Should I have to call Sequel.connect in puma's on_worker_boot?

[rgalanakis]

As per https://github.com/jeremyevans/sequel/blob/e0c99f619ab7d8baed1d3aba56b0fadf4b43d930/doc/fork_safety.rdoc#puma-, I am disconnecting my database in before_fork (I need to access the database in the puma parent process to load models).

However, accessing a model database still segfaults in the PG gem, as would happen if before_fork was not being used.

I have ensured the model.db.pool.size is zero before making the database call- active connections are not being carried over to the worker.

If I replace the model's database in on_worker_boot (model.db = Sequel.connect(...)), I do not get a segfault.

This is a bit cumbersome because Sequel errors since it cannot replace model.db once it's already set/the models were loaded, but I can work around it.

I'm wondering if I'm doing something wrong, or there is some stale state in Sequel#Database that isn't being reset on disconnect.

I was trying to create a tiny reproduction case, but cannot force a segfault to happen.

Sequel 5.93.0, sequel_pg 1.17.2, pg 1.5.9

[jeremyevans]

Disconnecting in before_fork should work. If you put together a minimal self contained reproducible example, I can take a look.