Duplicate key errors on mem0migrations table

[dcr1lud]

Hi mem0 team,

We are experiencing persistent duplicate key errors on the mem0migrations table when running mem0ai, even with only a single pod or process. The error log is as follows:

mem0.vector_stores.pgvector - INFO - Inserting 1 vectors into collection mem0migrations
mem0.vector_stores.pgvector - ERROR - Error occurred: duplicate key value violates unique constraint "mem0migrations_pkey"

This happens even when all migrations have already been applied and no other processes are writing to the table. It appears that mem0ai’s migration/version logic is not idempotent or is re-attempting inserts unnecessarily.

This issue occurs even in single-instance deployments. It would be helpful to have the migration/versioning logic reviewed to ensure it is fully idempotent and does not cause duplicate key errors.
Alternatively, please consider adding an environment variable (e.g., MEM0_SKIP_MIGRATIONS) to allow users to disable all runtime migration logic in mem0ai.

Thank you

[dcr1lud]

Maybe /mem0/vector_stores/pgvector.py could be changed to use safe INSERT statement with ON CONFLICT clause like this: sql_query = f"INSERT INTO {self.collection_name} (id, vector, payload) VALUES %s ON CONFLICT (id) DO NOTHING", query itself can be different depending on PSYCOPG_VERSION.
This won't fix the underlying bug of inserting into the wrong table, but it should make the application more resilient while a permanent fix is being developed.

By the way, the vector content which is being inserted into the mem0migrations table is just a long sequence of 0.1, 0.1, ... which looks more like a placeholder than a real vector.