Jelte Fennema-Nio

Easiest way to reproduce is to run the `shmem` example (which requires putting `shmem` in `shared_preload_libraries`). Then `kill -11 `.

Backtrace in case its helpful: ``` 2023-07-21 16:40:02.995 CEST [550346] LOG: all server processes terminated; reinitializing 2023-07-21 16:40:03.125 CEST [550346] FATAL: Can't attach, lock is not in an empty state:...

Based on my reading of the code for `PgLwLock` and `OnceCell`, it seems that `PgLwLock` should not be using `OnceCell` because postgres its reinitialization after a crash should reinitialize the...

> Should probably just use one of the other `cell` types or something like `RwLock`, no need to drop to `UnsafeCell`. Maybe I'm missing something/misunderstanding what you mean. But afaict...

If it's a big problem for you, you can install Citus and configure `track_extra_parameters` to include `search_path`. Creating a small Postgres extension that does the same trick as Citus also...

I'd expect such a failure to also show up on the client side, i.e. if the handshake of a client fails then the client would throw an error. I do...

To be clear, it's also possible that the handshake from PgBouncer to the Postgres server failed. If you don't see any ERROR in the logs after this warning, that actually...

Closing this because the issue is stale

Relevant pgsql-hackers thread that would allow updating the pgbouncer.pool_mode parameter for clients after their initial connection. Which could be useful for multilayered PgBouncer setups. https://www.postgresql.org/message-id/flat/CAGECzQTg2hcmb5GaU53uuWcdC7gCNJFLL6mnW0WNhWHgq9UTgw%40mail.gmail.com

@eulerto Do you still want to work on this?