One of our customer's deployment is affected by strange problem: "openstack server list" command takes almost 30 seconds to complete. We have collected proper debug logs and even strace, but it is hard to figure out root cause: it looks like problem is caused by sqlalchemy's failure to return connection back to the pool, but defining 
"[api_database]/pool_timeout=10" doesn't change anything.

I am looking for help with this situation: it looks like some advanced debugging/troubleshooting methods are needed to understand problem better from sqlalchemy/nova perspective. Please note that this is high-severity problem for customer.