[Bug]: Chroma collection is always empty for local persistence

[jma7889]

What happened?

I am using local persistence, files are saved. However, each time I load, it creates new files. The list_collections function always return none.

import chromadb
from chromadb.config import Settings

WORKING_DIR = '../..'
CHROMA_STORAGE_PATH = os.path.join(WORKING_DIR, 'bin', 'chromadb')

full_path = os.path.abspath(CHROMA_STORAGE_PATH)
print(full_path)
full_path = os.path.abspath(CHROMA_STORAGE_PATH)

assert os.path.exists(full_path), f"The path {full_path} does not exist."

chroma_client = chromadb.Client(
    Settings(
        chroma_db_impl="duckdb+parquet",
        persist_directory=CHROMA_STORAGE_PATH))
collections = chroma_client.list_collections()
        
if not collections:
    print('Chroma collection is empty')
for collection in collections:
    print('Retrieved chroma collection with name: %s %s',
                    collection.id, collection.name)

Versions

chromadb 0.3.26

Relevant log output

Chroma collection is empty

[jeffchuber]

@jma7889 are you using many clients? This usually happens when the way atExit handles cleanup... it will auto-persist the first client last which makes it look like it is wiping the db.

Sorry about this! we are considering taking out the magic entirely since it seems to bite a lot of people.

[jma7889]

Update1:
It seems code to get chroma_client can only be called once. Otherwise, it will create a new database. This is confusing. Once I call below code only once, i can see the collection is not empty. chromadb.Client function is not getting a client, it creates a instance of database!

Original:
@jeffchuber I don't intend to use many or more than 1 client. The only place I get client is using code above. I do use it in jupyter notebook, so the code might be called more than once. And each time it seems create a new set of files but list collections always return None.

            chroma_client = chromadb.Client(
                Settings(
                    chroma_db_impl="duckdb+parquet",
                    persist_directory=CHROMA_STORAGE_PATH))

is above code allowed to be called more than once? or every time it creates a client as well as a new database?