[Bug]: Postgres queries stuck in status:active freezing our apps that use Lucia

[MikeSpock]

Package

​@lucia-auth/adapter-postgresql

Describe the bug

We have a worrying issue on production.

We are using Lucia with a Postgres.js adapter. Queries from pg_stat_activity start to hang on status "active". They can be in that status for hours, and don't go away until the server is restarted / apps redeployed. The queries are the simplest they can be, SELECT by an indexed session id column [session validation]). They pile up quickly, and the app's pool thread limit is quickly reached (it is 10 by default), freezing the app.

• DB: AWS RDS Postgres
• Apps: SvelteKit apps
• Postgres adapter: PostgresJS

The issue:

• pg_stat_activity shows that queries pile up in status: active
• often with the same query_id
• the very same SELECT is executed under 10ms on the server (it is not a long-running query: "fetching: 7 ms")

Here is an example of stuck queries from the pg_stat_activity. As you may see from this example, the query_id is the same, and on is added 9 minutes after the first query.
I usually see queries duplicating like that in the pg_stat_activity query.

[
  {
    "datid": 16461,
    "datname": "auth-db",
    "pid": 21110,
    "leader_pid": null,
    "usesysid": 16460,
    "usename": "auth-db-user",
    "application_name": "postgres.js",
    "client_addr": "x.x.x.132",
    "client_hostname": null,
    "client_port": 37434,
    "backend_start": "2024-10-16 17:39:13.649736 +00:00",
    "xact_start": "2024-10-16 18:03:22.498475 +00:00",
    "query_start": "2024-10-16 18:03:22.498475 +00:00",
    "state_change": "2024-10-16 18:03:22.498478 +00:00",
    "wait_event_type": "Client",
    "wait_event": "ClientRead",
    "state": "active",
    "backend_xid": null,
    "backend_xmin": null,
    "query_id": -6194834080809489076,
    "query": "SELECT * FROM \"user_session\" WHERE id = $1",
    "backend_type": "client backend"
  },
  {
    "datid": 16461,
    "datname": "auth-db",
    "pid": 20583,
    "leader_pid": null,
    "usesysid": 16460,
    "usename": "auth-db-user",
    "application_name": "postgres.js",
    "client_addr": "x.x.x.132",
    "client_hostname": null,
    "client_port": 46586,
    "backend_start": "2024-10-16 17:30:48.670474 +00:00",
    "xact_start": "2024-10-16 18:03:28.508803 +00:00",
    "query_start": "2024-10-16 18:03:28.508803 +00:00",
    "state_change": "2024-10-16 18:03:28.508805 +00:00",
    "wait_event_type": "Client",
    "wait_event": "ClientRead",
    "state": "active",
    "backend_xid": null,
    "backend_xmin": null,
    "query_id": -6194834080809489076,
    "query": "SELECT * FROM \"user_session\" WHERE id = $1",
    "backend_type": "client backend"
  }
]

[pilcrowonpaper]

Are you sure this is a Lucia issue? You may need to close your database connections if you're using a lambda etc
https://github.com/porsager/postgres?tab=readme-ov-file#connection-timeout

[MikeSpock]

I'm not 100% sure, but we are not using the DB for any other stuff, just Lucia, which means that the only queries that the DB receives are Lucia queries.
Also, we are running the apps in ECS containers, not Lambda.
So I thought if there is any special connection handling necessary for not having stuck queries in the DB, it probably ​should be implemented in @lucia-auth/adapter-postgresql.

[pilcrowonpaper]

Are you by chance passing an undefined session ID? It's possible that Postgres is waiting for the value of $1

[pilcrowonpaper]

Can you try querying the database manually and see if you can recreate the issue? I kinda doubt this is an issue with Lucia and adapters aren't expected to manage connections