(filePath string) bool Read(filePath string) ([]byte, error) MultiRead(filePaths []string) ([][]byte, error) ReadWithPrefix(prefix string) ([]string, [][]byte, error) ReadAt(filePath string, off int64, length int64) (p []byte, err error) Remove(filePath string) error MultiRemove(filePaths []string) error RemoveWithPrefix(prefix string) error }
For this interface we will have three implementations, LocalChunkManager, MinioChunkmanager and VectorChunkManager.
VectorChunkManager is an optimized management class for vector reading under distributed milvus. It will use minio as a remote chunkManager storagechunkManager and local file system as a vector cachevectorChunkStorage. When reading a file, it will be downloaded from the minio to the local, and then the relevant data will be read from the local. In the standalone version, we will replace minioChunkManager with localChunkManager as the implementation of remoteChunkManagerstorageChunkManager.- ChunkManagerFactory
For Problem 1, a chunkManagerFactory similar to msgStream. Factory is added to generate chunkManagers with different configurations.
Code Block title ChunkManager Factory linenumbers true type ChunkManagerFactory struct { localStorageChunkStorage string remoteStorageVectorCacheStorage string } func NewChunkManagerFactory(localChunkStorage, remoteVectorCacheStorage string) *ChunkManagerFactory{} func (cmf *ChunkManagerFactory) NewLocalChunkManager(opts ...storage.Options){} func (cmf *ChunkManagerFactory) NewRemoteChunkManagerNewChunkStorage(opts ...storage.Options){ switch (cmf.remoteStorageChunkStorage){ case "s3": case "minio": ... } } func (cmf *ChunkManagerFactory) NewVectorCacheManager(opts ...storage.Options){}
Options is needed when generating a new chunkManager. The Options maybe like this.
Code Block title ChunkManager Config linenumbers true type config struct { address string bucketName string accessKeyID string secretAccessKeyID string useSSL bool createBucket bool rootPath string } type Option func(*config) func Address(addr string) Option { return func(c *config) { c.address = addr } }
This structure will have some redundancy. For example, local storage will not require parameters such as address and bucketname. but will be easier to reuse.
Use a more generic factory instead of the existing msgFactory to build nodes.
Code Block title Interface extened factory := newMsgFactory(localMsg) rc, err := components.NewRootCoord(ctx, factory) ↓↓↓↓↓↓↓↓↓↓↓↓↓ factory := newFactory(localMsg) rc, err := components.NewRootCoord(ctx, factory)
And the factory will be like this
Code Block title Factory Struct linenumbers true type Factory struct { msgF msgstream.Factory storageF storage.ChunkManagerFactory } func newDefaultFactory(opts ...Option) *Factory { c := newDefaultConfig() for opt := range opts { opt(c) } return &Factory{ MsgFactory: msgstream.NewFactory(c.msgstream), storageF: storage.NewChunkManagerFactory(c.localStoragevectorCacheStorage, c.remoteStoragechunkStorage), } } type config struct { msgStream string localStoragevectorCacheStorage string remoteStoragechunkStorage string } func newDefaultConfig() *config{} type Option func(*config) func localStoragevectorCacheStorage(localStoragevectorCacheStorage string) Option { return func(c *config) { c.localStoragevectorCacheStorage = localStoragevectorCachestorage } }
deploy in milvus.yaml will set the default storage engine for different deploy mode. The localStorage vectorCacheStorage and remoteStorage chunkStorage will be used to set what to use as the storage engine.
Code Block language yml title milvus.yaml linenumbers true deploy: standalone: localStoragechunkStorage: "local" remoteStoragevectorCacheStorage: "local" distributed: localStoragechunkStorage: "localminio" remoteStoragevectorCacheStorage: "miniolocal"
In standalone mode, remoteStorage chunkStorage using local is a more efficient choice. Of course you can also choose minio.
However, if the distributed version chooses local storage, because the nodes are located on different machines, the data stored locally is inconsistent. Therefore, careful consideration should be given when choosing a storage engine.
...