RE: LWLock SerializableFinishedList

Hi

 

I’m having trouble with PostgreSQL 16.8 on Windows where for maybe days it all works perfectly until the number of active connections start increasing, until over say 10 minutes all 97 connections are active but seemingly waiting on LWLock on SerializableFinishedList. They will remain in this locked state for some arbitrary period, up to 1 hour 40 minutes, after which the connections will all clear apparently simultaneously and the application continues.

 

The connections are opened, a few command executed and then closed and returned to the connection pool.

 

The application runs on the same server as the postgresql service.

 

Changing statement_timeout to 1 minute doesn’t seem to cancel these “active” connections.

 

Most of the application runs in READ COMMITTED isolation level, but the particular stored procedure that seems to cause the issue runs in SERIALIZED. We end up with “active” but hanging connections running this stored procedure or the associated commands generated by npgqql associated with opening, closing and returning connections to the pool.

 

I can’t find any information about SerializeableFinishedList or why all the standard timeout parameters seem to have no effect on the cancellation of these hung connections. The npgsql client configuration talks about clearing idle connections, but these are not idle, they are active but hung.

 

Any thoughts on the matter, or what might cause  a lock on SerializableFinishedList would be much appreciated.

 

Regards,

Alec