Pg-vector upsert indexing overwrites saved section, vector, documents rows and only maintains a single row

[obonyojimmy]

Pg vector indexing keeps overwriting saved rows for sections , vector tables .

[davidmezzetti]

I'll take a look and report back.

[davidmezzetti]

Sorry for the delay in checking this out. Could you share more on this?

For example, when I run the simple example below, it works as expected:

from txtai import Embeddings

# URL set in code for demo purposes. Use environment variables in production.
url = "postgresql+psycopg2://postgres:pass@localhost/postgres"

# Create embeddings
embeddings = Embeddings(
    content=url,
    backend="pgvector",
    pgvector={
        "url": url
    }
)

embeddings.index(["test"])
embeddings.upsert(["test2"])
embeddings.upsert(["test3"])
print(embeddings.search("test"))

# [
#   {'id': '0', 'text': 'test', 'score': 1.0},
#   {'id': '1', 'text': 'test2', 'score': 0.6994196176528931},
#   {'id': '2', 'text': 'test3', 'score': 0.5753536224365234}
# ]

And in Postgres.

postgres=# select * from sections;
 indexid | id | text  | tags |           entry            
---------+----+-------+------+----------------------------
       0 | 0  | test  |      | 2024-05-31 08:24:52.931027
       1 | 1  | test2 |      | 2024-05-31 08:24:53.09685
       2 | 2  | test3 |      | 2024-05-31 08:24:53.10798

postgres=# select indexid from vectors;
 indexid 
---------
       0
       1
       2

Are you by chance passing the same ids? In that case, it would overwrite the same rows.