[Question]: Passing node_dict to the retriever of the RecursiveRetriever building block reference is time-consuming

[hingkan]

Question Validation

• I have searched both the documentation and discord for an answer.

Question

When using the RecursiveRetriever building block reference retrieval, the node_dict passed can be very time-consuming when the index is large
vector_retriever_chunk = index.as_retriever(**vir_args) all_nodes = index.docstore.docs.values() all_nodes_dict = {n.node_id: n for n in all_nodes} retriever_chunk = RecursiveRetriever( "vector", retriever_dict={"vector": vector_retriever_chunk}, node_dict=all_nodes_dict, verbose=True, )
Among them, index is loaded from the local index using load_index_from_storage(storage_context)

[hingkan]

`from cachetools import LRUCache
import threading
import os
from pathlib import Path
import math

from llama_index.core.storage import StorageContext

json_cache = LRUCache(maxsize=20)
cache_lock = threading.Lock()

def fast_json_load(file_path):
"""Use the fast JSON loading function of orjson"""
import orjson
with open(file_path, 'rb') as f:
return orjson.loads(f.read())

def fast_json_load_with_cache(file_path, coll_id, filename=None):
"""Fast JSON loading with cache"""
if filename:
cache_key = f"{coll_id}::{filename}"
else:
filename = os.path.basename(file_path)
cache_key = f"{coll_id}::{filename}"

with cache_lock:
    if cache_key in json_cache:
        return json_cache[cache_key]

    try:
        data = fast_json_load(file_path)
        json_cache[cache_key] = data
        return data
    except Exception as e:
        raise

def patch_storage_with_fast_cached_loader(storage_context: StorageContext, coll_id):
"""Replace the loading function of the storage context with a cached version"""

def wrap_with_coll_id(filename):
    def wrapper(file_path):
        return fast_json_load_with_cache(file_path, coll_id, filename)

    return wrapper

if hasattr(storage_context.docstore, '_docstore'):
    storage_context.docstore._docstore._load_fn = wrap_with_coll_id("docstore.json")
if hasattr(storage_context.index_store, '_index_store'):
    storage_context.index_store._index_store._load_fn = wrap_with_coll_id("index_store.json")
if hasattr(storage_context.graph_store, '_graph_store'):
    storage_context.graph_store._graph_store._load_fn = wrap_with_coll_id("graph_store.json")
if hasattr(storage_context.vector_store, '_vector_store'):
    storage_context.vector_store._vector_store._load_fn = wrap_with_coll_id("image__vector_store.json")

load index

knowledge_id = '_ITwLLSJj'
persist_dir = r'/home/aj/index/_ITwLLSJj'
store = MilvusVectorStore(dim=1024, _overwrite=False)
storage_context = StorageContext.from_defaults(vector_store=store, persist_dir=persist_dir)
patch_storage_with_fast_cached_loader(storage_context, coll_id=knowledge_id)
index = storage_context.load_index_from_storage(store_nodes_overwrite=True)`
The above is part of the code where I want to replace StorageContext for loading the index file. I want to load the index from disk to memory. But it didn't work.