I wanted to create the issue so we don't forget it.
Basically due to this PR https://github.com/importcjj/mobc/pull/54, if the health check fails it will retry to get a connection.
This is a good behaviour in general, but prisma decided to fail on all broken connections and defer to the client what to do.
This will break this consistent behaviour since sometimes it will retry broken connections (mobc health check) and sometimes it won't (failed perform_io).
A quick fix is to set the max_bad_conn_retries to 0 (currently we use the default of two)
I wanted to create the issue so we don't forget it. Basically due to this PR https://github.com/importcjj/mobc/pull/54, if the health check fails it will retry to get a connection. This is a good behaviour in general, but prisma decided to fail on all broken connections and defer to the client what to do. This will break this consistent behaviour since sometimes it will retry broken connections (mobc health check) and sometimes it won't (failed perform_io). A quick fix is to set the
max_bad_conn_retries
to 0 (currently we use the default of two)