De ChromaDB a Qdrant: migrar tu RAG a produccion con Docker

┌─────────────┐     ┌──────────────┐     ┌─────────────────────┐
│  data/      │────▶│  ingest.py   │────▶│  Qdrant             │
│  (docs)     │     │  (chunking + │     │  (Docker:6333/6334) │
└─────────────┘     │  embedding)  │     │  collection:        │
                    └──────────────┘     │  "ops-docs"         │
                                         └──────────┬──────────┘
                                                    │
┌─────────────┐     ┌──────────────┐                │
│  User query │────▶│   rag.py     │◀───────────────┘
│             │     │  (search +   │     cosine similarity
└─────────────┘     │   LLM call)  │────▶ Grok API (xAI)
                    └──────────────┘

services:
  qdrant:
    image: qdrant/qdrant:latest
    container_name: qdrant
    ports:
      - "6333:6333"   # REST API + Web UI
      - "6334:6334"   # gRPC (mas rapido para produccion)
    volumes:
      - ./qdrant_data:/qdrant/storage
    environment:
      - QDRANT__SERVICE__GRPC_PORT=6334
      - QDRANT__LOG_LEVEL=INFO
    restart: unless-stopped
    # Opcional: limitar recursos
    deploy:
      resources:
        limits:
          memory: 2G

docker compose up -d
# Verificar que arranca
curl http://localhost:6333/healthz
# Respuesta: {"title":"qdrant - vector search engine","version":"1.x.x"}

services:
-  chromadb:
-    image: chromadb/chroma:latest
-    ports:
-      - "8001:8000"
-    volumes:
-      - ./chroma_data:/chroma/chroma
-    environment:
-      - ANONYMIZED_TELEMETRY=False
+  qdrant:
+    image: qdrant/qdrant:latest
+    ports:
+      - "6333:6333"
+      - "6334:6334"
+    volumes:
+      - ./qdrant_data:/qdrant/storage
+    restart: unless-stopped

# requirements.txt
  langchain>=0.3.25
  langchain-huggingface>=1.0.0
  langchain-openai>=0.3.12
  langchain-text-splitters>=0.3.0
- chromadb>=1.0.0
+ qdrant-client>=1.9.0
+ langchain-qdrant>=0.2.0
  sentence-transformers>=4.0.0
  python-dotenv>=1.0.0
  rich>=14.0.0

pip install qdrant-client langchain-qdrant

# Vector DB
VECTOR_DB=qdrant
QDRANT_HOST=localhost
QDRANT_PORT=6333
QDRANT_COLLECTION=ops-docs

# Embedding (sin cambios - sigue siendo local)
EMBEDDING_MODEL=all-MiniLM-L6-v2

# LLM (sin cambios)
LLM_PROVIDER=grok
GROK_API_KEY=your-key-here
GROK_MODEL=grok-3-mini

import os
from dotenv import load_dotenv

load_dotenv()

class Config:
    # Qdrant
    QDRANT_HOST = os.getenv("QDRANT_HOST", "localhost")
    QDRANT_PORT = int(os.getenv("QDRANT_PORT", 6333))
    QDRANT_COLLECTION = os.getenv("QDRANT_COLLECTION", "ops-docs")

    # Embedding
    EMBEDDING_MODEL = os.getenv("EMBEDDING_MODEL", "all-MiniLM-L6-v2")

    # LLM
    LLM_PROVIDER = os.getenv("LLM_PROVIDER", "grok")
    GROK_API_KEY = os.getenv("GROK_API_KEY")
    GROK_BASE_URL = os.getenv("GROK_BASE_URL", "https://api.x.ai/v1")
    GROK_MODEL = os.getenv("GROK_MODEL", "grok-3-mini")

#!/usr/bin/env python3
"""ingest.py - Ingestion de documentos en Qdrant"""

import sys
import uuid
from pathlib import Path

from langchain_huggingface import HuggingFaceEmbeddings
from langchain_text_splitters import RecursiveCharacterTextSplitter
from qdrant_client import QdrantClient
from qdrant_client.models import (
    Distance,
    PointStruct,
    VectorParams,
)
from rich.console import Console
from rich.progress import track

from config import Config

console = Console()

