old broken database connections aren't removed from the connection pool
Hello,
This problem will probably need more digging but, from a report from Nemo:
- if for some reasons, connections are broken (their case is: postgresql has been restarted), then they will be kept in the connection pooler
- prior to 3.28 this will mean having a the full already open queue of connections will all be broken and never replace and thus the instance is stucked
- I think that ince 3.28 this will mean a small memory leak because the pooler will open new connections but I'm not 100% sure
We'll probably need to also fix that in logilab-database
This doesn't look like an easy road for pysopg2 : https://stackoverflow.com/questions/1281875/making-sure-that-psycopg2-database-connection-alive