[codex] Add MinerU compose service

docker-compose-full.yml

@@ -2,8 +2,18 @@
 # This Sample File requires NVIDIA GPU for Milvus and VLLM services.
 # Copy `env.docker-compose-full` to `.env` before starting this compose file.
 # You can customize your setup using the Setup Wizard; for detailed instructions, please refer to docs/InteractiveSetup.md
+x-lightrag-no-proxy: &lightrag_no_proxy "localhost,127.0.0.1,::1,0.0.0.0,host.docker.internal,postgres,neo4j,milvus,milvus-etcd,milvus-minio,vllm-embed,vllm-rerank,mineru-api,LightRAG-postgres,LightRAG-neo4j,LightRAG-milvus,LightRAG-milvus-etcd,LightRAG-milvus-minio,LightRAG-vllm-embed,LightRAG-vllm-rerank,LightRAG-mineru-api,10.0.0.0/8,172.16.0.0/12,192.168.0.0/16,sefo.cc,registry.cn-shanghai.aliyuncs.com"
+x-lightrag-disable-proxy: &lightrag_disable_proxy
+  HTTP_PROXY: ""
+  HTTPS_PROXY: ""
+  ALL_PROXY: ""
+  http_proxy: ""
+  https_proxy: ""
+  all_proxy: ""
+
 services:
   lightrag:
+    container_name: LightRAG-servforce
     image: ghcr.io/hkuds/lightrag:latest
     build:
       context: .
@@ -25,10 +35,23 @@ services:
     extra_hosts:
       - "host.docker.internal:host-gateway"
     environment:
+      NO_PROXY: *lightrag_no_proxy
+      no_proxy: *lightrag_no_proxy
       HOST: "0.0.0.0"
       PORT: "9621"
+      WORKSPACE: "servforce"
       EMBEDDING_BINDING_HOST: "http://vllm-embed:8001/v1"
       RERANK_BINDING_HOST: "http://vllm-rerank:8000/rerank"
+      LIGHTRAG_PARSER: "pdf:mineru(local_parse_method=ocr)-iteP,*:native-iteP,*:legacy-R"
+      MINERU_API_MODE: "local"
+      MINERU_LOCAL_ENDPOINT: "http://mineru-api:8000"
+      MINERU_LOCAL_BACKEND: "pipeline"
+      MINERU_LOCAL_PARSE_METHOD: "ocr"
+      MINERU_LOCAL_IMAGE_ANALYSIS: "false"
+      MINERU_LANGUAGE: "ch"
+      MINERU_ENABLE_TABLE: "true"
+      MINERU_ENABLE_FORMULA: "true"
+      MAX_PARALLEL_PARSE_MINERU: "1"
       POSTGRES_HOST: "postgres"
       POSTGRES_PORT: "5432"
       NEO4J_URI: "neo4j://neo4j:7687"
@@ -52,9 +75,80 @@ services:
         condition: service_healthy
       milvus:
         condition: service_healthy
+      mineru-api:
+        condition: service_healthy
+        required: false
+
+  lightrag-sales:
+    container_name: LightRAG-sales
+    image: ghcr.io/hkuds/lightrag:latest
+    build:
+      context: .
+      dockerfile: Dockerfile
+      tags:
+        - ghcr.io/hkuds/lightrag:latest
+    ports:
+      - "${HOST:-0.0.0.0}:${SALES_PORT:-9622}:9621"
+    volumes:
+      - ./data/rag_storage:/app/data/rag_storage
+      - ./data/inputs:/app/data/inputs
+      - ./config.ini:/app/config.ini
+      - ./data/prompts:/app/data/prompts
+      - ./.env:/app/.env
+    deploy:
+      restart_policy:
+        condition: on-failure
+        max_attempts: 10
+    extra_hosts:
+      - "host.docker.internal:host-gateway"
+    environment:
+      NO_PROXY: *lightrag_no_proxy
+      no_proxy: *lightrag_no_proxy
+      HOST: "0.0.0.0"
+      PORT: "9621"
+      WORKSPACE: "sales"
+      EMBEDDING_BINDING_HOST: "http://vllm-embed:8001/v1"
+      RERANK_BINDING_HOST: "http://vllm-rerank:8000/rerank"
+      LIGHTRAG_PARSER: "pdf:mineru(local_parse_method=ocr)-iteP,*:native-iteP,*:legacy-R"
+      MINERU_API_MODE: "local"
+      MINERU_LOCAL_ENDPOINT: "http://mineru-api:8000"
+      MINERU_LOCAL_BACKEND: "pipeline"
+      MINERU_LOCAL_PARSE_METHOD: "ocr"
+      MINERU_LOCAL_IMAGE_ANALYSIS: "false"
+      MINERU_LANGUAGE: "ch"
+      MINERU_ENABLE_TABLE: "true"
+      MINERU_ENABLE_FORMULA: "true"
+      MAX_PARALLEL_PARSE_MINERU: "1"
+      POSTGRES_HOST: "postgres"
+      POSTGRES_PORT: "5432"
+      NEO4J_URI: "neo4j://neo4j:7687"
+      WORKING_DIR: "/app/data/rag_storage"
+      MILVUS_URI: "http://milvus:19530"
+      INPUT_DIR: "/app/data/inputs"
+      PROMPT_DIR: "/app/data/prompts"
+      MEMGRAPH_URI: "bolt://host.docker.internal:7687"
+      REDIS_URI: "redis://host.docker.internal:6379"
+      QDRANT_URL: "http://host.docker.internal:6333"
+      OPENSEARCH_HOSTS: "host.docker.internal:9200"
+      MONGO_URI: "mongodb://root:root@host.docker.internal:27017/"
+    depends_on:
+      vllm-embed:
+        condition: service_healthy
+      vllm-rerank:
+        condition: service_healthy
+      postgres:
+        condition: service_healthy
+      neo4j:
+        condition: service_healthy
+      milvus:
+        condition: service_healthy
+      mineru-api:
+        condition: service_healthy
+        required: false
 