# Extensiones soportadas
SUPPORTED_EXTENSIONS = {".md", ".yaml", ".yml", ".conf", ".sh", ".txt", ".json", ".toml"}


def load_documents(data_dir: str) -> list[dict]:
    """Carga documentos del directorio."""
    docs = []
    data_path = Path(data_dir)

    for filepath in sorted(data_path.rglob("*")):
        if filepath.suffix.lower() in SUPPORTED_EXTENSIONS and filepath.is_file():
            try:
                content = filepath.read_text(encoding="utf-8", errors="ignore")
                if content.strip():
                    docs.append({
                        "content": content,
                        "metadata": {
                            "source": str(filepath.relative_to(data_path)),
                            "filename": filepath.name,
                            "extension": filepath.suffix.lower(),
                        },
                    })
            except Exception as e:
                console.print(f"[yellow]Skip {filepath}: {e}[/]")

    return docs


def chunk_documents(docs: list[dict]) -> list[dict]:
    """Divide documentos en chunks."""
    splitter = RecursiveCharacterTextSplitter(
        chunk_size=1000,
        chunk_overlap=200,
        separators=["\n## ", "\n### ", "\n\n", "\n", " "],
    )

    chunks = []
    for doc in docs:
        splits = splitter.split_text(doc["content"])
        for i, text in enumerate(splits):
            chunks.append({
                "text": text,
                "metadata": {**doc["metadata"], "chunk_index": i},
            })

    return chunks


def main(data_dir: str = "data"):
    console.print(f"\n[bold blue]Ingestion Qdrant[/] - Directorio: {data_dir}\n")

    # 1. Cargar documentos
    docs = load_documents(data_dir)
    console.print(f"[green]Documentos cargados:[/] {len(docs)}")

    if not docs:
        console.print("[red]No se encontraron documentos.[/]")
        return

    # 2. Chunking
    chunks = chunk_documents(docs)
    console.print(f"[green]Chunks generados:[/] {len(chunks)}")

    # 3. Generar embeddings
    console.print(f"[yellow]Generando embeddings con {Config.EMBEDDING_MODEL}...[/]")
    embeddings_model = HuggingFaceEmbeddings(model_name=Config.EMBEDDING_MODEL)
    texts = [chunk["text"] for chunk in chunks]
    embeddings = embeddings_model.embed_documents(texts)
    vector_size = len(embeddings[0])
    console.print(f"[green]Embeddings generados:[/] {len(embeddings)} (dim={vector_size})")

    # 4. Conectar a Qdrant
    client = QdrantClient(host=Config.QDRANT_HOST, port=Config.QDRANT_PORT)

    # Recrear coleccion (full re-ingestion)
    client.recreate_collection(
        collection_name=Config.QDRANT_COLLECTION,
        vectors_config=VectorParams(
            size=vector_size,
            distance=Distance.COSINE,
        ),
    )
    console.print(f"[green]Coleccion '{Config.QDRANT_COLLECTION}' creada (dim={vector_size}, cosine)[/]")

    # 5. Insertar en batches
    batch_size = 100
    points = []
    for i, (chunk, embedding) in enumerate(zip(chunks, embeddings)):
        points.append(
            PointStruct(
                id=str(uuid.uuid4()),
                vector=embedding,
                payload={
                    "text": chunk["text"],
                    **chunk["metadata"],
                },
            )
        )

    for i in track(range(0, len(points), batch_size), description="Insertando..."):
        batch = points[i : i + batch_size]
        client.upsert(collection_name=Config.QDRANT_COLLECTION, points=batch)

    console.print(f"\n[bold green]Ingestion completada: {len(points)} chunks en '{Config.QDRANT_COLLECTION}'[/]\n")


if __name__ == "__main__":
    data_dir = sys.argv[1] if len(sys.argv) > 1 else "data"
    main(data_dir)

