Unlike other database adaptors, memcache adaptors can be shared among
transactions. So pool them in order to reduce the number of connections
established to memcached servers. And as the connections handle
thread-safety internally (by inheriting from threading.local), threads
can share the same connection instance.
Also, do not create a new connector each time we get disconnected, just
reconnect.
Also, memcached internally retries connection, so no need to retry it
ourselves.
Also, do not log when failing to update & delete entries in cache
server: this is a racy cache, each read much be checked anyway, so
failed writes are just business as usual.
Also, document a bit more why we catch "any" exception happening during
finish, and specify base exception class following python coding best
practices.
Also, use named constant for loggin levels for remaining log directives.
Also, display traceback when catching a connection issue during
__getitem__, as the original exception gets converted into a KeyError.