Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

7.2 KB
Raw Blame
//
//  Copyright 2026 The InfiniFlow Authors. All Rights Reserved.
//
//  Licensed under the Apache License, Version 2.0 (the "License");
//  you may not use this file except in compliance with the License.
//  You may obtain a copy of the License at
//
//      http://www.apache.org/licenses/LICENSE-2.0
//
//  Unless required by applicable law or agreed to in writing, software
//  distributed under the License is distributed on an "AS IS" BASIS,
//  WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
//  See the License for the specific language governing permissions and
//  limitations under the License.
//

package engine

import (
	"context"
	"time"

	"ragflow/internal/common"
	"ragflow/internal/engine/types"

	"gorm.io/gorm"
)

// EngineType document engine type
type EngineType string

const (
	EngineElasticsearch EngineType = "elasticsearch"
	EngineInfinity      EngineType = "infinity"
	EngineOceanBase     EngineType = "oceanbase"
	EngineSeekDB        EngineType = "seekdb"
	EngineSereneDB      EngineType = "serenedb"
	EngineVastbase      EngineType = "vastbase"
)

// DocEngine document storage engine interface
type DocEngine interface {
	// Chunk operations
	// CreateChunkStore creates the chunk store for a dataset. language is the
	// dataset language, needed by engines whose fulltext analyzer is fixed when
	// the index is created; the others ignore it.
	CreateChunkStore(ctx context.Context, baseName, datasetID string, vectorSize int, parserID, language string) error
	// InsertChunks writes chunks, creating the chunk store on first write.
	// language is the dataset language, needed there for the same reason as in
	// CreateChunkStore; engines that analyse client-side ignore it.
	InsertChunks(ctx context.Context, chunks []map[string]interface{}, baseName string, datasetID string, language string) ([]string, error)
	UpdateChunks(ctx context.Context, condition map[string]interface{}, newValue map[string]interface{}, baseName string, datasetID string) error
	DeleteChunks(ctx context.Context, condition map[string]interface{}, baseName string, datasetID string) (int64, error)
	Search(ctx context.Context, req *types.SearchRequest) (*types.SearchResult, error)
	GetChunk(ctx context.Context, baseName, chunkID string, datasetIDs []string) (interface{}, error)
	DropChunkStore(ctx context.Context, baseName, datasetID string) error
	ChunkStoreExists(ctx context.Context, baseName, datasetID string) (bool, error)

	// Document metadata operations
	CreateMetadataStore(ctx context.Context, tenantID string) error
	InsertMetadata(ctx context.Context, metadata []map[string]interface{}, tenantID string) ([]string, error)
	UpdateMetadata(ctx context.Context, docID string, datasetID string, metaFields map[string]interface{}, tenantID string) error
	DeleteMetadata(ctx context.Context, condition map[string]interface{}, tenantID string) (int64, error)
	DeleteMetadataKeys(ctx context.Context, docID string, datasetID string, keys []string, tenantID string) error
	DropMetadataStore(ctx context.Context, tenantID string) error
	MetadataStoreExists(ctx context.Context, tenantID string) (bool, error)
	SearchMetadata(ctx context.Context, req *types.SearchMetadataRequest) (*types.SearchMetadataResult, error)

	// Document operations (used by skill indexing)
	IndexDocument(ctx context.Context, indexName, docID string, doc interface{}) error
	DeleteDocument(ctx context.Context, indexName, docID string) error
	BulkIndex(ctx context.Context, indexName string, docs []interface{}) (interface{}, error)

	// Utility functions for search result processing
	GetFields(chunks []map[string]interface{}, fields []string) map[string]map[string]interface{}
	GetAggregation(chunks []map[string]interface{}, fieldName string) []map[string]interface{}
	GetHighlight(chunks []map[string]interface{}, keywords []string, fieldName string) map[string]string

	// RunSQL runs a SQL query
	RunSQL(ctx context.Context, sqlText string, format string) ([]map[string]interface{}, error)

	GetChunkIDs(chunks []map[string]interface{}) []string
	KNNScores(ctx context.Context, chunks []map[string]interface{}, queryVector []float64, topK int) (map[string]interface{}, error)
	GetScores(searchResult map[string]interface{}) map[string]float64

	// Ping check the engine is alive
	Ping(ctx context.Context) error
	Close() error

	// GetType returns the engine type
	GetType() string

	// SupportsPageRank reports whether the engine supports dataset-level pagerank.
	SupportsPageRank() bool

	// FilterDocIdsByMetaPushdown runs a metadata filter directly against
	// the doc metadata index, returning matching doc IDs or nil if push-down
	// is not supported (caller should fall back to in-memory filtering).
	// conditions is a list of filter objects with keys: key, op, value
	FilterDocIdsByMetaPushdown(ctx context.Context, sqlDB *gorm.DB, kbIDs []string, conditions []map[string]interface{}, logic string) []string
}

// Type returns the engine type (helper method for runtime type checking)
// This is a workaround since we can't import elasticsearch or infinity packages directly
func Type(docEngine DocEngine) EngineType {
	if docEngine == nil {
		return EngineType("unknown")
	}
	return EngineType(docEngine.GetType())
}

// IsOceanBaseFamily reports whether a configured engine uses the shared
// OceanBase/SeekDB SQL implementation.
func IsOceanBaseFamily(engineName string) bool {
	return engineName == string(EngineOceanBase) || engineName == string(EngineSeekDB)
}

// IsVastbase reports whether a configured engine is the Vastbase G100
// implementation. Vastbase is PostgreSQL-wire based and shares several
// service-layer behaviors with the OceanBase family (local rerank, chunk
// vectors in select fields, shared-table memory lifecycle) but is a separate
// SQL implementation, so it keeps its own predicate instead of joining
// IsOceanBaseFamily.
func IsVastbase(engineName string) bool {
	return engineName == string(EngineVastbase)
}

type MessageQueue interface {
	Init() error
	Type() string
	InitConsumer(subject string) error
	PublishTask(subject string, payload []byte) error
	PullMessages(ctx context.Context, messageCount int) ([]common.TaskHandle, error)
	PullMessage(ctx context.Context) (common.TaskHandle, error)
	ListMessages(messageType string, pending bool) ([]map[string]string, error)
	ShowMessageQueue() (map[string]string, error)
	CheckStatus() string

	// dataset-level compile consumer (§11) surface.
	InitKnowledgeCompileStream() error
	InitKnowledgeCompileConsumer() error
	PublishKnowledgeCompile(subject string, payload []byte) error
	FetchKnowledgeCompileMessages(n int) ([]common.RawMessage, error)
	InitKnowledgeCompileLeases() error
	AcquireKnowledgeCompileLease(key, holder string, ttl time.Duration) (uint64, bool, error)
	HeartbeatKnowledgeCompileLease(key, holder string, ttl time.Duration, revision uint64) (uint64, bool, error)
	ReleaseKnowledgeCompileLease(key, holder string, revision uint64) error

	// SubscribeNotify returns a channel of dataset ids pushed by
	// PublishKnowledgeCompile on the notify.kc.workers subject (Option E
	// §11.4: NATS as a wake-up, MySQL as the scheduling system of record).
	// Implementations return (nil, nil) when push wake-up is unavailable;
	// callers must fall back to periodic polling in that case.
	SubscribeNotify(ctx context.Context) (<-chan string, error)
}