#!/usr/bin/env python3
"""rag.py - RAG interactivo con Qdrant + Grok"""

from langchain_huggingface import HuggingFaceEmbeddings
from langchain_openai import ChatOpenAI
from qdrant_client import QdrantClient
from rich.console import Console
from rich.markdown import Markdown

from config import Config

console = Console()


def get_llm() -> ChatOpenAI:
    """Configura el LLM."""
    return ChatOpenAI(
        model=Config.GROK_MODEL,
        api_key=Config.GROK_API_KEY,
        base_url=Config.GROK_BASE_URL,
        temperature=0.3,
    )


def search_context(client: QdrantClient, query_embedding: list[float], top_k: int = 5) -> str:
    """Busca contexto relevante en Qdrant."""
    results = client.query_points(
        collection_name=Config.QDRANT_COLLECTION,
        query=query_embedding,
        limit=top_k,
        with_payload=True,
    )

    if not results.points:
        return ""

    context_parts = []
    for point in results.points:
        source = point.payload.get("source", "unknown")
        text = point.payload.get("text", "")
        score = point.score
        context_parts.append(f"[{source}] (score: {score:.3f})\n{text}")

    return "\n\n---\n\n".join(context_parts)


def main():
    console.print("\n[bold blue]RAG Interactivo[/] (Qdrant + Grok)")
    console.print("[dim]Escribe 'exit' para salir[/]\n")

    # Inicializar componentes
    embeddings_model = HuggingFaceEmbeddings(model_name=Config.EMBEDDING_MODEL)
    client = QdrantClient(host=Config.QDRANT_HOST, port=Config.QDRANT_PORT)
    llm = get_llm()

    # Verificar coleccion
    info = client.get_collection(Config.QDRANT_COLLECTION)
    console.print(f"[green]Coleccion:[/] {Config.QDRANT_COLLECTION} ({info.points_count} puntos)\n")

    while True:
        query = console.input("[bold cyan]Pregunta>[/] ").strip()
        if query.lower() in ("exit", "quit", "q"):
            break
        if not query:
            continue

        # 1. Embedding de la query
        query_embedding = embeddings_model.embed_query(query)

        # 2. Buscar contexto
        context = search_context(client, query_embedding)

        if not context:
            console.print("[yellow]No se encontro contexto relevante.[/]\n")
            continue

        # 3. Construir prompt con contexto
        prompt = f"""Responde la siguiente pregunta basandote UNICAMENTE en el contexto proporcionado.
Si el contexto no contiene informacion suficiente, dilo explicitamente.

## Contexto:
{context}

## Pregunta:
{query}

## Respuesta:"""

        # 4. Llamar al LLM
        response = llm.invoke(prompt)
        console.print()
        console.print(Markdown(response.content))
        console.print()


if __name__ == "__main__":
    main()

from qdrant_client.models import Filter, FieldCondition, MatchValue, Range

# Buscar solo en ficheros .yaml
results = client.query_points(
    collection_name="ops-docs",
    query=query_embedding,
    query_filter=Filter(
        must=[
            FieldCondition(key="extension", match=MatchValue(value=".yaml"))
        ]
    ),
    limit=5,
)

# Buscar en un fichero especifico
results = client.query_points(
    collection_name="ops-docs",
    query=query_embedding,
    query_filter=Filter(
        must=[
            FieldCondition(key="source", match=MatchValue(value="rules/REQUEST-942-APPLICATION-ATTACK-SQLI.conf"))
        ]
    ),
    limit=10,
)

# Combinar filtros: extension .conf Y chunk_index < 5 (primeras secciones)
results = client.query_points(
    collection_name="ops-docs",
    query=query_embedding,
    query_filter=Filter(
        must=[
            FieldCondition(key="extension", match=MatchValue(value=".conf")),
            FieldCondition(key="chunk_index", range=Range(lt=5)),
        ]
    ),
    limit=5,
)