   vllm-embed:
-    image: vllm/vllm-openai:latest
+    container_name: LightRAG-vllm-embed
+    image: ${VLLM_IMAGE:-vllm/vllm-openai:v0.7.3}
     runtime: nvidia
     command: >
       --model ${VLLM_EMBED_MODEL:-BAAI/bge-m3}
@@ -63,6 +157,8 @@ services:
       --api-key ${VLLM_EMBED_API_KEY}
       ${VLLM_EMBED_EXTRA_ARGS:-}
     environment:
+      NO_PROXY: *lightrag_no_proxy
+      no_proxy: *lightrag_no_proxy
       NVIDIA_VISIBLE_DEVICES: ${NVIDIA_VISIBLE_DEVICES:-all}
       NVIDIA_DRIVER_CAPABILITIES: ${NVIDIA_DRIVER_CAPABILITIES:-compute,utility}
     ports:
@@ -88,7 +184,8 @@ services:
     restart: unless-stopped
 
   vllm-rerank:
-    image: vllm/vllm-openai:latest
+    container_name: LightRAG-vllm-rerank
+    image: ${VLLM_IMAGE:-vllm/vllm-openai:v0.7.3}
     runtime: nvidia
     command: >
       --model ${VLLM_RERANK_MODEL:-BAAI/bge-reranker-v2-m3}
@@ -97,6 +194,8 @@ services:
       --api-key ${VLLM_RERANK_API_KEY}
       ${VLLM_RERANK_EXTRA_ARGS:-}
     environment:
+      NO_PROXY: *lightrag_no_proxy
+      no_proxy: *lightrag_no_proxy
       NVIDIA_VISIBLE_DEVICES: ${NVIDIA_VISIBLE_DEVICES:-all}
       NVIDIA_DRIVER_CAPABILITIES: ${NVIDIA_DRIVER_CAPABILITIES:-compute,utility}
     ports:
@@ -121,11 +220,40 @@ services:
               capabilities: [gpu]
     restart: unless-stopped
 
+  mineru-api:
+    container_name: LightRAG-mineru-api
+    image: ${MINERU_IMAGE:-mineru:latest}
+    pull_policy: ${MINERU_PULL_POLICY:-never}
+    entrypoint: mineru-api
+    command: >
+      --host 0.0.0.0
+      --port 8000
+    ports:
+      - "${MINERU_HOST_PORT:-8000}:8000"
+    environment:
+      NO_PROXY: *lightrag_no_proxy
+      no_proxy: *lightrag_no_proxy
+      MINERU_MODEL_SOURCE: ${MINERU_MODEL_SOURCE:-local}
+    volumes:
+      - mineru_cache:/root/.cache
+    ipc: host
+    ulimits:
+      memlock: -1
+      stack: 67108864
+    healthcheck:
+      test: ["CMD-SHELL", "curl -f http://localhost:8000/health || exit 1"]
+      interval: 10s
+      timeout: 5s
+      retries: 60
+      start_period: 30s
+    restart: unless-stopped
+
   postgres:
+    container_name: LightRAG-postgres
     # this image does not support PGGraphStorage
     image: pgvector/pgvector:pg18
-    # ports:
-    #   - "5432:5432"
+    ports:
+      - "${POSTGRES_HOST_PORT:-15432}:5432"
     volumes:
       - postgres_data:/var/lib/postgresql
     healthcheck:
@@ -138,11 +266,14 @@ services:
       start_period: 10s
     restart: unless-stopped
     environment:
+      NO_PROXY: *lightrag_no_proxy
+      no_proxy: *lightrag_no_proxy
       POSTGRES_USER: "rag"
       POSTGRES_PASSWORD: "rag"
       POSTGRES_DB: "rag"
 
   neo4j:
+    container_name: LightRAG-neo4j
     image: neo4j:5-community
     # ports:
     #   - "7474:7474"
@@ -159,15 +290,21 @@ services:
       start_period: 10s
     restart: unless-stopped
     environment:
+      NO_PROXY: *lightrag_no_proxy
+      no_proxy: *lightrag_no_proxy
       NEO4J_AUTH: ${NEO4J_USERNAME:?missing}/${NEO4J_PASSWORD:?missing}
       NEO4J_dbms_default__database: "neo4j"
 
   milvus:
+    container_name: LightRAG-milvus
     image: milvusdb/milvus:v2.6.11-gpu
     command: ["milvus", "run", "standalone"]
     security_opt:
       - seccomp:unconfined
     environment:
+      <<: *lightrag_disable_proxy
+      NO_PROXY: *lightrag_no_proxy
+      no_proxy: *lightrag_no_proxy
       ETCD_ENDPOINTS: milvus-etcd:2379
       MINIO_ADDRESS: milvus-minio:9000
       MINIO_ACCESS_KEY_ID: "${MINIO_ACCESS_KEY_ID:?missing}"
@@ -199,8 +336,11 @@ services:
     restart: unless-stopped
 
   milvus-etcd:
+    container_name: LightRAG-milvus-etcd
     image: quay.io/coreos/etcd:v3.5.25
     environment:
+      NO_PROXY: *lightrag_no_proxy
+      no_proxy: *lightrag_no_proxy
       ETCD_AUTO_COMPACTION_MODE: revision
       ETCD_AUTO_COMPACTION_RETENTION: "1000"
       ETCD_QUOTA_BACKEND_BYTES: "4294967296"
@@ -220,13 +360,19 @@ services:
     restart: unless-stopped
 
   milvus-minio:
+    container_name: LightRAG-milvus-minio
     image: minio/minio:RELEASE.2025-09-07T16-13-09Z
     environment:
+      NO_PROXY: *lightrag_no_proxy
+      no_proxy: *lightrag_no_proxy
       MINIO_ROOT_USER: "${MINIO_ACCESS_KEY_ID:?missing}"
       MINIO_ROOT_PASSWORD: "${MINIO_SECRET_ACCESS_KEY:?missing}"
     volumes:
       - milvus-minio_data:/minio_data
     command: minio server /minio_data --console-address ":9001"
+    ports:
+      - "${MINIO_HOST_PORT:-19000}:9000"
+      - "${MINIO_CONSOLE_HOST_PORT:-19001}:9001"
     healthcheck:
       test: ["CMD", "curl", "-f", "http://localhost:9000/minio/health/live"]
       interval: 30s
@@ -237,6 +383,7 @@ services:
 volumes:
   vllm_embed_cache:
   vllm_rerank_cache:
+  mineru_cache:
   postgres_data:
   neo4j_data:
   milvus_data:

lightrag/kg/milvus_impl.py

@@ -2333,13 +2333,13 @@ async def query(
         # Build search params from index config
         search_params_base = self.index_config.build_search_params()
 
-        # Merge with metric type and radius threshold
+        # Do not pass the threshold as Milvus `radius`: on some Milvus index /
+        # execution backends this switches the request to range search, which
+        # can fail with "RangeSearch not implemented". Keep Milvus on normal
+        # top-k search and apply the threshold consistently in LightRAG.
         search_params = {
             "metric_type": self.index_config.metric_type,
-            "params": {
-                **search_params_base.get("params", {}),
-                "radius": self.cosine_better_than_threshold,
-            },
+            "params": search_params_base.get("params", {}),
         }
 
         results = self._client.search(
@@ -2349,14 +2349,29 @@ async def query(
             output_fields=output_fields,
             search_params=search_params,
         )
+
+        metric_type = str(self.index_config.metric_type).upper()
+        if metric_type == "L2":
+            filtered_results = [
+                dp
+                for dp in results[0]
+                if dp["distance"] <= self.cosine_better_than_threshold
+            ]
+        else:
+            filtered_results = [
+                dp
+                for dp in results[0]
+                if dp["distance"] >= self.cosine_better_than_threshold
+            ]
+
         return [
             {
                 **dp["entity"],
                 "id": dp["id"],
                 "distance": dp["distance"],
                 "created_at": dp.get("created_at"),
             }
-            for dp in results[0]
+            for dp in filtered_results
         ]
 
     @staticmethod