#!/usr/bin/env python3
"""migrate_chroma_to_qdrant.py - Migrar vectores de ChromaDB a Qdrant"""

import uuid

import chromadb
from qdrant_client import QdrantClient
from qdrant_client.models import Distance, PointStruct, VectorParams
from rich.console import Console
from rich.progress import track

console = Console()

# Configuracion
CHROMA_HOST = "localhost"
CHROMA_PORT = 8001
CHROMA_COLLECTION = "ops-docs"

QDRANT_HOST = "localhost"
QDRANT_PORT = 6333
QDRANT_COLLECTION = "ops-docs"


def main():
    console.print("\n[bold blue]Migracion ChromaDB -> Qdrant[/]\n")

    # Conectar a ChromaDB
    chroma = chromadb.HttpClient(host=CHROMA_HOST, port=CHROMA_PORT)
    collection = chroma.get_collection(CHROMA_COLLECTION)

    # Obtener todos los datos
    data = collection.get(include=["documents", "embeddings", "metadatas"])
    total = len(data["ids"])
    console.print(f"[green]ChromaDB:[/] {total} vectores en '{CHROMA_COLLECTION}'")

    if total == 0:
        console.print("[red]Coleccion vacia.[/]")
        return

    # Detectar dimension
    vector_size = len(data["embeddings"][0])
    console.print(f"[green]Dimension:[/] {vector_size}")

    # Conectar a Qdrant
    qdrant = QdrantClient(host=QDRANT_HOST, port=QDRANT_PORT)

    # Crear coleccion en Qdrant
    qdrant.recreate_collection(
        collection_name=QDRANT_COLLECTION,
        vectors_config=VectorParams(size=vector_size, distance=Distance.COSINE),
    )

    # Migrar en batches
    batch_size = 100
    points = []

    for i in range(total):
        payload = {"text": data["documents"][i]}
        if data["metadatas"][i]:
            payload.update(data["metadatas"][i])

        points.append(
            PointStruct(
                id=str(uuid.uuid4()),
                vector=data["embeddings"][i],
                payload=payload,
            )
        )

    for i in track(range(0, len(points), batch_size), description="Migrando..."):
        batch = points[i : i + batch_size]
        qdrant.upsert(collection_name=QDRANT_COLLECTION, points=batch)

    console.print(f"\n[bold green]Migracion completada: {total} vectores[/]")
    console.print(f"[dim]ChromaDB ({CHROMA_HOST}:{CHROMA_PORT}) -> Qdrant ({QDRANT_HOST}:{QDRANT_PORT})[/]\n")


if __name__ == "__main__":
    main()

# Ejecutar con ambos servicios corriendo
docker compose up -d  # ChromaDB + Qdrant
python migrate_chroma_to_qdrant.py

# Metricas raw
curl http://localhost:6333/metrics

# Dashboard web (incluido)
open http://localhost:6333/dashboard

scrape_configs:
  - job_name: 'qdrant'
    static_configs:
      - targets: ['qdrant:6333']
    metrics_path: '/metrics'

services:
  qdrant:
    image: qdrant/qdrant:latest
    container_name: qdrant
    ports:
      - "6333:6333"
      - "6334:6334"
    volumes:
      - qdrant_data:/qdrant/storage
    environment:
      - QDRANT__SERVICE__GRPC_PORT=6334
      - QDRANT__LOG_LEVEL=INFO
      # Opcional: API key para produccion
      # - QDRANT__SERVICE__API_KEY=your-secret-key
    restart: unless-stopped
    healthcheck:
      test: ["CMD", "curl", "-f", "http://localhost:6333/healthz"]
      interval: 30s
      timeout: 5s
      retries: 3
    deploy:
      resources:
        limits:
          memory: 2G
          cpus: "2"

volumes:
  qdrant_data:
    driver: local

# En docker-compose, descomentar:
QDRANT__SERVICE__API_KEY=your-secret-key

# En tu .env:
QDRANT_API_KEY=your-secret-key

# En el cliente Python:
client = QdrantClient(
    host="localhost",
    port=6333,
    api_key="your-secret-key",
)